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

Abstract and Figures

Achieving the desired accuracy in time series forecasting has become a binding domain, and developing a forecasting framework with a high degree of accuracy is one of the most challenging tasks in this area. Combining different forecasting methods to construct efficient hybrid models has been widely reported in the literature regarding this challenge. Various types of hybrid models have been developed and successfully employed to improve forecasting accuracy. The well-known hybrid models can be generally categorized into four classes: (1) preprocessing-based, (2) parameter optimization-based, (3) components combination-based, and (4) postprocessing-based hybrid models. Despite the significant successes of hybrid models, efforts to access more accurate results face continued growth. Hybridization of hybrid models is a novel idea proposed to obtain extreme accuracy in recent literature, in which two or more hybrid classes are combined instead of conjoining the conventional individual forecasting methods. Although, in many publications, the aforementioned classes of hybrid models have been reviewed and analyzed in a wide variety of forecasting fields; no study is conducted to review the hybridization of hybrid models. This paper’s main contribution is to fill this gap and provide classification and comprehensive review of the current endeavors done in the hybridization of hybrid models in time series forecasting areas. Our searches indicate that more than 250 papers have been published in recent years utilizing hybridization of hybrid models. In this paper, these published papers have been classified regarding their different used combination strategies into four main categories, including (1) Hybridization with preprocessing-based hybrid models (HPH), (2) Hybridization with parameter optimization-based hybrid models (HOH), (3) Hybridization with components combination-based hybrid models (HCH) and, (4) Hybridization with postprocessing-based hybrid models (HSH). Each hybridization of the hybrid class is evaluated regarding the usage frequency, specific merits, and limitations. It can be inferred from reviewing articles that the hybridization of the hybrid concept, as a recent advancement in time series forecasting, can significantly improve traditional hybrid models’ accuracy. Furthermore, each category’s research gaps and some future research directions are identified in this paper.
This content is subject to copyright. Terms and conditions apply.
Vol.:(0123456789)
Artificial Intelligence Review
https://doi.org/10.1007/s10462-022-10199-0
1 3
Hybridization ofhybrid structures fortime series forecasting:
areview
ZahraHajirahimi1· MehdiKhashei1
Accepted: 26 April 2022
© The Author(s), under exclusive licence to Springer Nature B.V. 2022
Abstract
Achieving the desired accuracy in time series forecasting has become a binding domain,
and developing a forecasting framework with a high degree of accuracy is one of the most
challenging tasks in this area. Combining different forecasting methods to construct effi-
cient hybrid models has been widely reported in the literature regarding this challenge.
Various types of hybrid models have been developed and successfully employed to
improve forecasting accuracy. The well-known hybrid models can be generally categorized
into four classes: (1) preprocessing-based, (2) parameter optimization-based, (3) compo-
nents combination-based, and (4) postprocessing-based hybrid models. Despite the sig-
nificant successes of hybrid models, efforts to access more accurate results face contin-
ued growth. Hybridization of hybrid models is a novel idea proposed to obtain extreme
accuracy in recent literature, in which two or more hybrid classes are combined instead of
conjoining the conventional individual forecasting methods. Although, in many publica-
tions, the aforementioned classes of hybrid models have been reviewed and analyzed in
a wide variety of forecasting fields; no study is conducted to review the hybridization of
hybrid models. This paper’s main contribution is to fill this gap and provide classification
and comprehensive review of the current endeavors done in the hybridization of hybrid
models in time series forecasting areas. Our searches indicate that more than 250 papers
have been published in recent years utilizing hybridization of hybrid models. In this paper,
these published papers have been classified regarding their different used combination
strategies into four main categories, including (1) Hybridization with preprocessing-based
hybrid models (HPH), (2) Hybridization with parameter optimization-based hybrid models
(HOH), (3) Hybridization with components combination-based hybrid models (HCH) and,
(4) Hybridization with postprocessing-based hybrid models (HSH). Each hybridization of
the hybrid class is evaluated regarding the usage frequency, specific merits, and limitations.
