Conference Paper

A New Entropy-based Feature Selection Method for Load Forecasting in Smart Homes

Authors:
  • Edo State University Iyamho
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

This paper addresses the challenges of load forecasting that occur due to the complex nature of load in different predicting horizons and as well as the total consumption within these horizons. It is not often easy to accurately fit the several complex factors that are faced with demand for electricity into the predicting models. More so, due to the dynamic nature of these complex factors (i.e., temperature, humidity and other factors that influence consumption), it is difficult to derive an accurate demand forecast based on these parameters. As a consequence, a model that uses hourly electricity loads and temperature data to forecast the next hourly loads is proposed. The model is based on modified entropy mutual information based feature selection to remove irrelevancy and redundancy from the dataset. Conditional restricted Boltzmann machine (CRBM) is investigated to perform load forecasting; accuracy and convergence are improved to reduce the CRBM's forecast error via a Jaya based meta-heuristic optimization algorithm. The proposed model is implemented on the publicly available dataset of GEFCom2012 of the US utility. Comparative analysis is carried out on an existing accurate, fast converging short-term load forecasting (AFC-STLF) model since it has a similar architecture to the proposed model. Simulation results confirm that the proposed model improves the accuracy up to 56.32% as compared to 43.67% of AFC-STLF. Besides, the proposed model reduces the average execution time up to 53.87% as compared to 46.12% of AFC-STLF.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... All abbreviations used throughout this work are presented at the end of the paper. This paper is the extension of our previous work [20]. The contributions of our paper are as follows. ...
... (16-18 June 2007) is higher than the electrical load consumption during the working days, which occur when cooling loads are mostly active. The sudden decrease of consumers' load means that the utility restricts the use of power by means of switching off the energy supply in these periods (20)(21)(22)(23). Another plausible explanation could be that a higher temperature can spur load demand as the air conditioning continues to operate. ...
Article
Full-text available
Over the last decades, load forecasting is used by power companies to balance energy demand and supply. Among the several load forecasting methods, medium-term load forecasting is necessary for grid’s maintenance planning, settings of electricity prices, and harmonizing energy sharing arrangement. The forecasting of the month ahead electrical loads provides the information required for the interchange of energy among power companies. For accurate load forecasting, this paper proposes a model for medium-term load forecasting that uses hourly electrical load and temperature data to predict month ahead hourly electrical loads. For data preprocessing, modified entropy mutual information-based feature selection is used. It eliminates the redundancy and irrelevancy of features from the data. We employ the conditional restricted Boltzmann machine (CRBM) for the load forecasting. A meta-heuristic optimization algorithm Jaya is used to improve the CRBM’s accuracy rate and convergence. In addition, the consumers’ dynamic consumption behaviors are also investigated using a discrete-time Markov chain and an adaptive k-means is used to group their behaviors into clusters. We evaluated the proposed model using GEFCom2012 US utility dataset. Simulation results confirm that the proposed model achieves better accuracy, fast convergence, and low execution time as compared to other existing models in the literature.
... In this chapter, the solutions to problems in Section 1.4 are addressed and published in [116,117]. Here, the problems are poor feature selection, inefficient forecasting model and dynamism of the behavior of consumers. So, an entropy mutual information based feature selection method that can handle both linear and non-linear electricity load data is proposed. ...
