No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Explainability and Interpretability in Electric Load Forecasting Using Machine Learning Techniques – A Review
... Although this technique demonstrates efficacy in enhancing NWP accuracy, its reliance primarily on data input and applying an artificial intelligence model running in post-processing presents limitations. This approach does not facilitate a comprehensive understanding of the physical and mathematical dynamics underlying these enhanced outcomes, as the use of conventional ML methods in this context stages the challenge due to their inherent difficulties in interpretability and explainability [11]. ...
... In the context of this study, explainability is the analysis of how the model is trained and what mathematical representations of its internal logic and processes can be extracted from the trained model. Interpretability, on the other hand, involves analyzing how these formulas can be understood, focusing on the intuition behind the results derived from the formulas that help in understanding wind behavior from the GFS spatial point to the LPMA runway point [11,25]. ...
... In table 1, the performance of each model is detailed for the test dataset, which was previously unseen by FFNN and KAN models. For the KAN approach, the results represent the application of equations (11) and (12), therefore denoting an immediate calculation. ...
This study examines the application of machine learning to enhance wind nowcasting by using a Kolmogorov-Arnold Network model to improve predictions from the Global Forecast System at Madeira International Airport, a site affected by complex terrain. The research addresses the limitations of traditional numerical weather prediction models, which often fail to accurately forecast localized wind patterns. Using the Kolmogorov-Arnold Network model led to a substantial reduction in wind speed and direction forecast errors, with a performance that reached a 48.5% improvement to the Global Forecast System 3 h nowcast, considering the mean squared error. A key outcome of this study comes from the model’s ability to generate mathematical formulas that provide insights into the physical and mathematical dynamics influencing local wind patterns and improve the transparency, explainability, and interpretability of the employed machine learning models for atmosphere modeling.
... The studies surveyed largely relied on traditional machine learning techniques as has also been noted in another recent survey [9]. The main advantages of such methods are that they are inherently interpretable [10] and tend to have lower computation complexity compared to their deep learning counterparts. ...
... domain specific features) as well as contextual and behavioural. Together with selected interpretable machine learning models, we study the contribution and importance of each feature category on the final performance of the models for household electricity forecasting while also reflecting on individual feature influence using a feature importance based explainability technique [10]. ...
The transition from traditional power grids to smart grids, significant increase in the use of renewable energy sources, and soaring electricity prices has triggered a digital transformation of the energy infrastructure that enables new, data driven, applications often supported by machine learning models. However, the majority of the developed machine learning models rely on univariate data. To date, a structured study considering the role meta-data and additional measurements resulting in multivariate data is missing. In this paper we propose a taxonomy that identifies and structures various types of data related to energy applications. The taxonomy can be used to guide application specific data model development for training machine learning models. Focusing on a household electricity forecasting application, we validate the effectiveness of the proposed taxonomy in guiding the selection of the features for various types of models. As such, we study of the effect of domain, contextual and behavioral features on the forecasting accuracy of four interpretable machine learning techniques and three openly available datasets. Finally, using a feature importance techniques, we explain individual feature contributions to the forecasting accuracy.
... The increasing availability of big data in buildings presents an excellent opportunity to apply machine learning models to analyze and model electricity consumption in buildings [24]. Numerous studies have reviewed these applications, focusing on aspects such as the algorithms used, input variables, building types, data resolution, forecast horizons, etc. [24][25][26]. Given the extensive literature in this field, we focused on the most relevant research to our work in this area. ...
... Over time, it has been observed that the complexity of machine learning and neural network architectures increases, making the meaningfulness and transparency of forecasts more critical. There is a growing interest in interpretable and explainable load forecasting methods, leading to increased use of probabilistic models, time series methods, and classically interpretable models [26]. Therefore, in the following sections, regression-based approaches are presented, which remain among the most popular methods for short-term and long-term electricity forecasting [34][35][36][37]. ...
Wood stoves are commonly used for the space heating of residential buildings, especially in Norway. While their influence on building energy consumption has been extensively studied, the contribution of wood stoves to building power consumption has barely been addressed in the scientific literature. Wood stoves typically have a relatively high nominal power (6–8 kW) and are common in most Norwegian homes. Thus, the effect of wood stoves on the heating power of households is expected to be large. However, this effect has never been demonstrated using measurement data. For this purpose, the paper compares the aggregated hourly electricity use of detached residential houses with and without wood stoves. The baseline space-heating system is electric panels, but air-to-air heat pumps are also considered. The results confirm that the contribution of wood stoves to the reduction of electric power is large (that is, up to
10 W/m² at 10°C), especially during peak hours when the occupants are present and active. However, wood stoves also decrease electrical power in the middle of the night, when occupants are not expected to operate the wood stove. This suggests that the ownership of a wood stove could also influence user behavior, such as the desired indoor temperature. These findings highlight the critical role of wood stoves in alleviating the stress of power demand on the electricity grid by replacing electricity with biomass heating. In conclusion, wood stoves play an important role in household energy use and have broader implications for power grid management and peak load reduction.
... In this context, two key concepts are interpretability and explainability. Interpretability refers to how easily humans can understand a model's predictions based solely on its structure and design, while explainability focuses on how well humans can reason about the factors influencing the model's predictions [3]. ...
... (5) Note that equation (5) is similar to the one defined for the linear model, i.e., (3). Essentially, the evaluation of a feature value occurs by averaging the differences in predictions made when fixing the values of the coalition features and randomly sorting through the values of features that are not included in the coalition. ...
This paper explores the application of Explainable AI (XAI) techniques to improve the transparency and understanding of predictive models in control of automated supply air temperature (ASAT) of Air Handling Unit (AHU). The study focuses on forecasting of ASAT using a linear regression with Huber loss. However, having only a control curve without semantic and/or physical explanation is often not enough. The present study employs one of the XAI methods: Shapley values, which allows to reveal the reasoning and highlight the contribution of each feature to the final ASAT forecast. In comparison to other XAI methods, Shapley values have solid mathematical background, resulting in interpretation transparency. The study demonstrates the contrastive explanations--slices, for each control value of ASAT, which makes it possible to give the client objective justifications for curve changes.
... In order to predict power demand more accurately, researchers have proposed a variety of load forecasting methods [1]. Usually, short-term load forecasting models can be divided into three categories: statistical models, deep learning models and hybrid models [2], [3]. ...
... With increasing the number of n and df , the TCN has the ability to expand the receptive field, which makes the output of top layer to learn from much effective input information. When n = 2 and df = [1,2,4], the structure of the dilated causal convolution stack is shown in Figure 2. With adding the dilated convolution stack, the output y t can get and learn from the information of inputs x t−7 , x t−6 , x t−5 , ..., x t . ...
Accurate short-term load forecasting (LF) under extreme weather is vital for the sustainable development of energy systems. This paper proposes a basic framework for future load forecasting researches of sustainable energy systems under extreme weather events and provides new direction for membrane computing model in terms of load forecasting. Inspired by nonlinear spiking mechanisms in nonlinear spiking neural P systems, the gated spiking neural P (GSNP) model is a new recurrent-like network. In this study, we develop an innovative membrane computing model, termed FATCN-LF-FAGSNP model. Frequency enhanced channel attention mechanism (FECAM) is utilized to enhance the features extraction ability of temporal convolutional network (TCN) and improve prediction ability of GSNP systems. Frequency Attention-TCN (FATCN) fully extracts the temporal relationship of features, the features of each channel interact with each frequency component to learn more temporal information effectively and comprehensively in frequency domain. Moreover, adding FECAM to extract features from the data fully reveals the relationship between influencing factors and the load series, which improves the quality of data features and the forecasting accuracy of the FAGSNP model. Then inspired by the interaction mechanism of impulses between biological neuronal cells, FAGSNP is able to consider the load variability and effectively predict load trends. In addition, to address load prediction challenges posed by extreme weather and promote the sustainable development of power systems, the proposed model integrated many models to solve this problem. First, optimized variational mode decomposition (VMD) is used to decompose the load series and the sub-sequences are combined with relevant features, to form the different input sequences of the prediction model. Then, FATCN-LF-FAGSNP model is developed to accurately forecast each high frequency components. Subsequently inverted Transformer model and Informer model are utilized to predict low frequency components and residual component, respectively. Finally all predicted components are reconstructed to get the final predicted results.We conducted extensive comparative experiments with ten baseline models on three real-world datasets, compared with GSNP model and TCN-GSNP model, the R
2
of the FATCN-LF-FAGSNP model increased and MAPE, MAE and RAE reduced, the LF accuracy (measured by R
2
) of the proposed hybrid model gets 0.997 in seasonal LF task. In addition, the proposed hybrid model gets the best in MAPE, MAE, R
2
and RAE metrics in all cases, which demonstrated the effectiveness of the proposed model in LF tasks under both extreme weather scenarios and seasonal prediction scenarios.
... As a single generic model cannot effectively address all issues, the forecasting problem is categorized into short-, medium-, and long-term load forecasting [2]. Machine Learning (ML) and Deep Learning (DL) based load predictions have experienced explosive growth in recent years due to their ability to handle nonlinearity, large data, feature extraction automation, and good performance [3]. ...
... It is nonsymmetric and does not penalize large errors. All errors are equally weighted, as shown in (3). Figure 3 illustrates the workflow, which consists of three sections, the data collection and processing phase, various missing data imputation processes, and the implementation of forecasting models on the datasets. ...