It can be inferred from reviewing articles that the hybridization of the hybrid concept, as a
recent advancement in time series forecasting, can significantly improve traditional hybrid
models’ accuracy. Furthermore, each category’s research gaps and some future research
directions are identified in this paper.
* Mehdi Khashei
Khashei@cc.iut.ac.ir
Extended author information available on the last page of the article
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
... A hybrid model combines the advantages of each distinct model, leading to high prediction performance for time series with shorter time scales and longer lead times (Fung et al., 2019). Hybridisation has emerged as a promising technique for overcoming numerous drawbacks of standalone methods while also improving prediction accuracy (Hajirahimi & Khashei, 2022). There are several types of hybrid models, e.g., the hybrid model that combines a physical model and a machine learning model (Nualtong et al., 2021;Xu et al., 2019), and the hybrid stochastic-ML model proposed by Koutsoyiannis and Montanari (Koutsoyiannis & Montanari, 2022). ...
... These hybrid models are typically made up of various procedures, with one serving as the principal technique and the others as pre-or post-processing methods (Zubaidi, Ortega-Martorell, Al-Bugharbee et al., 2020). According to the reviewed papers, hybrid models can be classified into four groups: the components combination-based hybrid models (CBH), pre-processing-based hybrid models (PBH), parameter optimisation-based hybrid models (OBH), and hybridisation of parameter optimisation-based with preprocessing-based hybrid models (HOPH) as in Hajirahimi and Khashei (Hajirahimi & Khashei, 2022) (see Figure 3) and (Table 2). ...
... While the primary aim of the filter and denoising-based approaches is to detect and remove existing noise in the underlying time series. In the second step, the screened time series is forecasted by the appropriate individual model (Hajirahimi & Khashei, 2022). ...
Article
Full-text available
The community’s well-being and economic livelihoods are heavily influenced by the water level of watersheds. The changes in water levels directly affect the circulation processes of lakes and rivers that control water mixing and bottom sediment resuspension, further affecting water quality and aquatic ecosystems. Thus, these considerations have made the water level monitoring process essential to save the environment. Machine learning hybrid models are emerging robust tools that are successfully applied for water level monitoring. Various models have been developed, and selecting the optimal model would be a lengthy procedure. A timely, detailed, and instructive overview of the models’ concepts and historical uses would be beneficial in preventing researchers from overlooking models’ potential selection and saving significant time on the problem. Thus, recent research on water level prediction using hybrid machines is reviewed in this article to present the “state of the art” on the subject and provide some suggestions on research methodologies and models. This comprehensive study classifies hybrid models into four types algorithm parameter optimisation-based hybrid models (OBH), pre-processing-based hybrid models (PBH), the components combination-based hybrid models (CBH), and hybridisation of parameter optimisation-based with preprocessing-based hybrid models (HOPH); furthermore, it explains the pre-processing of data in detail. Finally, the most popular optimisation methods and future perspectives and conclusions have been discussed.
... In recent years, combined models have arisen as a way to construct flexible and efficient models and improve the forecasting accuracy of individual algorithms [5,55]. The hybrid models can be classified into four types, namely: the components combination-based hybrid models (CBH), parameter optimisation-based hybrid models (OBH), pre-processing-based hybrid models (PBH), and hybridisation of hybrid models as in Hajirahimi and Khashei [22]. There are different studies in the hybrid models shown in Figure 3. ...
... In this section, ML models were combined to correct the relative incompetency of the individual models. The CBH models aim to improve prediction performance by enabling the remarkable capacity of individual prediction models regardless of combination structures [22]. For example, Lola et al. [56] developed a combined technique to forecast daily WQ data (DO, water T, pH, and salinity) using ARIMA and ANN. ...
... The hybridisation of hybrid models is a novel idea proposed to improve forecasting precision over traditional hybrid classes [22]. ...