Thesis
Full-text available
This thesis examines the privacy preserving energy management issue, taking into account both energy generation units and responsive demand in the smart grids. Firstly, because of the inherent stochastic behavior of the distributed energy resources, an optimal energy management problem is studied. Distributed energy resources are used in the decentralization of energy systems. Large penetration of distributed energy resources without the precise cybersecurity measures, such as privacy, monitoring and trustworthy communication may jeopardize the energy system and cause outages, and reliability problem for consumers. Therefore, a blockchain based decentralized energy system to accelerate electrification by improving service delivery while minimizing the cost of generation and addressing historical antipathy and cybersecurity risk is proposed. A case study of sub-Sahara Africa is considered. Also, a blockchain based energy trading system is proposed, which includes price negotiation and incentive mechanisms to address the imbalance of order. Besides, the Internet of energy makes it possible to integrate distributed energy resources and consumers. However, as the number of users involved in energy transactions increases, some factors are restricting conventional centralized energy trading. These factors include lack of trust, privacy, fixed energy pricing, and demurrage fees dispute. Therefore, additive homomorphic encryption and consortium blockchain are explored in this thesis to provide privacy and trust. Additionally, a dynamic energy pricing model is formulated based on the load demand response ratio of prosumers to address the fixed energy pricing problem. The proposed dynamic pricing model includes demurrage fees, which is a monetary penalty imposed on a prosumer if it failed to deliver energy within the agreed duration. Also, a new threat model is designed and analyzed. Secondly, mobile prosumers, such as electric vehicles offer a wide range of sophisticated services that contribute to the robustness and energy efficiency of the power grid. As the number of vehicles in the smart grid grows, it potentially exposes vehicle owners to a range of location related privacy threats. For example, when making payments, the location of vehicles is typically revealed during the charging process. Also, fixed pricing policy and lack of trust may restrict energy trading between vehicles and charging stations. Therefore, a private blockchain system is proposed to preserve the privacy of vehicle owners from linking based attack while a public blockchain system is established to enhance energy trading. Various parameters are used to formulate a demand based pricing policy for vehicles, such as time of demand, types of vehicles and locations. Using the demand based pricing policy, an optimal scheduling method is designed to maximize the vehicles both social welfare and utility. An improved consensus energy management algorithm is proposed to protect the privacy of vehicle owners by applying differential privacy. The proposed system is robust against temporal and spatial location based privacy related attacks. Thirdly, blockchain is an evolving decentralized data collection technology, which costeffectively exploits residential homes to collate large amounts of data. The problems of blockchain are the inability to withstand malicious nodes, which provide misleading information that destabilize the entire network, lack of privacy for individual node and shared data inaccuracy. Therefore, a secure system for energy users to share their multi-data using the consortium blockchain is proposed. In this system, a credibility based Byzantine fault tolerance algorithm is employed as the blockchain consensus mechanism to achieve the fault tolerance of the system. Also, a recurrent neural network is used by certain honest users with credibility to forecast the energy usage of other honest users. A recurrent neural network operates on the collated data without revealing the private information about honest users and its gradient parameters. Moreover, additive homomorphic encryption is used in the recurrent neural network to secure the collated data and the gradient parameters of the network. Also, a credibility management system is proposed to prevent malicious users from attacking the system and it consists of two layers: upper and lower. The upper layer manages global credibility that reflects the overall readiness of honest users to engage in multi-data sharing. The lower layer performs local credibility that reflects certain feedback of honest users on the accuracy of the forecast data. Lastly, combining blockchain mining and application intensive tasks increases the computational cost for resource constrained energy users. Besides, the anonymity and privacy problems of the users are not completely addressed in the existing literature. Therefore, this thesis proposes an improved sparse neural network to optimize computation offloading cost for resource constrained energy users. Furthermore, a blockchain system based on garlic routing, known as GarliChain, is proposed to solve the problems of anonymity and privacy for energy users during energy trading in the smart grid. Furthermore, a trust method is proposed to enhance the credibility of nodes in the GarliChain network. Simulations evaluate the theoretical results and prove the effectiveness of the proposed solutions. From the simulation results, the performance of the proposed model and the least-cost option varies with the relative energy generation cost of centralized, decentralized and blockchain based decentralized system infrastructure. Case studies of Burkina Faso, Cote d’Ivoire, Gambia, Liberia, Mali, and Senegal illustrate situations that are more suitable for blockchain based decentralized system. For other sub-Sahara Africa countries, the blockchain based decentralized system can cost-effectively service a large population and regions. Additionally, the proposed blockchain based levelized cost of energy reduces energy costs by approximately 95% for battery and 75% for the solar modules. The future blockchain based levelized cost of energy varies across sub-Sahara Africa on an average of about 0.049 USD/kWh as compared to 0.15 USD/kWh of an existing system in the literature. The proposed model achieves low transaction cost, the minimum execution time for block creation, the transactional data privacy of prosumers and dispute resolution of demurrage fees. Moreover, the proposed system reduces the average system overhead cost up to 66.67% as compared to 33.43% for an existing scheme. Additionally, the proposed blockchain proof of authority consensus average hash power is minimized up to 82.75% as compared to 60.34% for proof of stake and 56.89% for proof of