Electricity load forecasting is an important aspect of power system management. Improving forecasting accuracy ensures reliable electricity supply, grid operations, and cost savings. Often, collected data consist of Missing Values (MVs), anomalies, outliers, or other inconsistencies caused by power failures, metering errors, data collection errors, hardware failures, network failures, or other unexpected events. This study uses real-world data to investigate the possibility of using synthetically generated data as an alternative to filling in MVs. Three datasets were created from an original one based on different imputation methods. The imputation methods employed were linear interpolation, imputation using synthetic data, and a proposed hybrid method based on linear interpolation and synthetic data. The performance of the three datasets was compared using deep learning, machine learning, and statistical models and verified based on forecasting accuracy improvements. The findings demonstrate that the hybrid dataset outperformed the other interpolation methods based on the forecasting accuracy of the models.
... Given that batteries are among the most complicated components that power EV engines, having viable battery management systems is essential to mitigate performance and battery degradation problems, thus ensuring battery performance within electric vehicle fleets. Many other problems, such as limited vehicle efficiency and aspects of operation, can be also addressed with viable battery management systems [1], [7]. ...
... It is critical for some machine learning algorithms, such as neural networks, because without this step, it may take a long time for the algorithm to converge . At the same, normalization allows for scaling the feature values to a specific range, for instance, from 0 to 1 , to ensure better convergence and prevent features with a higher scale from dominating over others [4], [7]. ...
The objective of this research is to apply machine learning techniques to optimize electric vehicle battery management and balance to attain maximum battery performance. Here, we will assess and compare the efficiency and accuracy of decision tree classifier, ANN, and Naive Bayes classifiers in: predicting two optimal charging and discharging battery management strategies, prediction of the battery's performance and detecting any abnormalities in it.. In this context, one machine learning model will be proposed to have the most advantageous performance. The experimental results details that such findings are attained, and the precision rates, recall rates, and F1-scores attained by net models exceed 98%. Additionally, the decision tree, as well as the Naive Bayes classifier, have impressive performance, and their accuracy rates exceed 90%. Decision trees along with Naive Bayes classifiers have also important impacts on the identification of classification of battery's responses through probabilistic classification in the former. Our findings have important implications for ensuring electric vehicle batteries are managed more properly for optimal performance. Additionally, based on the results obtained, they can be utilized to help relevant stakeholders in the EV industry and other industries with battery usage implement practical measures that optimize battery performance, increase battery life, and reduce the incurred operational expenses of electric vehicle fleets.
... One of the major issues is the lack of interpretability in most deep learning models, which make failure predictions without explaining which sensor features contribute to the decision. This lack of transparency limits their adoption in safety-critical applications [11]. Additionally, traditional machine learning models require large amounts of labeled failure data [12], which are often unavailable or imbalanced, as turbine failures are relatively rare. ...
Predictive maintenance is essential for ensuring the reliability and efficiency of wind energy systems. Traditional deep learning models for sensor fault detection rely solely on data-driven patterns, often lacking interpretability and robustness. This paper proposes a Physics-Guided Bayesian Neural Network (PINN-BNN) model that integrates physics-informed learning with Bayesian inference to improve fault detection in wind turbines. The proposed approach enforces domain-specific constraints to ensure physically consistent predictions while quantifying uncertainty for risk-aware decision-making. The model is evaluated using a real-world wind turbine sensor dataset, achieving an accuracy of 97.6%, a recall of 91.8%, and an AUC-ROC of 0.987. The SHapley Additive exPlanations (SHAP) analysis reveals that gearbox temperature, blade vibration, and generator torque are the most critical features influencing failure predictions. Bayesian uncertainty estimation further improves interpretability by assigning confidence levels to each prediction. A comparative study with ten baseline models, including Long Short-Term Memory (LSTM), Transformer-based models, and traditional machine learning classifiers, demonstrates that the PINN-BNN model outperforms existing approaches while maintaining computational efficiency with a training time of 39.8 minutes and an inference time of 1.7 ms per sample. The integration of physics-informed learning ensures that the model generalizes well to varying environmental conditions, reducing false negatives and minimizing unexpected system failures. The proposed methodology presents a step toward interpretable and reliable predictive maintenance in wind energy systems.
... Ensemble learning techniques combine multiple models to improve predictive accuracy. Breiman (1996) introduced bagging as one of the earliest ensemble techniques, which improves model performance by reducing variance and bias [63]. Ensemble methods are highly effective in improving generalization and robustness, making them suitable for complex forecasting tasks like load demand prediction [64]. ...
Enhancement prediction of load demand is crucial for effective energy management and resource allocation in modern power systems and especially in medical segment. Proposed method leverages strengths of ANFIS in learning complex nonlinear relationships inherent in load demand data. To evaluate the effectiveness of the proposed approach, researchers conducted hybrid methodology combine LHS with ANFIS, using actual load demand readings. Comparative analysis investigates performing various machine learning models, including Adaptive Neuro-Fuzzy Inference Systems (ANFIS) alone, and ANFIS combined with Latin Hypercube sampling (LHS), in predicting electrical load demand. The paper explores enhancing ANFIS through LHS compared with Monte Carlo (MC) method to improve predictive accuracy. It involves simulating energy demand patterns over 1000 iterations, using performance metrics through Mean Squared Error (MSE). The study shows superior predictive performance of ANFIS-LHS model, achieving higher accuracy and robustness in load demand prediction across different time horizons and scenarios. Thus, findings of this research contribute to advanced developments rather than previous research by introducing a combined predictive methodology that leverages LHS to ensure solving limitations of previous methods like structured, stratified sampling of input variables, reducing overfitting and enhancing adaptability to varying data sizes. Additionally, it incorporates sensitivity analysis and risk assessment, significantly improving predictive accuracy. Using Python and Simulink Matlab, Combined LHS with ANFIS showing accuracy of 96.42% improvement over the ANFIS model alone.
... Inexperienced model interpretability and validation: Many electricity demand forecasting model especially ML/AI and hybrid models are considered "black boxes", because they provide predictions without offering a clear explanation of the underlying decision-making processes (Md et al., 2024). This lack of transparency makes it challenging for regulators, operators, and stakeholders to trust the model's outputs, or understand why a particular forecast was carried out (Lukas et al., 2024). For utilities and electricity providers, there are often regulatory requirements to justify the predictions used for pricing, grid planning and energy policy. ...
Electricity demand forecasting has emerged as a critical area of research in recent times, driven by the necessity for accurate predictions of future load requirements. Such predictions are essential for effectively operating and planning electric power systems. Various forecasting methodologies and approaches have been employed to estimate electricity demand, emphasizing the need for precision and informed analysis in electricity management. Accordingly, diverse approaches have been utilized within the research community to provide optimal estimates for future electricity demand. This study evaluates the global trends and advancements in electricity demand forecasting methodologies through a comprehensive review and analysis of existing literature relating to electricity demand management, electricity forecasting methodologies and applications. The forecasting methodologies are categorized into statistical, Machine Learning/Artificial Intelligence (ML/AI), and hybrid models. The findings indicate that while ML/AI-based models are applied more in electricity demand forecasting as compared with statistical models, hybrid models are preferred for their sustained accuracy, enhanced abilities in flexibility, productivity, talent pool, cost saving, precision, and reduced volatility. This emerging reliance on hybrid models is attributed to the integration of the forecasting capabilities of different models. The review finally recapped the challenges and opportunities for future research in electricity demand forecasting in Nigeria and globally.
... Explainability pertains to a systems capacity to consider its outputs in a manner that is clear, comprehensible, and trustworthy for human users [12]. Interpretability refers to the extent to which a human can understand the process of decision making and prediction of models based solely on its design, without requiring additional explanations or external information [13]. During the designing and developmental phase of DNN models, joint design and reviews among development teams are more effective and goal-driven when models are inherently explainable. ...
... An alternative approach to enhance time series forecasting models and gain insight into their internal mechanisms is the attention mechanism. To visualise attention, attention maps are an effective tool [6]. As the explainability of the models is enhanced, their trustworthiness is increased, thereby facilitating the decision-making process for relevant stakeholders, such as building managers. ...
The role of heating load forecasts in the energy transition is significant, given the considerable increase in the number of heat pumps and the growing prevalence of fluctuating electricity generation. While machine learning methods offer promising forecasting capabilities, their black-box nature makes them difficult to interpret and explain. The deployment of explainable artificial intelligence methodologies enables the actions of these machine learning models to be made transparent.
In this study, a multi-step forecast was employed using an encoder–decoder model to forecast the hourly heating load for an multifamily residential building and a district heating system over a forecast horizon of 24-hour. By using 24 instead of 48 lagged hours, the simulation time was reduced from 92.75 s to 45.80 s and the forecast accuracy was increased. The feature selection was conducted for four distinct methods. The Tree and Deep SHAP method yielded superior results in feature selection. The application of feature selection according to the Deep SHAP values resulted in a reduction of 3.98 % in the training time and a 8.11 % reduction in the NRMSE. The utilisation of local Deep SHAP values enables the visualisation of the influence of past input hours and individual features. By mapping temporal attention, it was possible to demonstrate the importance of the most recent time steps in a intrinsic way.
The combination of explainable methods enables plant operators to gain further insights and trustworthiness from the purely data-driven forecast model, and to identify the importance of individual features and time steps.
... Over the years, notable data-driven techniques such ARMA models [8], support vector regression (SVR) [9], graph convolutional networks (GCNs) [10], fuzzy models [11,12], neurofuzzy models [13,14], and graph-based artifcial neural networks (graph-ANN) [15] have been utilized to capture spatial and temporal dependencies within environmental data and energy networks obtaining remarkable results in short-and long-term prediction [16][17][18][19]. ...