Article
Full-text available
Water quality has a significant influence on human health. As a result, water quality parameter modelling is one of the most challenging problems in the water sector. Therefore, the major factor in choosing an appropriate prediction model is accuracy. This research aims to analyse hybrid techniques and pre-processing data methods in freshwater quality modelling and forecasting. Hybrid approaches have generally been seen as a potential way of improving the accuracy of water quality modelling and forecasting compared with individual models. Consequently, recent studies have focused on using hybrid models to enhance forecasting accuracy. The modelling of dissolved oxygen is receiving more attention. From a review of relevant articles, it is clear that hybrid techniques are viable and precise methods for water quality prediction. Additionally, this paper presents future research directions to help researchers predict freshwater quality variables.
... Different types of hybrid models have been created and successfully used to increase prediction accuracy [33,34]. Therefore, the hybrid models are classified into three categories: ...
... Few studies employed this type of hybrid model, such as Khan et al. [31] and Wu et al. [40]. Different types of hybrid models have been created and successfully used to increase prediction accuracy [33,34]. Therefore, the hybrid models are classified into three categories: ...
... (2) The parameter optimisation-based hybrid models (OBH): The primary idea behind the OBH models is to use optimisation algorithms to identify the optimal parameters of the models [34], and combine the models with nature-inspired algorithms for hydrological drought prediction such as Adnan, et al. [32], Nabipour et al. [20], Banadkooki et al. [38], and Kisi et al. [39]. ...
Article
Full-text available
Drought is a prolonged period of low precipitation that negatively impacts agriculture, animals, and people. Over the last decades, gradual changes in drought indices have been observed. Therefore, understanding and forecasting drought is essential to avoid its economic impacts and appropriate water resource planning and management. This paper presents a recent literature review, including a brief description of data pre-processing, data-driven modelling strategies (i.e., univariate or multivariate), machine learning algorithms (i.e., advantages and disadvantages), hybrid models, and performance metrics. Combining various prediction methods to create efficient hybrid models has become the most popular use in recent years. Accordingly, hybrid models have been increasingly used for predicting drought. As such, these models will be extensively reviewed, including preprocessing-based hybrid models, parameter optimisation-based hybrid models, and hybridisation of components combination-based with preprocessing-based hybrid models. In addition, using statistical criteria, such as RMSE, MAE, NSE, MPE, SI, BIC, AIC, and AAD, is essential to evaluate the performance of the models.
... Recently, Hajirahimi and Khashei [42] reviewed the hybridization of hybrid structures for time series predicting, and the study demonstrated that pre-processing data and optimization algorithms are crucial components of hybridization. The hybridization of hybrid models, wherein two or more hybrid classes are merged as opposed to combining the typical individual forecasting methods, is a novel idea suggested in recent literature in order to reach a high accuracy. ...
Article
Full-text available
Accurate streamflow prediction is significant when developing water resource management and planning, forecasting floods, and mitigating flood damage. This research developed a novel methodology that involves data pre-processing and an artificial neural network (ANN) optimised with the coefficient-based particle swarm optimisation and chaotic gravitational search algorithm (CPSOCGSA-ANN) to forecast the monthly water streamflow. The monthly streamflow data of the Tigris River at Amarah City, Iraq, from 2010 to 2020, were used to build and evaluate the suggested methodology. The performance of CPSOCGSA was compared with the slim mold algorithm (SMA) and marine predator algorithm (MPA). The principal findings of this research are that data preprocessing effectively improves the data quality and determines the optimum predictor scenario. The hybrid CPSOCGSA-ANN outperformed both the SMA-ANN and MPA-ANN algorithms. The suggested methodology offered accurate results with a coefficient of determination of 0.91, and 100% of the data were scattered between the agreement limits of the Bland–Altman diagram. The research results represent a further step toward developing hybrid models in hydrology applications.