work consensus mechanisms. Simulations are also performed to evaluate the efficacy of the proposed demand based pricing policy for mobile prosumers. From the simulation results, the proposed demand based pricing policy is efficient in terms of both low energy price and average cost, high utility and social welfare maximization as compared to existing schemes in the literature. It means that about 89.23% energy price reduction is achieved for the proposed demand based pricing policy as compared to 83.46% for multi-parameter pricing scheme, 73.86% for fixed pricing scheme and 53.07% for the time of use pricing scheme. The vehicles minimize their operating costs up to 81.46% for the proposed demand based pricing policy as compared to 80.48% for multi-parameter pricing scheme, 69.75% for fixed pricing scheme and 68.29% for the time of use pricing scheme. Also, the proposed system outperforms an existing work, known as blockchain based secure incentive scheme in terms of low energy prices and high utility. Furthermore, the proposed system achieves an average block transaction cost of 1.66 USD. Besides, after applying the differential privacy, the risk of privacy loss is minimum as compared to existing schemes. Furthermore, higher privacy protection of vehicles is attained with a lower information loss against multiple background knowledge of an attacker. To analyze the efficiency of the proposed system regarding multi-data sharing, an experimental assessment reveals that about 85% of honest users share their data with stringent privacy measures. The remaining 15% share their data without stringent privacy measures. Moreover, the proposed system operates at a low operating cost while the credibility management system is used to detect malicious users in the system. Security analysis shows that the proposed system is robust against 51% attack, transaction hacking attack, impersonation attack and the double spending attack. To evaluate the proposed system regarding energy management of resource constrained blockchain energy users, a Jaya optimization algorithm is used to accelerate the error convergence rate while reducing the number of connections between different layers of the neurons for the proposed improved sparse neural network. Furthermore, the security of the users is ensured using blockchain technology while security analysis shows that the system is robust against the Sybil attack. Moreover, the probability of a successful Sybil attack is zero as the number of attackers’ identities and computational capacities increases. Under different sizes of data to be uploaded, the proposed improved sparse neural network scheme has the least average computational cost and data transmission time as compared to deep reinforcement learning combined with genetic algorithm, and sparse evolutionary training and multi-layer perceptron schemes in the literature. Simulation results of the proposed GarliChain system show that the system remains stable as the number of path requests increases. Also, the proposed trust method is 50.56% efficient in detecting dishonest behavior of nodes in the network as compared to 49.20% of an existing fuzzy trust model. Under different sizes of the blocks, the computational cost of the forwarding nodes is minimum. Security analysis shows that the system is robust against both passive and active attacks. Malicious nodes are detected using the path selection model. Moreover, a comparative study of the proposed system with existing systems in the literature is provided.
Article
Full-text available
Background: With the development of smart grids, accurate electric load forecasting has become increasingly important as it can help power companies in better load scheduling and reduce excessive electricity production. However, developing and selecting accurate time series models is a challenging task as this requires training several different models for selecting the best amongst them along with substantial feature engineering to derive informative features and finding optimal time lags, a commonly used input features for time series models. Methods: Our approach uses machine learning and a long short-term memory (LSTM)-based neural network with various configurations to construct forecasting models for short to medium term aggregate load forecasting. The research solves above mentioned problems by training several linear and non-linear machine learning algorithms and picking the best as baseline, choosing best features using wrapper and embedded feature selection methods and finally using genetic algorithm (GA) to find optimal time lags and number of layers for LSTM model predictive performance optimization. Results: Using France metropolitan’s electricity consumption data as a case study, obtained results show that LSTM based model has shown high accuracy then machine learning model that is optimized with hyperparameter tuning. Using the best features, optimal lags, layers and training various LSTM configurations further improved forecasting accuracy. Conclusions: A LSTM model using only optimally selected time lagged features captured all the characteristics of complex time series and showed decreased Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) for medium to long range forecasting for a wider metropolitan area.
Article
Full-text available
This paper presents a model for optimal energy management under the time-of-use (ToU) and critical peak price (CPP) market in a microgrid. The microgrid consists of intermittent dispatchable distributed generators, energy storage systems, and multi-home load demands. The optimal energy management problem is a challenging task due to the inherent stochastic behavior of the renewable energy resources. In the past, medium-sized distributed energy resource generation was injected into the main grid with no feasible control mechanism to prevent the waste of power generated by a distributed energy resource which has no control mechanism, especially when the grid power limit is altered. Thus, a Jaya-based optimization method is proposed to shift dispatchable distributed generators within the ToU and CPP scheduling horizon. The proposed model coordinates the power supply of the microgrid components, and trades with the main grid to reduce its fuel costs, production costs, and also maximize the monetary profit from sales revenue. The proposed method is implemented on two microgrid operations: the standalone and grid-connected modes. The simulation results are compared with other optimization methods: enhanced differential evolution (EDE) and strawberry algorithm (SBA). Finally, simulation results show that the Jaya-based optimization method minimizes the fuel cost by up to 38.13%, production cost by up to 93.89%, and yields a monetary benefit of up to 72.78% from sales revenue.