Renewable energy forecasting is crucial for pollution prevention, management, and long-term sustainability. In response to the challenges associated with energy forecasting, the simultaneous deployment of several data-processing approaches has been used in a variety of studies in order to improve the energy–time-series analysis, finding that, when combined with the wavelet analysis, deep learning techniques can achieve high accuracy in energy forecasting applications. Consequently, we investigate the implementation of various wavelets within the structure of a long short-term memory neural network (LSTM), resulting in the new LSTM wavelet (LSTMW) neural network. In addition, and as an improvement phase, we modeled the uncertainty and incorporated it into the forecast so that systemic biases and deviations could be accounted for (LSTMW with luster: LSTMWL). The models were evaluated using data from six renewable power generation plants in Chile. When compared to other approaches, experimental results show that our method provides a prediction error within an acceptable range, achieving a coefficient of determination (R2) between 0.73 and 0.98 across different test scenarios, and a consistent alignment between forecasted and observed values, particularly during the first 3 prediction steps.
... As several authors have observed, despite their utility, XAI methods have notable limitations. For instance, many methods struggle to provide meaningful explanations for correlated features [47,48]. Additionally, these methods can be computationally expensive, and the approximating white-box models they rely on may sometimes produce misleading results [49]. ...
Accurate weather prediction and electrical load modeling are critical for optimizing energy systems and mitigating environmental impacts. This study explores the integration of the novel Mean Background Method and Background Estimation Method with Explainable Artificial Intelligence (XAI) with the aim to enhance the evaluation and understanding of time-series models in these domains. The electrical load or temperature predictions are regression-based problems. Some XAI methods, such as SHAP, require using the base value of the model as the background to provide an explanation. However, in contextualized situations, the default base value is not always the best choice. The selection of the background can significantly affect the corresponding Shapley values. This paper presents two innovative XAI methods designed to provide robust context-aware explanations for regression and time-series problems, addressing critical gaps in model interpretability. They can be used to improve background selection to make more conscious decisions and improve the understanding of predictions made by models that use time-series data.
... Each technique uses historical data to generate forecasts; yet, they all employ overly simplistic frameworks, fail to explore data connections, and consider past and future data merely as mathematical equations, resulting in inaccurate predictions [16]. The rapid advancement of machine learning has led to several accomplishments in the domain of health [17], automatic detection [18], security [19], and energy load forecasting [20,21]. Typical machine learning techniques include decision trees [22], image classification, and support vector machines [23], highlighting the capabilities of ML algorithms in proposing an appropriate solution for different problems. ...
The energy sector plays a vital role in driving environmental and social advancements. Accurately predicting energy demand across various time frames offers numerous benefits, such as facilitating a sustainable transition and planning of energy resources. This research focuses on predicting energy consumption using three individual models: Prophet, eXtreme Gradient Boosting (XGBoost), and long short-term memory (LSTM). Additionally, it proposes an ensemble model that combines the predictions from all three to enhance overall accuracy. This approach aims to leverage the strengths of each model for better prediction performance. We examine the accuracy of an ensemble model using Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) through means of resource allocation. The research investigates the use of real data from smart meters gathered from 5567 London residences as part of the UK Power Networks-led Low Carbon London project from the London Datastore. The performance of each individual model was recorded as follows: 62.96% for the Prophet model, 70.37% for LSTM, and 66.66% for XGBoost. In contrast, the proposed ensemble model, which combines LSTM, Prophet, and XGBoost, achieved an impressive accuracy of 81.48%, surpassing the individual models. The findings of this study indicate that the proposed model enhances energy efficiency and supports the transition towards a sustainable energy future. Consequently, it can accurately forecast the maximum loads of distribution networks for London households. In addition, this work contributes to the improvement of load forecasting for distribution networks, which can guide higher authorities in developing sustainable energy consumption plans.
... Baur et al. [4] reviews literature on explainable and interpretable machine learning methods for electric load forecasting, identifying trends and techniques to improve forecast transparency and interpretability. An example of an MTLF model with interpretability features is presented in [16], where the transformer model is capable of explaining the contribution of each input feature to the predicted load at different times of the day. ...
This paper presents an enhanced N-BEATS model, N-BEATS*, for improved mid-term electricity load forecasting (MTLF). Building on the strengths of the original N-BEATS architecture, which excels in handling complex time series data without requiring preprocessing or domain-specific knowledge, N-BEATS* introduces two key modifications. (1) A novel loss function -- combining pinball loss based on MAPE with normalized MSE, the new loss function allows for a more balanced approach by capturing both L1 and L2 loss terms. (2) A modified block architecture -- the internal structure of the N-BEATS blocks is adjusted by introducing a destandardization component to harmonize the processing of different time series, leading to more efficient and less complex forecasting tasks. Evaluated on real-world monthly electricity consumption data from 35 European countries, N-BEATS* demonstrates superior performance compared to its predecessor and other established forecasting methods, including statistical, machine learning, and hybrid models. N-BEATS* achieves the lowest MAPE and RMSE, while also exhibiting the lowest dispersion in forecast errors.
... Probabilistic modeling enhances the interpretability of short-term load forecasting by quantifying uncertainties and incorporating them into predictions. This approach provides confidence intervals and quantiles, offering a clearer understanding of prediction variability, but can introduce complexity depending on the underlying model design [14]. ...
Effective short-term load forecasting (STLF) is essential for optimizing electricity grid operations. This study focuses on refining STLF for day-ahead predictions using Bayesian multiple linear regression (BMLR). This study’s originality lies in its innovative use of BMLR combined with data clustering techniques to improve prediction accuracy, a method not previously explored in existing literature. We address the critical issue of input data clustering, highlighting its impact on prediction accuracy. Four clustering methods based on temporality were examined, with clustering by weekday and hour proving most effective for BMLR-based STLF. Predictors included historical load, temperature, season, weekday, and hour, selected using the Akaike information criterion (AIC). Linear regression assumptions were verified, and solutions were proposed for deviations, notably addressing heteroscedasticity. Autocorrelation in residuals was addressed to improve forecasting efficiency. Time-cross validation and performance metrics demonstrated model effectiveness. Second-degree polynomial terms are included for better fitting. Clustering by weekday and hour is optimal for BMLR-based STLF, aiding accurate load forecasts. The main objectives of this research are to determine the optimal clustering method for BMLR in STLF and to provide practical insights into the application of Bayesian techniques in load forecasting. This research significantly contributes to the field of STLF by providing practical insights into data clustering and model refinement, offering valuable perspectives for enhanced energy management and grid stability.
... With the continuous maturation of emerging technologies such as artificial intelligence and machine learning, the intelligent application of the primary processing production process has gradually become a major research focus in the industry [1][2]. Steam, as an essential secondary energy source in the tobacco industry, relies heavily on non-renewable energy sources, including natural gas, coal, and petroleum. ...
... Explainability refers to the ability of a model to provide a rationale for its outputs that can be easily understood and trusted by humans [8]. Interpretability refers to how well a human can comprehend a model's prediction and decision-making solely by model's design, without additional information [9]. At architectural design phase of DNN models, collaborative review and joint design between development teams become more efficient and outcomeoriented when models are more explainable (see Section 4.1). ...
At present Deep Neural Networks (DNN) have a dominant role in the AI-driven Autonomous driving approaches. This paper focuses on the potential safety risks of deploying DNN classifiers in Advanced Driver Assistance System (ADAS) systems. In our experience, many theoretically sound AI-driven solutions tested and deployed in ADAS have shown serious safety flaws in practice. A brief review of practice and theory of automotive safety standards and related body of knowledge is presented. It is followed by a comparative analysis between DNN classifiers and safety standards developed in the automotive industry. The output of the study provides advice and recommendations for filling the current gaps within the complex and interrelated factors pertaining to the safety of Autonomous Road Vehicles (ARV). This study may assist ARV’s safety, system, and technology providers during the design, development, and implementation life cycle. The contribution of this work is to highlight and link the learning rules enforced by risk factors when DNN classifiers are expected to provide a near real-time safer Vehicle Navigation Solution (VNS).
... Particularly with p-type semiconductors, a Schottky barrier can develop at the interface between the anodic metal and the semiconductor membrane in an SJ-SIMFC device. This barrier generates a built-in electric field (BIEF) that directs energy from the metal side toward the semiconductor by generating a depletion layer at the interface [220]. This BIEF lowers electrode polarization loss by preventing electron crossover from the anode to the membrane and promoting the passage of ions (O 2À ) from the cathode to the membrane. ...
Mixed ionic-eelectronic conductors (MIECs) play a crucial role in the landscape of energy
conversion and storage technologies, with a pronounced focus on electrode materials’
application in solid oxide fuel cells (SOFCs) and proton ceramic fuel cells (PCFCs). In parallel,
the emergence of semiconductor ionic materials (SIMs) has introduced a new paradigm in the
field of functional materials, particularly for both electrode and electrolyte in the development
of low-temperature, 300-550 °C, SOFC, and PCFC. This review article critically delves into
the intricate mechanisms underpinning the synergistic relationship between MIECs and SIMs,
with a particular emphasis on elucidating the fundamental working principles of semiconductor
ionic material fuel cells (SIMFCs). By exploring critical facets such as ion-coupled electron
transfer/transport, junction effects, energy band alignments, and theoretical computations, it
casts an illuminating spotlight on the transformative potential of MIECs, also involving triple
charge conducting oxides (TCOs) in the context of SIMs and advanced fuel cells (FCs). The
insights and findings articulated herein contribute substantially to the advancement of SIMs
and SIMFCs by tailoring MIECs (TCOs) as promising avenues in the emergence of highperformance SIMFCs. This scientific quest not only addresses the insistent challenges
surrounding efficient charge transfer, ionic transport and power output but also unlocks the
profound potential for the widespread commercialization of FC technology.
... Linear Autoregressive (LAR) models are the first choice among the data-driven models for time-series forecasting due to their simplicity and interpretability. The linear autoregression predicts the forthcoming value of a target time series as a linear combination of its past values; thus, the weighting coefficients thoroughly explain the model's decisionmaking [16]. Despite the acceptable performance on short-time hydrological forecasting, the augmented randomness and nonlinearity on longer prediction horizons overwhelm the LAR simplicity [17]. ...