Article
Full-text available
Salinity is a classic problem in water quality management since it is directly associated with low water quality indices. Debate continues about selecting the best model for water quality forecasting, it remains a major challenge and causes much uncertainty. Accordingly, identifying the optimal modelling that can capture the salinity behavior is becoming a common trend in recent water quality research. This study applies novel combined techniques, including data preprocessing and artificial neural network (ANN) optimized with constriction coefficient-based particle swarm optimization and chaotic gravitational search algorithm (CPSOCGSA) to forecast monthly salinity data. Historical monthly total dissolved solids (TDS) and electrical conductivity (EC) data of the Euphrates River at AlMusayyab, Babylon, and climatic factors from 2010 to 2019 were used to build and validate the methodology. Additionally, for more validation, the CPSOCGSA-ANN was compared with the slime mould algorithm (SMA-ANN), particle swarm optimization (PSO-ANN) and multi-verse optimizer (MVO-ANN). The results reveal that the preprocessing data approaches improved data quality and selected the best predictors’ scenario. The CPSOCGSA-ANN algorithm is the best based on several statistical criteria. The proposed methodology accurately simulated the TDS and EC time series based on R2 = 0.99 and 0.97, respectively, and SI = 0.003 for both parameters.
Article
Full-text available
Water level (WL) forecasting has become a difficult undertaking due to spatiotemporal fluctuations in climatic factors and complex physical processes. This paper proposes a novel hybrid machine learning model based on an artificial neural network (ANN) and the Marine Predators algorithm (MPA) for modeling monthly water levels of the Tigris River in Al-Kut, Iraq. Data preprocessing techniques are employed to enhance data quality and determine the optimal input model. Historical data for water level and climatic factors data are utilized from 2011 to 2020 to build and assess the model. MPA-ANN algorithm’s performance is compared with recent constriction coefficient-based particle swarm optimization and chaotic gravitational search algorithm (CPSOCGSA-ANN) and slime mold algorithm (SMA-ANN) to reduce uncertainty and raise the prediction range. The finding demonstrated that singular spectrum analysis is a highly effective method to denoise time series. MPA-ANN outperformed CPSOCGSA-ANN and SMA-ANN algorithms based on different statistical criteria. The suggested novel methodology offers good results with scatter index (SI) = 0.0009 and coefficient of determination (R2 = 0.98).
Article
Wind energy is an emerging source of renewable energy in Brazil. Nevertheless, it already accounts for 17% of the National Interconnected Network. Due to the great intricacy of wind speed variations, it is difficult to predict wind energy with high accuracy. This research offers, in these circumstances, an ensemble learning model based on variational mode decomposition and singular spectrum analysis in decomposition in multiple stages, using stacking-ensemble learning. The proposed model is tested and applied in short-term wind speed data from a wind farm located in Parazinho in the northeast region of Brazil, using a multi-step-ahead forecasting strategy. The selected models for forecasting were the machine learning models partial least squares regression, k-nearest neighbors, cubist regression, support vector regression, and ridge regression. The results of the study were divided into three comparative experiments: comparisons with (i) dual decomposed models, (ii) single decomposed models, and (iii) decomposed models. Concerning performance improvement, in the first experiment, the model was compared to dual decomposition models with an average performance between 3.71% and 21.38%. In the second experiment, mean performance improved between 37.18 and 52.47 percent compared to single decomposition. Lastly, the proposed model delivered, on average, 54.98% better results than the models without decomposition. In summary, all compared models in all forecasting horizons were surpassed by the proposed model, with an average improvement between 3.69 and 56.61 percent, showing that the dual decomposition ensemble learning model is an effective and accurate approach for forecasting wind speed.