Article
Full-text available
Energy management systems are designed to monitor, optimize, and control the smart grid energy market. Demand-side management, considered as an essential part of the energy management system, can enable utility market operators to make better management decisions for energy trading between consumers and the operator. In this system, a priori knowledge about the energy load pattern can help reshape the load and cut the energy demand curve, thus allowing a better management and distribution of the energy in smart grid energy systems. Designing a computationally intelligent load forecasting (ILF) system is often a primary goal of energy demand management. This study explores the state of the art of computationally intelligent (i.e., machine learning) methods that are applied in load forecasting in terms of their classification and evaluation for sustainable operation of the overall energy management system. More than 50 research papers related to the subject identified in existing literature are classified into two categories: namely the single and the hybrid computational intelligence (CI)-based load forecasting technique. The advantages and disadvantages of each individual techniques also discussed to encapsulate them into the perspective into the energy management research. The identified methods have been further investigated by a qualitative analysis based on the accuracy of the prediction, which confirms the dominance of hybrid forecasting methods, which are often applied as metaheurstic algorithms considering the different optimization techniques over single model approaches. Based on extensive surveys, the review paper predicts a continuous future expansion of such literature on different CI approaches and their optimizations with both heuristic and metaheuristic methods used for energy load forecasting and their potential utilization in real-time smart energy management grids to address future challenges in energy demand management.
Article
Full-text available
With the development of smart power grids, communication network technology and sensor technology, there has been an exponential growth in complex electricity load data. Irregular electricity load fluctuations caused by the weather and holiday factors disrupt the daily operation of the power companies. To deal with these challenges, this paper investigates a day-ahead electricity peak load interval forecasting problem. It transforms the conventional continuous forecasting problem into a novel interval forecasting problem, and then further converts the interval forecasting problem into the classification forecasting problem. In addition, an indicator system influencing the electricity load is established from three dimensions, namely the load series, calendar data, and weather data. A semi-supervised feature selection algorithm is proposed to address an electricity load classification forecasting issue based on the group method of data handling (GMDH) technology. The proposed algorithm consists of three main stages: (1) training the basic classifier; (2) selectively marking the most suitable samples from the unclassified label data, and adding them to an initial training set; and (3) training the classification models on the final training set and classifying the test samples. An empirical analysis of electricity load dataset from four Chinese cities is conducted. Results show that the proposed model can address the electricity load classification forecasting problem more efficiently and effectively than the FW-Semi FS (forward semi-supervised feature selection) and GMDH-U (GMDH-based semi-supervised feature selection for customer classification) models.
Article
Full-text available
Short-term power load forecasting is an important basis for the operation of integrated energy system, and the accuracy of load forecasting directly affects the economy of system operation. To improve the forecasting accuracy, this paper proposes a load forecasting system based on wavelet least square support vector machine and sperm whale algorithm. Firstly, the methods of discrete wavelet transform and inconsistency rate model (DWT-IR) are used to select the optimal features, which aims to reduce the redundancy of input vectors. Secondly, the kernel function of least square support vector machine LSSVM is replaced by wavelet kernel function for improving the nonlinear mapping ability of LSSVM. Lastly, the parameters of W-LSSVM are optimized by sperm whale algorithm, and the short-term load forecasting method of W-LSSVM-SWA is established. Additionally, the example verification results show that the proposed model outperforms other alternative methods and has a strong effectiveness and feasibility in short-term power load forecasting.
Article
Full-text available
The process of modernizing smart grid prominently increases the complexity and uncertainty in scheduling and operation of power systems, and, in order to develop a more reliable, flexible, efficient and resilient grid, electrical load forecasting is not only an important key but is still a difficult and challenging task as well. In this paper, a short-term electrical load forecasting model, with a unit for feature learning named Pyramid System and recurrent neural networks, has been developed and it can effectively promote the stability and security of the power grid. Nine types of methods for feature learning are compared in this work to select the best one for learning target, and two criteria have been employed to evaluate the accuracy of the prediction intervals. Furthermore, an electrical load forecasting method based on recurrent neural networks has been formed to achieve the relational diagram of historical data, and, to be specific, the proposed techniques are applied to electrical load forecasting using the data collected from New South Wales, Australia. The simulation results show that the proposed hybrid models can not only satisfactorily approximate the actual value but they are also able to be effective tools in the planning of smart grids.