Accurate streamflow forecasting is crucial for effectively managing water resources, particularly in countries like Colombia, where hydroelectric power generation significantly contributes to the national energy grid. Although highly interpretable, traditional deterministic, physically-driven models often suffer from complexity and require extensive parameterization. Data-driven models like Linear Autoregressive (LAR) and Long Short-Term Memory (LSTM) networks offer simplicity and performance but cannot quantify uncertainty. This work introduces Sparse Variational Gaussian Processes (SVGPs) for forecasting streamflow contributions. The proposed SVGP model reduces computational complexity compared to traditional Gaussian Processes, making it highly scalable for large datasets. The methodology employs optimal hyperparameters and shared inducing points to capture short-term and long-term relationships among reservoirs. Training, validation, and analysis of the proposed approach consider the streamflow dataset from 23 geographically dispersed reservoirs recorded during twelve years in Colombia. Performance assessment reveals that the proposal outperforms baseline Linear Autoregressive (LAR) and Long Short-Term Memory (LSTM) models in three key aspects: adaptability to changing dynamics, provision of informative confidence intervals through Bayesian inference, and enhanced forecasting accuracy. Therefore, the SVGP-based forecasting methodology offers a scalable and interpretable solution for multi-output streamflow forecasting, thereby contributing to more effective water resource management and hydroelectric planning.
... While load decomposition is not a novel concept [20][21][22][23][24], its application and number of publications has increased in the field of electric load forecasting in recent years. Decomposing the electrical load offers valuable insights and data explanations [25], which can prove beneficial for ML practitioners in STLF. ...
As the demand for electricity, electrification, and renewable energy rises, accurate forecasting and flexible energy management become imperative. Distribution network operators face capacity limits set by regional grids, risking economic penalties if exceeded. This study examined data-driven approaches of load forecasting to address these challenges on a city scale through a use case study of Eskilstuna, Sweden. Multiple Linear Regression was used to model electric load data, identifying key calendar and meteorological variables through a rolling origin validation process, using three years of historical data. Despite its low cost, Multiple Linear Regression outperforms the more expensive non-linear Light Gradient Boosting Machine, and both outperform the “weekly Naïve” benchmark with a relative Root Mean Square Errors of 32–34% and 39–40%, respectively. Best-practice hyperparameter settings were derived, and they emphasize frequent re-training, maximizing the training data size, and setting a lag size larger than or equal to the forecast horizon for improved accuracy. Combining both models into an ensemble could the enhance accuracy. This paper demonstrates that robust load forecasts can be achieved by leveraging domain knowledge and statistical analysis, utilizing readily available machine learning libraries. The methodology for achieving this is presented within the paper. These models have the potential for economic optimization and load-shifting strategies, offering valuable insights into sustainable energy management.
... Particularly with p-type semiconductors, a Schottky barrier can develop at the interface between the anodic metal and the semiconductor membrane in an SJ-SIMFC device. This barrier generates a built-in electric field (BIEF) that directs energy from the metal side toward the semiconductor by generating a depletion layer at the interface[220]. This BIEF lowers electrode polarization loss by preventing electron crossover from the anode to the membrane and promoting the passage of ions (O 2-) from the cathode to the membrane. ...
Highly active and stable electrocatalysts are mandatory for developing high-performance and long-lasting fuel cells. The current study demonstrates a high oxygen reduction reaction (ORR) electrocatalytic activity of a novel spinel-structured LaFe2O4 via a self-doping strategy. The LaFe2O4 demonstrated excellent ORR activity in a protonic ceramic fuel cell (PCFC) at temperature range of 350−500 °C. The high ORR activity of LaFe2O4 is mainly attributed to the facile release of oxide and proton ions, and improved synergistic incorporation abilities associated with interplay of multivalent Fe³⁺/Fe²⁺ and La³⁺ ions. Using LaFe2O4 as cathode over proton conducting BaZr0.4Ce0.4Y0.2O3 (BZCY) electrolyte, the fuel cell has delivered a high-power density of 806 mW/cm² operating at 500 °C. Different spectroscopic and calculations methods such as UV−visible, Raman, X-ray photoelectron spectroscopy and Density Functional Theory (DFT) calculations are performed to screen the potential application of LaFe2O4 as cathode. This study would help in developing functional Cobalt-free ORR electrocatalysts for LT−PCFCs and SOFCs applications.
The precise prediction of power demand is of utmost importance for optimizing power system operations, particularly in the domain of the increasing integration of renewable energy resources. Conventional statistical and machine learning techniques encounter difficulties in capturing complex temporal correlations within load data. The objective of this research is to examine the utilization of Long Short‐Term Memory – Recurrent Neural Networks (LSTM‐RNNs) in load prediction and perform an extensive comparison analysis with the well‐established XG‐Boost and other conventional techniques. The incorporation of demand response and distributed renewable energy sources is of paramount importance in ensuring the stability of smart grids and the accurate assessment of power demand. However, the task of making precise energy forecasts faces various obstacles that stem from climate conditions, societal influences, and seasonal variations. The precision of our LSTM‐RNN model is evaluated using actual demand data obtained from a prominent utility company in Germany. The findings indicate that the LSTM‐RNN model consistently exhibits superior performance compared to standard machine learning techniques and XG‐Boost in both short‐term (1–24 h) and long‐term (yearly) load forecasting. The LSTM‐RNN has a notable level of resilience in generating accurate predictions, particularly when confronted with inadequate or noisy input data. The aforementioned results highlight the potential of LSTM‐RNN in enhancing load forecasting in smart grids, hence enabling the efficient incorporation of demand response mechanisms and renewable energy sources. This study offers valuable insights and presents a comprehensive methodology for improving power demand estimation in contemporary power systems.
Renewable energy sources (RESs) such as solar, wind, and hydroelectric power play an increasingly significant role in the global energy landscape, offering sustainable alternatives to conventional fossil fuels. However, the intermittent and unpredictable nature of renewable energy generation poses challenges for effective integration into the power grid and reliable energy supply. In recent years, machine learning (ML) techniques have emerged as powerful tools for forecasting renewable energy (RE) generation, enabling improved planning, management, and grid integration strategies. This chapter provides a comprehensive survey of ML applications in renewable energy forecasting, covering various techniques, methodologies, and case studies across different RESs. It begins with an overview of traditional forecasting methods and their limitations, followed by an exploration of ML algorithms, including artificial neural networks, support vector machines, decision trees, and ensemble methods, among others, employed for RE prediction tasks. The chapter also discusses data preprocessing techniques, feature selection methods, and evaluation metrics commonly used in ML-based forecasting models. Furthermore, it presents a detailed analysis of case studies and real-world applications in solar, wind, and hydroelectric power forecasting, highlighting the strengths, limitations, and potential areas for future research in this field. Through this survey, researchers, practitioners, and policymakers gain insights into the current state-of-the-art in ML-based RE forecasting, paving the way for enhanced utilization and integration of RESs into the modern power grid.
This study uses the Scopus and Web of Science databases to review quantitative methods to forecast electricity consumption from 2015 to 2024. Using the PRISMA approach, 175 relevant publications were identified from an initial set of 821 documents and subsequently subjected to bibliometric analysis. This analysis examined publication trends, citation metrics, and collaboration patterns across various countries and institutions. Over the period analyzed, the number of articles has steadily increased, with a more rapid rise observed after 2020. Although China dominates this research field, strong bibliographic coupling worldwide indicates significant international collaboration. The study suggests that no single method consistently outperforms others across all contexts and that forecasting methods should be adapted to regional contexts, considering specific economic, social, and environmental factors. Furthermore, we emphasize that review papers should compare methods and results regarding both time horizon and temporal resolution, as these aspects are crucial for the accuracy and applicability of the forecasts.
This study uses the Scopus and Web of Science databases to review the forecast of electricity consumption from 2015 to 2024. Based on the keywords used in the article title, which are associated with electricity consumption forecasts, the study retrieved 821 docu-ments for additional analysis using various techniques. For frequency analysis, we used Mi-crosoft Excel; for data visualisation, we used VOSviewer; and for citation metrics and analysis, we used Harzing's Publish or Perish. Standard bibliometric variables, including publication growth, authorship patterns, collaboration, prolific authors, national contribution, most active institutions, favourite journals, and highly cited articles, are used in this study's reporting of the findings. Our analysis indicates that over ten years, starting in 2015, publications on the forecasting of electricity consumption have continued to increase. China was the country that contributed the most to research on electricity consumption. Most of the research publications on electricity consumption were published in the Journal of Energy.
This research presents a comprehensive case study on medium-term load forecasting (MTLF) in the intricate dynamics of Pakistan’s power sector, Gujranwala Electric Power Company (GEPCO). The time horizon for MTLF ranges from a few weeks to one year and it has applications in energy management and planning. The deep-learning networks (DLNs), proposed in recent years, have a black-box nature, which reduces the interpretability of the results. Therefore, in the proposed study, a mathematical model, Power Market Survey (PMS), is implemented to forecast month-ahead load for GEPCO division dataset. The historical load data, incorporating relevant features such as growth rate, load factor, schedule of load shedding, etc., are fed as input to the proposed model for efficient load forecasting. The PMS is trained with a bottom-up approach and its performance is validated through mean absolute percentage error (MAPE), mean absolute error (MAE), root mean square error (RMSE), mean square error (MSE), and confidence interval with 95% of probability. The results demonstrate that the proposed model performed well while incorporating the effects of planned load shedding as compared to the case in which load-shedding effects are neglected. Moreover, while incorporating the load-shedding effects, the proposed model recorded MAPE and MAE as 11.91% and 0.22, respectively, for the fiscal year 2022–2023. However, when the proposed model is implemented while neglecting the effects of load shedding, the MAPE and MAE of 13.64% and 0.24 are recorded by the PMS technique, correspondingly.