Article
The fluctuation of agricultural commodity prices has attracted a considerable amount of attention. However, the complexity of the agricultural futures market and the variability of influencing factors makes the prediction of agricultural commodity futures prices difficult. We address the nonlinear characteristics of agricultural commodity futures price series by proposing a forecast combination approach based on a global optimization method, called the Artificial Bee Colony Algorithm (ABC), for forecasting soybean and corn futures prices. Firstly, we used three denoising techniques, namely singular spectral analysis (SSA), empirical mode decomposition (EMD), and variational mode decomposition (VMD), to filter the external noise in the original price series. Then, we generated diverse forecasting sub-models by combining denoising techniques and five popular forecasting models: autoregressive integrated moving average regression (ARIMA), support vector regression (SVR), recurrent neural network (RNN), gated recurrent neural network (GRU), and long-short term memory neural network (LSTM). Finally, we present an ABC approach for three forecast combinations: heterogeneous, semi-heterogeneous, and homogeneous combination. Experimental results indicate that the semi-heterogeneous forecast combination is superior to other combination strategies. For corn and soybean prices, ABC-based semi-heterogeneous forecast combinations have error reductions of 53.3% and 50.0% of MAPE and improvements of 32.4% and 34.5% in Dstat compared to the best single models, respectively.
Article
The growing human population in this modern society hugely depends on the energy to fulfill their day-to-day needs and activities. Renewable energy sources, especially solar energy, can satisfy the global power demand while reducing global warming caused by conventional sources. Solar irradiance is an essential component in solar power applications. The availability of solar irradiance is influenced by several factors, such as forecasting horizon, weather classification, and performance evaluation metrics, which also need consideration. The accurate forecasting of solar irradiance is of utmost importance for the power system designers and grid operators for efficient management of solar energy systems. The intermittent and non-stationary nature of solar irradiance makes many existing statistical and machine learning approaches less competent in providing accurate predictions. In this context, deep learning models have been proposed by several researchers to reduce the limitations of existing machine learning models and improve prediction accuracy. In this work, an extensive and comprehensive review of deep learning-based solar irradiance forecasting models is presented. The effectiveness and efficacy of several deep learning models, including long short-term memory, deep belief network, echo state network, convolution neural network, etc. have been reviewed. The results obtained in the reported studies proved the superiority of deep learning models in solar forecasting applications. Few researchers have proposed deep hybrid models to improve the prediction performance further. A study reported that the hybrid of CNN-LSTM can enhance the prediction accuracy by 3.62%, 25.29%, 34.66%, 37.37% and 26.20% over CNN, LSTM, GRU, RNN and DNN, respectively. Overall, this paper offers preliminary guidelines for a detailed view of deep learning techniques that researchers and engineers can use to improve the solar photovoltaic plant’s modeling and planning.
Article
Due to the strong randomness of wind speed, wind power generation is difficult to integrate into the grid. It is very important to predict wind speed reliably and accurately so that wind energy can be utilized effectively. In this study, to obtain accurate wind speed prediction results, a combined VMD-D-ESN model based on variational mode decomposition (VMD), a double-layer staged training echo state network (D-ESN) and genetic algorithm (GA) optimization is proposed. First, the VMD-D-ESN model preprocesses the original wind speed data with VMD and then uses the D-ESN model to predict each decomposed subsequence. Lastly, the final prediction value is obtained by combining all of the predicted subsequences. In the D-ESN model’s double-layer structure, the first layer selects the length of the training set, and the second layer has the ability to correct the prediction error in the first layer. In a practical application of case prediction using six different data collection sites, ten models are established to compare the performance of the proposed model. Compared with other traditional models, the results show that the model that combines the VMD decomposition algorithm with the D-ESN structure achieves high prediction accuracy and strong stability on all available datasets. Additionally, the model also shows that the use of the VMD decomposition algorithm strongly improves the prediction ability of the model.
Article
Accurate wind speed forecasting and effective wind energy conversion can reduce the operating cost of wind farms. However, many previous studies have been restricted to analyses of wind speed forecasting and wind energy conversion, which may result in poor decisions and inaccurate power scheduling for wind farms. This study develops a wind energy decision system based on forecasting and simulation, which includes two modules: wind speed forecasting and wind energy conversion. In the wind speed forecasting module, an effective secondary denoising strategy based on singular spectrum analysis and ensemble empirical mode decomposition was used to eliminate chaotic noise and extract important features from the original data. Then, a model selection called weighted information criterion was applied to select optimal sub-models for the combined model. To improve the forecasting performance of the combined model, a modified multi-objective grey wolf optimisation algorithm was adopted to optimise the parameters of the sub-models and the weight of the combined model. In the wind energy conversion module, a wind energy conversion curve was established by simulating historical electrical energy data and wind speed data, which can effectively analyse the power generation at each site. The numerical results show that compared with the mean absolute percentage error values of the single models, that of the combined model is reduced by up to 35.57%. Moreover, the standard deviation of the absolute percentage error is decreased by up to 49.88% for wind speed forecasting, and the R² of the wind energy conversion curve is more than 0.9. Therefore, the proposed combined method can serve as an effective tool for wind farm management and decision-making.