Article
Full-text available
Integration of energy systems with information technologies has facilitated the realization of smart energy systems that utilize information to optimize system operation. To that end, crucial in optimizing energy system operation is the accurate, ahead-of-time forecasting of load demand. In particular, load forecasting allows planning of system expansion, and decision making for enhancing system safety and reliability. In this paper, the application of two types of kernel machines for medium term load forecasting (MTLF) is presented and their performance is recorded based on a set of historical electricity load demand data. The two kernel machine models and more specifically Gaussian process regression (GPR) and relevance vector regression (RVR) are utilized for making predictions over future load demand. Both models, i.e., GPR and RVR, are equipped with a Gaussian kernel and are tested on daily predictions for a 30-day-ahead horizon taken from the New England Area. Furthermore, their performance is compared to the ARMA(2,2) model with respect to mean average percentage error and squared correlation coefficient. Results demonstrate the superiority of RVR over the other forecasting models in performing MTLF.
Article
Full-text available
A simple yet powerful optimization algorithm is proposed in this paper for solving the constrained and unconstrained optimization problems. This algorithm is based on the concept that the solution obtained for a given problem should move towards the best solution and should avoid the worst solution. This algorithm requires only the common control parameters and does not require any algorithm-specific control parameters. The performance of the proposed algorithm is investigated by implementing it on 24 constrained benchmark functions having different characteristics given in Congress on Evolutionary Computation (CEC 2006) and the performance is compared with that of other well-known optimization algorithms. The results have proved the better effectiveness of the proposed algorithm. Furthermore, the statistical analysis of the experimental work has been carried out by conducting the Friedman’s rank test and Holm-Sidak test. The proposed algorithm is found to secure first rank for the ‘best’ and ‘mean’ solutions in the Friedman’s rank test for all the 24 constrained benchmark problems. In addition, for solving the constrained benchmark problems, the algorithm is also investigated on 30 unconstrained benchmark problems taken from the literature and the performance of the algorithm is found better.
Article
Full-text available
Selection of input features plays an important role in developing models for short-term load forecasting (STLF). Previous studies along this line of research have focused pre-dominantly on filter and wrapper methods. Given the potential value of a hybrid selection scheme that includes both filter and wrapper methods in constructing an appropriate pool of features, coupled with the general lack of success in employing filter or wrapper methods individually, in this study we propose a hybrid filter-wrapper approach for STLF feature selection. This proposed approach, which is believed to have taken full advantage of the strengths of both filter and wrapper methods, first uses the Partial Mutual Information based filter method to filter out most of the irrelevant and redundant features, and subsequently applies a wrapper method, implemented via a firefly algorithm, to further reduce the redundant features without degrading the forecasting accuracy. The well-established support vector regression is selected as the modeler to implement the proposed hybrid feature selection scheme. Real-world electricity load datasets from a North-American electric utility and the Global Energy Forecasting Competition 2012 have been used to test the performance of the proposed approach, and the experimental results show its superiority over selected counterparts.
Article
Short-term load forecasting (STLF) models are very important for electric industry in the trade of energy. These models have many applications in the day-today operations of electric utility(ies) like energy generation planning, load switching, energy purchasing, infrastructure maintenance and contract evaluation. A large variety of STLF models have been developed which trade-off between forecast accuracy and convergence rate. This paper presents an accurate and fast converging STLF (AFC-STLF) model for industrial applications in a smart grid. In order to improve the forecast accuracy, modifications are devised in two popular techniques: (1) mutual information (MI) based feature selection, and (2) enhanced differential evolution (EDE) algorithm based error minimization. On the other hand, the convergence rate of the overall forecast strategy is enhanced by devising modifications in the heuristic algorithm and in the training process of the artificial neural network (ANN). Simulation results show that accuracy of the newly proposed forecast model is 99.5% with moderate execution time, i.e., we have decreased the average execution of the existing Bi-level forecast strategy by 52.38%.