The home energy management (HEM) sector is going through an enormous change that includes important elements like incorporating green power, enhancing efficiency through forecasting and scheduling optimization techniques, employing smart grid infrastructure, and regulating the dynamics of optimal energy trading. As a result, ecosystem players need to clarify their roles, develop effective regulatory structures, and experiment with new business models. Peer-to-Peer (P2P) energy trading seems to be one of the viable options in these conditions, where consumers can sell/buy electricity to/from other users prior to totally depending on the utility. P2P energy trading enables the exchange of energy between consumers and prosumers, thus provide a more robust platform for energy trading. This strategy decentralizes the energy market more than it did previously, opening up new possibilities for improving energy trade between customers and utility. Considering above scenarios, this research provides an extensive insight of P2P energy trading structure, procedure, market design, trading platform, pricing mechanism, P2P approaches, forecasting techniques, scheduling topologies and possible futuristic techniques, while examining their characteristics, pros and cons with the primary goal of determining whichever approach is most appropriate in a given situation for P2P HEMs. Moreover, an optimal and robust P2P HEMs load scheduling framework simulation model is also proposed to analyze the P2P HEMs network critically, thus paving futuristic technical research directions for the scientific researchers. With this cooperation, a new age of technological advancements ushering in a more intelligent, more interconnected, and reactive urban environment are brought to life. In this sense, the path to smart living entails reinventing the urban environment as well as how people interact with and perceive their dwellings in the larger framework of a smart city. Finally, this research work also provides a comprehensive overview of technical challenges in P2P HEMs in terms of load forecasting and scheduling strategies, their possible solutions, and future prospects.
This study investigates the efficacy of Explainable Artificial Intelligence (XAI) methods, specifically Gradient-weighted Class Activation Mapping (Grad-CAM) and Shapley Additive Explanations (SHAP), in the feature selection process for national demand forecasting. Utilising a multi-headed Convolutional Neural Network (CNN), both XAI methods exhibit capabilities in enhancing forecasting accuracy and model efficiency by identifying and eliminating irrelevant features. Comparative analysis revealed Grad-CAM's exceptional computational efficiency in high-dimensional applications and SHAP's superior ability in revealing features that degrade forecast accuracy. However, limitations are found in both methods, with Grad-CAM including features that decrease model stability, and SHAP inaccurately ranking significant features. Future research should focus on refining these XAI methods to overcome these limitations and further probe into other XAI methods' applicability within the time-series forecasting domain. This study underscores the potential of XAI in improving load forecasting, which can contribute significantly to the development of more interpretative, accurate and efficient forecasting models.
Residential load forecasting is of great significance to improve the energy efficiency of smart home services. Deep-learning techniques, i.e., long short-term memory (LSTM) neural networks, can considerably improve the performance of prediction models. However, these black-box networks are generally unexplainable, which creates an obstacle for the customer to deeply understand forecasting results and rapidly respond to uncertain circumstances, as practical engineering requires a high standard of prediction reliability. In this paper, an interpretable deep-learning method is proposed to solve the multi-step residential load forecasting problem which is referred to as explainable artificial intelligence (XAI). An encoder–decoder network architecture based on multi-variable LSTM (MV-LSTM) is developed for the multi-step probabilistic-load forecasting. The mixture attention mechanism is introduced in each prediction time step to better capture the different temporal dynamics of multivariate sequence in an interpretable form. By evaluating the contribution of each variable to the forecast, multi-quantile forecasts at multiple future time steps can be generated. The experiments on the real data set show that the proposed method can achieve good prediction performance while providing valuable explanations for the prediction results. The findings help end users gain insights into the forecasting model, bridging the gap between them and advanced deep-learning techniques.
In recent years, machine learning, especially deep learning, has developed rapidly and has shown remarkable performance in many tasks of the smart grid field. The representation ability of machine learning algorithms is greatly improved, but with the increase of model complexity, the interpretability of machine learning algorithms is worse. The smart grid is a critical infrastructure area, so machine learning models involving it must be interpretable in order to increase user trust and improve system reliability. Unfortunately, the black-box nature of most machine learning models remains unresolved, and many decisions of intelligent systems still lack explanation. In this paper, we elaborate on the definition, motivations, properties, and classification of interpretability. In addition, we review the relevant literature addressing interpretability for smart grid applications. Finally, we discuss the future research directions of interpretable machine learning in the smart grid.
To extract strong correlations between different energy loads and improve the interpretability and accuracy for load forecasting of a regional integrated energy system (RIES), an explainable framework for load forecasting of an RIES is proposed. This includes the load forecasting model of RIES and its interpretation. A coupled feature extracting strategy is adopted to construct coupled features between loads as the input variables of the model. It is designed based on multi-task learning (MTL) with a long short-term memory (LSTM) model as the sharing layer. Based on SHapley Additive exPlanations (SHAP), this explainable framework combines global and local interpretations to improve the interpretability of load forecasting of the RIES. In addition, an input variable selection strategy based on the global SHAP value is proposed to select input feature variables of the model. A case study is given to verify the effectiveness of the proposed model, constructed coupled features, and input variable selection strategy. The results show that the explainable framework intuitively improves the interpretability of the prediction model.
Time series forecasting remains a central challenge problem in almost all scientific disciplines. We introduce a novel load forecasting method in which observed dynamics are modeled as a forced linear system using Dynamic Mode Decomposition (DMD) in time delay coordinates. Central to this approach is the insight that grid load, like many observables on complex real-world systems, has an “almost-periodic” character, i.e., a continuous Fourier spectrum punctuated by dominant peaks, which capture regular (e.g., daily or weekly) recurrences in the dynamics. The forecasting method presented takes advantage of this property by (i) regressing to a deterministic linear model whose eigenspectrum maps onto those peaks, and (ii) simultaneously learning a stochastic Gaussian process regression (GPR) process to actuate this system. Our forecasting algorithm is compared against state-of-the-art forecasting techniques not using additional explanatory variables and is shown to produce superior performance. We further show that this joint approach offers marked improvement over existing methods that use purely DMD or GPR. Moreover, its use of linear intrinsic dynamics offers a number of desirable properties in terms of interpretability and parsimony. Results are presented for a test case using load data from an electrical grid. Load forecasting is an essential challenge in power systems engineering, with major implications for real-time control, pricing, maintenance, and security decisions.
There is a lot of research on the neural models used for short-term load forecasting (STLF), which is crucial for improving the sustainable operation of energy systems with increasing technical, economic, and environmental requirements. Neural networks are computationally powerful; however, the lack of clear, readable and trustworthy justification of STLF obtained using such models is a serious problem that needs to be tackled. The article proposes an approach based on the local interpretable model-agnostic explanations (LIME) method that supports reliable premises justifying and explaining the forecasts. The use of the proposed approach makes it possible to improve the reliability of heuristic and experimental neural modeling processes, the results of which are difficult to interpret. Explaining the forecasting may facilitate the justification of the selection and the improvement of neural models for STLF, while contributing to a better understanding of the obtained results and broadening the knowledge and experience supporting the enhancement of energy systems security based on reliable forecasts and simplifying dispatch decisions.
At present, the prediction accuracy of the power load prediction model is not high and the black box model prediction results are not explanatory, so this paper proposes an improved XGBoost based grid search for power load forecasting model (GS-XGBoost), and uses SHAP for model analysis and interpretation to show the impact of the feature set on the model in a visual way according to the marginal contribution of power load feature samples The impact of the prediction results is shown in a visual way to improve the interpretability of the model, and the key factors affecting the electricity load are identified based on the ranking results of the feature importance to provide decision support for the power system equipment maintenance plan, guide the formulation of strategies related to electricity production, resource procurement and pricing, and avoid market risks. In this paper, the electricity consumption dataset of an enterprise in Jiangsu is used as the research object, and the dataset is pre-processed with unique thermal coding, outlier analysis, null filling, and standardization. Through experimental comparison, the GS-XGBoost load forecasting model has the best prediction accuracy compared to SVR, random forest, decision tree, and XGBoost machine learning models, and then SHAP is applied to interpret the model and improve the model interpretability.
Accurate long-term and midterm electricity load forecasting play an essential role in electric power system planning. Drawing on the seasonal-trend forecasting capacity of Fourier series and LOESS transformation, this paper applies modified Fourier series transformation (MFST) and modified seasonal-trend decomposition using LOESS transformation (MSTLT) to electricity load forecasting and compares the performance of two alternative models: the ARIMA(p,d,q) SARIMA(P,D,Q) model and the support vector regression (SVR) model. The data comprise monthly electricity consumption volumes between 2002 and 2019. The data between 2002 and 2018 are utilized to construct the forecasting model, while those in 2019 are employed to test the accuracy of the predicted values. The results confirm the validity of the proposed model in terms of forecasting accuracy and interpretability.
Daily peak load forecasting (DPLF) and total daily load forecasting (TDLF) are essential for optimal power system operation from one day to one week later. This study develops a Cubist-based incremental learning model to perform accurate and interpretable DPLF and TDLF. To this end, we employ time-series cross-validation to effectively reflect recent electrical load trends and patterns when constructing the model. We also analyze variable importance to identify the most crucial factors in the Cubist model. In the experiments, we used two publicly available building datasets and three educational building cluster datasets. The results showed that the proposed model yielded averages of 7.77 and 10.06 in mean absolute percentage error and coefficient of variation of the root mean square error, respectively. We also confirmed that temperature and holiday information are significant external factors, and electrical loads one day and one week ago are significant internal factors.
With the emergence of advanced computational technologies, the capacity to process data for developing machine learning-based predictive models has increased multifold. However, reliance on the model’s mere accuracy has swiftly shifted attention away from its interpretability. Resultantly, a need has emerged amongst forecasters and academics to have predictive models that are not only accurate but also interpretable as well. Therefore, to facilitate energy forecasters, this paper advances the knowledge of short-term load forecasting through generalized regression analysis using high degree polynomials and cross terms. To predict the irregularly changing energy demand at the consumer level, the proposed model uses a time series of an hourly load of three years of an electricity distribution company in Pakistan. Two variants of regression analysis are used: (a) generalized linear regression model (GLRM), and (b) generalized linear regression model with polynomials and cross-terms (GLRM-PCT) for comparative reasons. Experiments revealed that GLRM-PCT showed higher forecasting accuracy across a variety of performance metrics such as mean absolute percentage error (MAPE), mean absolute error (MAE), root mean squared error (RMSE), and r-squared values. Moreover, the enhanced interpretability of GLRM-PCT also explained a wide range of combinations of weather variables, public holidays, as well as lagged load and climatic variables.
Accurately forecasting electricity load have great impact on a number of departments in power systems. Compared to electricity load simulation (white-box model), electricity load forecasting (black-box model) does not require expertise in building construction. The development cycle of the electricity load forecasting model is much shorter than the design cycle of the electricity load simulation. Recent developments in machine learning have lead to the creation of models with strong fitting and accuracy to deal with nonlinear characteristics. Based on the real load data set, this paper evaluates and compares the two mainstream short-term load forecasting techniques. Before the experiment, this paper first enumerates the common methods of short-term load forecasting and explains the principles of Long Short-term Memory Networks (LSTMs) and Support Vector Machines (SVM) used in this paper. Secondly, based on the characteristics of the electricity load dataset, data pre-processing and feature selection takes place. This paper describes the results of a controlled experiment to study the importance of feature selection. The LSTMs model and SVM model are applied to one-hour ahead load forecasting and one-day ahead peak and valley load forecasting. The predictive accuracy of these models are calculated based on the error between the actual and predicted loads, and the runtime of the model is recorded. The results show that the LSTMs model have a higher prediction accuracy when the load data is sufficient. However, the overall performance of the SVM model is better when the load data used to train the model is insufficient and the time cost is prioritised.
The intelligent management of power in electrical utilities depends on the high significance of load forecasting models. Since the industries are digitalized, power generation is supported by a variety of resources. Therefore, the forecasting accuracy of different models varies. The power utilities with different generation modalities (DGM) experience complexities and a noticeable amount of error in predicting future electrical consumption. To effectively manage the power flow with negligible power interruptions, a utility must utilize the forecasting tools to predict the future electricity demand with minimum error. Since the current literature supports individual and limited power sources involved in generation for load forecasting, thus the utilities with multiple power sources or DGM remain unexplored. Therefore, exploration of existing literature is required relating to analyzing the existing models which could be considered in load forecasting for DGM. This paper explores state-of-art methods recently utilized for electrical load forecasting highlighting the common practices, recent advances, and exposure of areas available for improvement. The review investigates the methods, parameters, and respective sectors considered for load forecasting. It performs in-depth analysis and discusses the strengths, weaknesses, and error percentages of models. It also highlights the peculiarities of methods used in residential, commercial, industrial, grid, and off-grid sectors aiming to help the researchers to appraise the common practices. Moreover, trends and research gaps are also discussed.
Many clinicians remain wary of machine learning due to long-standing concerns about “black box” models. “Black box” is shorthand for models that are sufficiently complex that they are not straightforwardly interpretable to humans. Lack of interpretability in predictive models can undermine trust in those models, especially in health care where so many decisions are literally life and death. There has recently been an explosion of research in the field of explainable machine learning aimed at addressing these concerns. The promise of explainable machine learning is considerable, but it is important for cardiologists who may encounter these techniques in clinical decision support tools or novel research papers to have a critical understanding of both their strengths and their limitations. This paper reviews key concepts and techniques in the field of explainable machine learning as they apply to cardiology. Key concepts reviewed include interpretability versus explainability and global versus local explanations. Techniques demonstrated include permutation importance, surrogate decision trees, local interpretable model-agnostic explanations, and partial dependence plots. We discuss several limitations with explainability techniques, focusing on the how the nature of explanations as approximations may omit important information about how black box models work and why they make certain predictions. We conclude by proposing a rule of thumb about when it is appropriate to use black box models with explanations, rather than interpretable models.
In recent years, there has been a rapidly expanding focus on explaining the predictions made by black-box AI systems that handle image and tabular data. However, considerably less attention has been paid to explaining the predictions of opaque AI systems handling time series data. In this paper, we advance a novel model-agnostic, case-based technique – Native Guide – that generates counterfactual explanations for time series classifiers. Given a query time series, , for which a black-box classification system predicts class, c, a counterfactual time series explanation shows how could change, such that the system predicts an alternative class, . The proposed instance-based technique adapts existing counterfactual instances in the case-base by highlighting and modifying discriminative areas of the time series that underlie the classification. Quantitative and qualitative results from two comparative experiments indicate that Native Guide generates plausible, proximal, sparse and diverse explanations that are better than those produced by key benchmark counterfactual methods.
Deep neural networks such as Convolutional Neural Networks (CNNs) have been successfully applied to a wide variety of tasks, including time series forecasting.
In this paper, we propose a novel approach for online deep CNN selection using saliency maps in the task of time series forecasting. We start with an arbitrarily set of different CNN forecasters with various architectures. Then, we outline a gradient-based technique for generating saliency maps with a coherent design to make it able to specialize the CNN forecasters across different regions in the input \timeseries using a performance-based ranking.
In this framework, the selection of the adequate model is performed in an online fashion and the computation of saliency maps responsible for the model selection is achieved adaptively following drift detection in the \timeseries. In addition, the saliency maps can be exploited to provide suitable explanations for the reason behind selecting a specific model at a certain time interval or instant.
An extensive empirical study on various real-world datasets demonstrates that our method achieves excellent or on par results in comparison to the state-of-the-art approaches as well as several baselines.
Load forecasting has always been an essential part of power system planning and operation. In recent decades, the competition of the market and the requirements of renewable integration lead more attention to probabilistic load forecasting methods, which can reflect forecasting uncertainties through prediction intervals and hence benefit decision-making activities in system operation. Moreover, with the development of smart grid and power metering techniques, power companies have collected enormous load data about electricity customers and substations. The abundant load data allow us to utilize medium voltage measurement data to achieve better accuracy in high voltage transmission substation load forecasting. In this paper, a bottom-up probabilistic forecasting method is proposed for high voltage transmission substation short-term load forecasting, in which the probability distributions of medium voltage day-ahead load forecasting values are estimated and added up to form high voltage load predictions. Two bottom-up frameworks based on load patterns collected from medium voltage outgoing lines and substations are proposed respectively, in which mismatches between load data at different levels are estimated for correcting high voltage predictions. The comparison of predictions obtained by traditional and bottom-up methods demonstrates that the proposed method obtains high voltage load forecasting more accurately and give narrower prediction intervals.
Smart electricity meters are a central feature of any future smart grid, and therefore represent a rapid and significant household energy transition, growing by our calculations from less than 23.5 million smart meters in 2010 to an estimated 729.1 million in 2019, a decadal growth rate of 3013%. What are the varying economic, governance, and energy and climate sustainability aspects associated with the diffusion of smart meters for electricity? What lessons can be learned from the ongoing rollouts of smart meters around the world? Based on an original dataset twice as comprehensive as the current state of the art, this study examines smart meter deployment across 41 national programs and 61 subnational programs that collectively target 1.49 billion installations involving 47 countries. In addition to rates of adoption and the relative influence of factors such as technology costs, we examine adoption requirements, modes of information provision, patterns of incumbency and management, behavioral changes and energy savings, emissions reductions, policies, and links to other low-carbon transitions such as energy efficiency or renewable energy. We identify numerous weak spots in the literature, notably the lack of harmonized datasets as well as inconsistent scope and quality within national cost-benefit analyses of smart meter programs. Most smart meters have a lifetime of only 20 years, leading to future challenges concerning repair, care, and waste. National-scale programs (notably China) account for a far larger number of installations than subnational ones, and national scale programs also install smart meters more affordably, i.e. with lower general costs. Finally, the transformative effect of smart meters may be oversold, and we find that smart electricity meters are a technology that is complementary, rather than disruptive or transformative, one that largely does not challenge the dominant practices and roles of electricity suppliers, firms, or network operators.
Probabilistic load forecasting (PLF) has become necessary for power system operators to do efficient planning across power transmission and distribution systems. However, there are not many PLF models, and those that exist take a lot of computation time and are not efficient, especially in multiple loads. This paper proposes a novel algorithm for spatially correlated multiple loads wherein a global parameter is learned from state-space parameters of individual loads by an amalgamation of deep neural networks and state-space models. The proposed model employs complex pattern learning capabilities of recurrent neural networks and temporal pattern extraction of innovation state-space models. It is tested on GEFCom-14 and ISO-NE datasets, one with a single load and multiple loads. Different case studies are conducted to examine the involvement of temperature for load forecasting. It has been observed that in the case of multivariate loads, temperature variable doesn’t make much difference in PLF, but in the case of univariate loads, forecasting results are four-times better. The proposed method is highly interpretable and can be employed in areas where limited training data is available to the areas where colossal data is available. The proposed model has outperformed several benchmarks present in the literature on the same datasets.