Article
The ultimate issue in electricity loads modelling is to improve forecasting accuracy as well as guarantee a robust prediction result, which will save considerable manual labor material resources and economic consumption. For addressing this challenge, many researchers are committed to investigating highly accurate forecasting models, and feature selection (FS) technologies are considered as a powerful tool to improve performance of models in many literatures. However, common FS technologies applied for Short-term load forecasting (STLF) ignore to select date information of the observed series as feature candidates and pay less attention to reduction rates of feature candidates, which will result in loss of date information and redundancy of features. Both drawbacks provide a significant roadblock for improving forecasting accuracy. Aiming to overcome both drawbacks and develop an effective model for STLF, this paper successfully investigates the date-framework strategy (DFS) to construct the pool of features and develops an FS technology, genetic algorithm binary improved cuckoo search (GABICS), to search a solution with the lowest reduction rate. Assigning the extreme learning machine (ELM) to be the forecast, GABICS-DFS-ELM not only obtains a minimum and effective subset of features but also has a satisfactory forecasting result with high accuracy and robustness.
Article
In this paper, a new forecasting model based on artificial immune system (AIS) is proposed. The model is used for short-term electrical load forecasting as an example of forecasting time series with multiple seasonal cycles. Artificial immune system learns to recognize antigens (AGs) representing two fragments of the time series: 1) fragment preceding the forecast (input vector) and 2) forecasted fragment (output vector). Antibodies as recognition units recognize AGs by selected features of input vectors and learn output vectors. In the test procedure, new AG with only input vector is recognized by some antibodies (ABs). Its output vector is reconstructed from activated ABs. The unique feature of the proposed AIS is the embedded property of local feature selection. Each AB learns in the clonal selection process its optimal subset of features (a paratope) to improve its recognition and prediction abilities. In the simulation studies the proposed model was tested on real power system data and compared with other AIS-based forecasting models as well as neural networks, autoregressive integrated moving average, and exponential smoothing. The obtained results confirm good performance of the proposed model.
Article
Temperature plays a key role in driving the electricity demand. We adopt the “recency effect”, a term drawn from psychology, to represent the fact that the electricity demand is affected by the temperatures of the preceding hours. In the load forecasting literature, the temperature variables are often constructed in the form of lagged hourly temperatures and moving average temperatures. In the past, computing power has limited the amount of temperature variables that can be used in a load forecasting model. In this paper, we present a comprehensive study to model the recency effect using a big data approach. We take advantage of modern computing power to answer a fundamental question: how many lagged hourly temperatures and/or moving average temperatures are needed in a regression model in order to capture the recency effect fully without compromising the forecasting accuracy? Using a case study based on data from the load forecasting track of the Global Energy Forecasting Competition 2012, we first demonstrate that a model with the recency effect outperforms its counterpart (a.k.a. Tao’s Vanilla Benchmark Model) by 18% to 21% for forecasting the load series at the top (aggregated) level. We then model the recency effect in order to customize load forecasting models at the bottom level of a geographic hierarchy, again showing a superiority over the benchmark model of 12% to 15% on average. Finally, we discuss four different implementations of the recency effect modeling by hour of a day. In addition, this paper also presents two interesting findings: 1) the naive models are not useful for benchmark purposes in load forecasting at aggregated level due to their lack of accuracy; and 2) slicing the data into 24 pieces to develop one model for each hour is not necessarily better than building one interaction regression model using all 24 hours together.
Article
To improve the design of the electricity infrastructure and the efficient deployment of distributed and renewable energy sources, a new paradigm for the energy supply chain is emerging, leading to the development of smart grids. There is a need to add intelligence at all levels in the grid, acting over various time horizons. Predicting the behavior of the energy system is crucial to mitigate potential uncertainties. An accurate energy prediction at the customer level will reflect directly in efficiency improvements in the whole system. However, prediction of building energy consumption is complex due to many influencing factors, such as climate, performance of thermal systems, and occupancy patterns. Therefore, current state-of-the-art methods are not able to confine the uncertainty at the building level due to the many fluctuations in influencing variables. As an evolution of artificial neural network (ANN)-based prediction methods, deep learning techniques are expected to increase the prediction accuracy by allowing higher levels of abstraction. In this paper, we investigate two newly developed stochastic models for time series prediction of energy consumption, namely Conditional Restricted Boltzmann Machine (CRBM) and Factored Conditional Restricted Boltzmann Machine (FCRBM). The assessment is made on a benchmark dataset consisting of almost four years of one minute resolution electric power consumption data collected from an individual residential customer. The results show that for the energy prediction problem solved here, FCRBM outperforms ANN, Support Vector Machine (SVM), Recurrent Neural Networks (RNN) and CRBM.