Microgrids have recently emerged as a building block for smart grids combining
distributed renewable energy sources (RESs), energy storage devices, and load
management methodologies. The intermittent nature of RESs brings several
challenges to the smart microgrids, such as reliability, power quality, and balance between supply and demand. Thus, forecasting power generation from
RESs, such as wind turbines and solar panels, is becoming essential for the
efficient and perpetual operations of the power grid and it also helps in attaining optimal utilization of RESs. Energy demand forecasting is also an integral
part of smart microgrids that helps in planning the power generation and energy trading with commercial grid. Machine learning (ML) and deep learning
(DL) based models are promising solutions for predicting consumers’ demands
and energy generations from RESs. In this context, this manuscript provides
a comprehensive survey of the existing DL-based approaches, which are developed for power forecasting of wind turbines and solar panels as well as electric
power load forecasting. It also discusses the datasets used to train and test
the different DL-based prediction models, enabling future researchers to identify appropriate datasets to use in their work. Even though there are a few
related surveys regarding energy management in smart grid applications, they
are focused on a specific production application such as either solar or wind.
Moreover, none of the surveys review the forecasting schemes for production
and load side simultaneously. Finally, previous surveys do not consider the
datasets used for forecasting despite their significance in DL-based forecasting
approaches. Hence, our survey work is intrinsically different due to its data centered view, along with presenting DL-based applications for load and energy
generation forecasting in both residential and commercial sectors. The comparison of different DL approaches discussed in this manuscript reveals that the
efficiency of such forecasting methods is highly dependent on the amount of
the historical data and thus a large number of data storage devices and high
processing power devices are required to deal with big data. Finally, this study
raises several open research problems and opportunities in the area of renewable
energy forecasting for smart microgrids.
As new grid edge technologies emerge—such as rooftop solar panels, battery storage, and controllable water heaters—quantifying the uncertainties of building load forecasts is becoming more critical. The recent adoption of smart meter infrastructures provided new granular data streams, largely unavailable just ten years ago, that can be utilized to better forecast building-level demand. This paper uses Bayesian Structural Time Series for probabilistic load forecasting at the residential building level to capture uncertainties in forecasting. We use sub-hourly electrical submeter data from 120 residential apartments in Singapore that were part of a behavioral intervention study. The proposed model addresses several fundamental limitations through its flexibility to handle univariate and multivariate scenarios, perform feature selection, and include either static or dynamic effects, as well as its inherent applicability for measurement and verification. We highlight the benefits of this process in three main application areas: (1) Probabilistic Load Forecasting for Apartment-Level Hourly Loads; (2) Submeter Load Forecasting and Segmentation; (3) Measurement and Verification for Behavioral Demand Response. Results show the model achieves a similar performance to ARIMA, another popular time series model, when predicting individual apartment loads, and superior performance when predicting aggregate loads. Furthermore, we show that the model robustly captures uncertainties in the forecasts while providing interpretable results, indicating the importance of, for example, temperature data in its predictions. Finally, our estimates for a behavioral demand response program indicate that it achieved energy savings; however, the confidence interval provided by the probabilistic model is wide. Overall, this probabilistic forecasting model accurately measures uncertainties in forecasts and provides interpretable results that can support building managers and policymakers with the goal of reducing energy use.
Numerous deep learning architectures have been developed to accommodate the diversity of time-series datasets across different domains. In this article, we survey common encoder and decoder designs used in both one-step-ahead and multi-horizon time-series forecasting—describing how temporal information is incorporated into predictions by each model. Next, we highlight recent developments in hybrid deep learning models, which combine well-studied statistical models with neural network components to improve pure methods in either category. Lastly, we outline some ways in which deep learning can also facilitate decision support with time-series data.
This article is part of the theme issue ‘Machine learning for weather and climate modelling’.
Various literature surveys state and confirm a rapid increase in research on explainable artificial intelligence (XAI) in recent years. One possible motivation for this change are legal regulations, including the general data protection regulation (GDPR) but also similar regulations outside of Europe. Another possible reason is the decreasing trust in machine learning systems since both their algorithms and they models they include are often opaque. The desire to retrieve an explanation for a given decision reaches back to the era of expert systems in the 1980s. Decisions made by experts often rely on their stored experiences, yet most XAI approaches cannot provide explanations based on specific experiences because they do not retain them. In contrast, explainable case-based reasoning (XCBR) approaches can provide such explanations , and thus is of interest to XAI researchers. We present a taxonomy of XCBR approaches by categorizing and presenting current methodologies and implementations based on an extensive literature review. This taxonomy can be used by XAI researchers and CBR researchers who are explicitly interested in the generation and use of explanations.
Over the last two decades, Artificial Intelligence (AI) approaches have been applied to various applications of the smart grid, such as demand response, predictive maintenance, and load forecasting. However, AI is still considered to be a "black-box" due to its lack of explainability and transparency, especially for something like solar photovoltaic (PV) forecasts that involves many parameters. Explainable Artificial Intelligence (XAI) has become an emerging research field in the smart grid domain since it addresses this gap and helps understand why the AI system made a forecast decision. This paper presents several use cases of solar PV energy forecasting using XAI tools, such as LIME, SHAP, and ELI5, which can contribute to adopting XAI tools for smart grid applications. Understanding the inner workings of a prediction model based on AI can give insights into the application field. Such insight can provide improvements to the solar PV forecasting models and point out relevant parameters.
The economic growth of every nation is highly related to its electricity infrastructure,
network, and availability since electricity has become the central part of everyday life
in this modern world. Hence, the global demand for electricity for residential and commercial purposes have seen an incredible increase. On the other side, electricity prices keep fluctuating over the past years and not mentioning the inadequacy in electricity generation to meet global demand. As a solution to this, numerous studies aimed at estimating future electrical energy demand for residential and commercial purposes to enable electricity generators, distributors, and suppliers to plan effectively ahead and promote energy conservation among the users. Notwithstanding, load forecasting is one of the major problems facing the power industry since the inception of electric power. The current study tried to undertake a systematic and critical review of about seventy-seven (77) relevant previous works reported in academic journals over nine
years (2010–2020) in electricity demand forecasting. Specifically, attention was given
to the following themes: (i) The forecasting algorithms used and their fitting ability in
this field, (ii) the theories and factors affecting electricity consumption and the origin
of research work, (iii) the relevant accuracy and error metrics applied in electricity load
forecasting, and (iv) the forecasting period. The results revealed that 90% out of the
top nine models used in electricity forecasting was artificial intelligence-based, with
artificial neural network (ANN) representing 28%. In this scope, ANN models were primarily used for short-term electricity forecasting where electrical energy consumption
patterns are complicated. Concerning the accuracy metrics used, it was observed that
root-mean-square error (RMSE) (38%) was the most used error metric among electricity
forecasters, followed by mean absolute percentage error MAPE (35%). The study further
revealed that 50% of electricity demand forecasting was based on weather and eco‑
nomic parameters, 8.33% on household lifestyle, 38.33% on historical energy consumption, and 3.33% on stock indices. Finally, we recap the challenges and opportunities for further research in electricity load forecasting locally and globally.
Recent years have seen an increasing interest in Demand Response (DR) as a means to provide flexibility, and hence improve the reliability of energy systems in a cost-effective way. Yet, the high complexity of the tasks associated with DR, combined with their use of large-scale data and the frequent need for near real-time decisions, means that Artificial Intelligence (AI) and Machine Learning (ML) — a branch of AI — have recently emerged as key technologies for enabling demand-side response. AI methods can be used to tackle various challenges, ranging from selecting the optimal set of consumers to respond, learning their attributes and preferences, dynamic pricing, scheduling and control of devices, learning how to incentivise participants in the DR schemes and how to reward them in a fair and economically efficient way. This work provides an overview of AI methods utilised for DR applications, based on a systematic review of over 160 papers, 40 companies and commercial initiatives, and 21 large-scale projects. The papers are classified with regards to both the AI/ML algorithm(s) used and the application area in energy DR. Next, commercial initiatives are presented (including both start-ups and established companies) and large-scale innovation projects, where AI methods have been used for energy DR. The paper concludes with a discussion of advantages and potential limitations of reviewed AI techniques for different DR tasks, and outlines directions for future research in this fast-growing area.
Deep learning (DL) approaches are being increasingly used for time-series forecasting, with many efforts devoted to designing complex DL models. Recent studies have shown that the DL success is often attributed to effective data representations, fostering the fields of feature engineering and representation learning. However, automated approaches for feature learning are typically limited with respect to incorporating prior knowledge, identifying interactions among variables, and choosing evaluation metrics to ensure that the models are reliable. To improve on these limitations, this paper contributes a novel visual analytics framework, namely
TimeTuner
, designed to help analysts understand how model behaviors are associated with localized correlations, stationarity, and granularity of time-series representations. The system mainly consists of the following two-stage technique: We first leverage counterfactual explanations to connect the relationships among time-series representations, multivariate features and model predictions. Next, we design multiple coordinated views including a partition-based correlation matrix and juxtaposed bivariate stripes, and provide a set of interactions that allow users to step into the transformation selection process, navigate through the feature space, and reason the model performance. We instantiate
TimeTuner
with two transformation methods of smoothing and sampling, and demonstrate its applicability on real-world time-series forecasting of univariate sunspots and multivariate air pollutants. Feedback from domain experts indicates that our system can help characterize time-series representations and guide the feature engineering processes.
Reinforcement learning (RL) has shown significant success in sequential decision making in fields like autonomous vehicles, robotics, marketing and gaming industries. This success has attracted the attention to the RL control approach for building energy systems which are becoming complicated due to the need to optimize for multiple, potentially conflicting, goals like occupant comfort, energy use and grid interactivity. However, for real world applications, RL has several drawbacks like requiring large training data and time, and unstable control behavior during the early exploration process making it infeasible for an application directly to building control tasks. To address these issues, an imitation learning approach is utilized herein where the RL agents starts with a policy transferred from accepted rule based policies and heuristic policies. This approach is successful in reducing the training time, preventing the unstable early exploration behavior and improving upon an accepted rule-based policy — all of these make RL a more practical control approach for real world applications in the domain of building controls.
This study utilizes machine learning and, more specifically, reinforcement learning (RL) to allow for an optimized, real-time operation of large numbers of decentral flexible assets on private household scale in the electricity domain. The potential and current obstacles of RL are demonstrated and a guide for interested practitioners is provided on how to tackle similar tasks without advanced skills in neuronal network programming. For the application in the energy domain it is demonstrated that state-of-the-art RL algorithms can be trained to control potentially millions of small-scale assets in private households. In detail, the applied RL algorithm outperforms common heuristic algorithms and only falls slightly short of the results provided by linear optimization, but at less than a thousandth of the simulation time. Thus, RL paves the way for aggregators of flexible energy assets to optimize profit over multiple use cases in a smart energy grid and thus also provide valuable grid services and a more sustainable operation of private energy assets.
Methods to forecast electric loads and generation timeseries are widely applied in power system operation and balancing. For this purpose, most sophisticated forecasting methods are complex, since the net electricity consumption is not dependent on and explainable by a single cause. The increasing complexity furthers a trend towards machine-learning methods that achieve more accurate load forecasts with exogenous features. However, a well-known downside of machine learning is that forecast users will face non-transparent, and inexplicable models that are difficult to relate to. In this paper, we propose a graphical representation of the most salient exogenous features as functions of time for day-ahead residual load forecasting. The presented work extends a sequence-to-sequence recurrent neural network model to visually accommodate the explanation of the produced forecast. The resulting saliency maps reduce the high complexity of the input–output relationships to the two-dimensional plane of features over time steps.
Integrated with the state-of-the-art technologies, Artificial Intelligence (AI) has been successfully applied to diverse industries thanks to the increased availability of data and computing power. As AI applications become more challenging to solve real world problems, various explainable AI techniques have also been developed to interpret AI models. In this paper, we conduct SHAP value analysis, which is one of the explainable AI, to assess the feature importance of the load forecasting model. We point out the limitation of the conventional SHAP value-based feature importance metric and propose a new metric which incorporates the coefficient of determination to consider the distribution of the SHAP values. To evaluate the proposed metric, we conduct feature importance experiments on the XGBoost-based 24-h load forecasting model trained with Korea Power Exchange data. The experimental results show that the proposed SHAP value-based feature importance metric is more relevant in terms of the performance of load forecasting.
With synergies among multiple energy sectors, integrated energy systems (IESs) have been recognized lately as an effective approach to accommodate large-scale renewables and achieve environmental sustainability. The core of IES operation is to keep energy balance between supply and demand, where accurate load forecasting serves as one of the most crucial cornerstones. Recent advances in data-driven techniques have spawned a whole new branch of solution for load forecasting in IESs, which urges the need for a timely review accordingly. First, this overview reveals the uniqueness of the IES load forecasting problem compared with the conventional problem in electric power systems. The influential factors are much more complicated and volatile, while multivariate load series are forecasted simultaneously to address the coupling among different energy sectors. This uniqueness has contributed to increasing works and early breakthroughs for the IES load forecasting problem. Then, following the application and implementation procedures, essential issues of data-driven techniques in current works are reviewed with respect to the IES settings such as the variable decision, data preparation, feature engineering, model identification, and augmentation strategy adoption. The procedures are summarized according to current works and have covered all of the effective solutions for accurate forecasts. Finally, future trends and prospects of advanced topics therein are identified beyond current breakthroughs. Compatible with the distributed structure of IESs, federated learning is a promising solution for coordinated load forecasting among diverse energy sectors. On the other hand, automated machine learning builds deep learning and other data-driven models more intelligently to extremely improve load forecasting in complex IESs. The limited data issue in IESs also warrants further research efforts.
Despite widespread adoption and outstanding performance, machine learning models are considered as “black boxes”, since it is very difficult to understand how such models operate in practice. Therefore, in the power systems field, which requires a high level of accountability, it is hard for experts to trust and justify decisions and recommendations made by these models. Meanwhile, in the last couple of years, Explainable Artificial Intelligence (XAI) techniques have been developed to improve the explainability of machine learning models, such that their output can be better understood. In this light, it is the purpose of this paper to highlight the potential of using XAI for power system applications. We first present the common challenges of using XAI in such applications and then review and analyze the recent works on this topic, and the on-going trends in the research community. We hope that this paper will trigger fruitful discussions and encourage further research on this important emerging topic.
Deep learning based forecasting methods have become the methods of choice in many applications of time series prediction or forecasting often outperforming other approaches. Consequently, over the last years, these methods are now ubiquitous in large-scale industrial forecasting applications and have consistently ranked among the best entries in forecasting competitions (e.g., M4 and M5). This practical success has further increased the academic interest to understand and improve deep forecasting methods. In this article we provide an introduction and overview of the field: We present important building blocks for deep forecasting in some depth; using these building blocks, we then survey the breadth of the recent deep forecasting literature.
Memristive LSTM networks have been proven as a powerful Neuromorphic Computing Architecture (NCA) for various time series forecasting tasks and are recognized as the next generation of AI. However, a lack of model explainability makes it hard to properly interpret forecasting results for existing memristive LSTM networks, which makes this NCA unreliable, unaccountable and untrustworthy. In this paper, an interpretable memristive (IM) LSTM network design is proposed for time series forecasting, where the mixture attention technique is embedded into IM-LSTM cells for characterizing the variable-wise feature and the temporal importance. The updating rules and training approach are also presented for this interpretable memristive LSTM network. We evaluate this approach on a probabilistic residential load forecasting task incorporating PV. By improving model interpretability, the most influential predictive factors can be verified by Built Environment domain experts, demonstrating the effectiveness of our design.
The surge of machine learning and increasing data accessibility in buildings provide great opportunities for applying machine learning to building energy system modeling and analysis. Building load prediction is one of the most critical components for many building control and analytics activities, as well as grid-interactive and energy efficiency building operation. While a large number of research papers exist on the topic of machine-learning-based building load prediction, a comprehensive review from the perspective of machine learning is missing. In this paper, we review the application of machine learning techniques in building load prediction under the organization and logic of the machine learning, which is to perform tasks T using Performance measure P and based on learning from Experience E.
Firstly, we review the applications of building load prediction model (task T). Then, we review the modeling algorithms that improve machine learning performance and accuracy (performance P). Throughout the papers, we also review the literature from the data perspective for modeling (experience E), including data engineering from the sensor level to data level, pre-processing, feature extraction and selection. Finally, we conclude with a discussion of well-studied and relatively unexplored fields for future research reference. We also identify the gaps in current machine learning application and predict for future trends and development.
Hierarchical load forecasting (HLF) is an approach to generate forecasts for hierarchical load time series. The performance of HLF can be improved by optimizing the
forecasting model
and the
hierarchical structure
. Previous studies mostly focus on the forecasting model while the hierarchical structure is usually formed by clustering of natural attributes like geography, customer type, or the similarities between load profiles. A major limitation of these natural hierarchical structures is the mismatched objectives between clustering and forecasting. Clustering aims to minimize the dissimilarity among customers of a group while forecasting aims to minimize their forecasting errors. The two independent optimizations could limit the overall forecasting performance. Hence, this paper attempts to integrate the hierarchical structure and the forecasting model by a novel closed-loop clustering (CLC) algorithm. It links the objectives of forecasting and clustering by a feedback mechanism to return the goodness-of-fit as the criterion for the clustering. In this way, the hierarchical structure is enhanced by re-assigning the cluster membership and the parameters of the forecasting models are updated iteratively. The method is comparatively assessed with existing HLF methods. Using the same forecasting model, the proposed hierarchical structure outperforms the bottom-up structure by 52.20%, ensemble-based structure by 26.89%, load-profile structures by 19.90%, respectively.
Industrial load takes a big portion of the total electricity demand. Skilled probabilistic industrial load forecasts allow for optimally exploiting energy resources, managing the reserves, and market bidding, which are beneficial to transmission and distribution system operators and their industrial customers. Despite its importance, industrial load forecasting has never been a popular subject in the literature. Most existing methods operate on the active power alone, partially or totally neglecting the reactive power. This paper proposes a multivariate approach to probabilistic industrial load forecasting, which addresses active and reactive power simultaneously. The proposed method is based on a two-level procedure, which consists of generating probabilistic forecasts individually for active and reactive power through univariate probabilistic models, and combining these forecasts in a multivariate approach based on a multivariate quantile regression model. The procedure to estimate the parameters of the multivariate quantile regression model is posed in this paper under a linear programming problem, to facilitate the convergence to the optimal solution. The proposed method is validated using actual load data collected at an Italian factory, under comparison with several probabilistic benchmarks. The proposed multivariate method enhances the skill of forecasts by 6% to 13.5%, with respect to univariate benchmarks.
Over the past decades, electric power systems (EPSs) have undergone an evolution from an ordinary bulk structure to intelligent flexible systems by way of advanced electronics and control technologies. Moreover, EPS has become a more complex, unstable and nonlinear structure with the integration of distributed energy resources in comparison with traditional power grids. Unlike classical approaches, physical methods, statistical approaches and computer calculation techniques are commonly used to solve EPS problems. Artificial intelligent (AI) techniques have especially been used recently in many fields. Deep neural networks have become increasingly attractive as an AI approach due to their robustness and flexibility in handling nonlinear complex relationships on large scale data sets. Major deep learning concepts addressing some problems in EPS have been reviewed in the present study by a comprehensive literature survey. The practices of deep learning and its combinations are well organized with up‐to‐date references in various fields such as load forecasting, wind and solar power forecasting, power quality disturbances detection and classifications, fault detection power system equipment, energy security, energy management and energy optimization. Furthermore, the difficulties encountered in implementation and the future trends of this method in EPS are discussed subject to the findings of current studies. It concludes that deep learning has a huge application potential on EPS, due to smart technologies integration that will increase considerably in the future.