Article
Plenty of feature selection methods are available in literature due to the availability of data with hundreds of variables leading to data with very high dimension. Feature selection methods provides us a way of reducing computation time, improving prediction performance, and a better understanding of the data in machine learning or pattern recognition applications. In this paper we provide an overview of some of the methods present in literature. The objective is to provide a generic introduction to variable elimination which can be applied to a wide array of machine learning problems. We focus on Filter, Wrapper and Embedded methods. We also apply some of the feature selection techniques on standard datasets to demonstrate the applicability of feature selection techniques.
Article
The load forecasting problem is a complex nonlinear problem linked with social considerations, economic factors, and weather variations. In particular, load forecasting for holidays is a challenging task as only a small number of historical data is available for holidays compared with what is available for normal weekdays and weekends. This paper presents a fuzzy polynomial regression method with data selection based on Mahalanobis distance incorporating a dominant weather feature for holiday load forecasting. Selection of past weekday data relevant to a given holiday is critical for improvement of the accuracy of holiday load forecasting. In the paper, a data selection process incorporating a dominant weather feature is also proposed in order to improve the accuracy of the fuzzy polynomial regression method. The dominant weather feature for selection of historical data is identified by evaluating mutual information between various weather features and loads from season to season. The results of case studies are presented to show the effectiveness of the proposed method.
Article
This paper presents the development of a dynamic artificial neural network model (DAN2) for medium term electrical load forecasting (MTLF). Accurate MTLF provides utilities information to better plan power generation expansion (or purchase), schedule maintenance activities, perform system improvements, negotiate forward contracts and develop cost efficient fuel purchasing strategies. We present a yearly model that uses past monthly system loads to forecast future electrical demands. We also show that the inclusion of weather information improves load forecasting accuracy. Such models, however, require accurate weather forecasts, which are often difficult to obtain. Therefore, we have developed an alternative: seasonal models that provide excellent fit and forecasts without reliance upon weather variables. All models are validated using actual system load data from the Taiwan Power Company. Both the yearly and seasonal models produce mean absolute percent error (MAPE) values below 1%, demonstrating the effectiveness of DAN2 in forecasting medium term loads. Finally, we compare our results with those of multiple linear regressions (MLR), ARIMA and a traditional neural network model.
Article
Mid-term load forecasting (MTLF) becomes an essential tool for today power systems, mainly in those countries whose power systems operate in a deregulated environment. Among different kinds of MTLF, this paper focuses on the prediction of daily peak load for one month ahead. This kind of load forecast has many applications like maintenance scheduling, mid-term hydro thermal coordination, adequacy assessment, management of limited energy units, negotiation of forward contracts, and development of cost efficient fuel purchasing strategies. However, daily peak load is a nonlinear, volatile, and nonstationary signal. Besides, lack of sufficient data usually further complicates this problem. The paper proposes a new methodology to solve it, composed of an efficient data model, preforecast mechanism and combination of neural network and evolutionary algorithm as the hybrid forecast technique. The proposed methodology is examined on the EUropean Network on Intelligent TEchnologies (EUNITE) test data and Iran’s power system. We will also compare our strategy with the other MTLF methods revealing its capability to solve this load forecast problem.
Article
Medium range forecasts of power system load for a span of one day to one week are often required for the planning of short term maintenance schedules of unit auxiliaries and peaking stations as well as for maintaining security constraints and minimizing operational costs. A time series model of multiplicative SARIMA (seasonal autoregressive integrated moving average) type suffers from a divergent error level in the multistep ahead forecast and the representation of all the specific features of the different days of the week becomes quite complex. In the present paper a unique approach has been made by grouping the data into various subgroups, characterizing each weekday separately. Each group is replaced by an orthogonal transform known as the Walsh transform, the components of which were then processed by fading memory Kalman filter algorithm and forecasts were made by taking the inverse transform of the predicted value of the component.
Conference Paper
The latest research and results published in the literature show that the interest in load forecasts has moved from traditional techniques to models that use neural networks and fuzzy systems. This paper describes a new approach to the medium-term load forecasting which uses artificial neural networks and fuzzy techniques. The medium-term load forecast must be understood as the assessment of the power consumption as load profiles with hourly values for 48 characteristic days within a year