Article
Short-term wind speed forecast is of great importance to wind farm regulation and its early warning. Previous studies mainly focused on the prediction at a single location but few extended the task to 2-D wind plane. In this study, a novel deep learning model was proposed for a 2-D regional wind speed forecast, using the combination of the auto-encoder of convolutional neural network (CNN) and the long short-term memory unit (LSTM). The 12-hidden-layer deep CNN was adopted to encode the high dimensional 2-D input into the embedding vector and inversely, to decode such latent representation after it was predicted by the LSTM module based on historical data. The model performance was compared with parallel models under different criteria, including MAE, RMSE and R², all showing stable and considerable enhancements. For instance, the overall MAE value dropped to 0.35 m/s for the current model, which is 32.7%, 28.8% and 18.9% away from the prediction results using the persistence, basic ANN and LSTM model. Moreover, comprehensive discussions were provided from both temporal and spatial views of analysis, revealing that the current model can not only offer an accurate wind speed forecast along timeline (R² equals to 0.981), but also give a distinct estimation of the spatial wind speed distribution in 2-D wind farm.
Article
Wind power generation is uncontrollable and non-adjustable green energy. The accurate prediction of wind speed is of great significance to the operation and maintenance of wind farms and the safety of the power grid. A novel wind speed prediction approach based on mode decomposition combined with integrated multi-model fusion is proposed. The purpose of the integrated multi-model fusion method is to improve the accuracy of wind speed prediction. Two decomposition algorithms, empirical mode decomposition and local mean decomposition are used to decompose the wind speed into several components respectively. After decomposition, the odd and even sequences are extracted to form two new sequences. According to the characteristics of the components obtained from the decomposition of odd and even sequences, stochastic configuration network and support vector machine are selected as the prediction models. Meanwhile, the particle swarm optimization algorithm is introduced to optimize the parameters of support vector machine. After all decomposition components of the odd and even sequence are predicted by stochastic configuration networks and support vector machine, the predicted values are superimposed to obtain the final predicted value of the odd and even sequence. Finally, the predicted values of odd and even series are integrated to get the final predicted value of wind speed. By comparing the actual and predicted values of short-term wind speed, RMSE, R², MAE, MAPE, Pearson’s test, prediction error distribution box-plot, and Taylor diagram are used to judge the performance indicators of the prediction model. The case study results show that, compared with the state-of-the-art prediction models, RMSE decreases by about 0.1, MAPE decreases by about 0.05%, MAE decreases by about 0.2, and R² increases by about 0.2. The other performance indicators of the proposed approach are also better than other models, which prove that the integrated multi-model fusion model can improve the accuracy and effectiveness of wind speed prediction. The proposed prediction approach has high prediction accuracy, can correctly reflect the short-term wind speed law, and has a good application prospect.
Article
To improve the prediction accuracy of futures price, this paper proposes a hybrid approach based on gradient boosting decision tree (GBDT), correlation analysis, and empirical wavelet transform (EWT) and applies it to predict the settlement price of the London Metal Exchange (LME) Nickel. Firstly, the Spearman correlation coefficient is adopted to analyze the relationship between each input variable and the settlement price. Then, the EWT likewise used to decompose each variable time series into an independent component. And those components are paralleled as new input variables. Next, the GBDT is carried for forecasting the settlement price of LME Nickel. To validate the effectiveness of the EWT, comparison experiments between EMD (empirical mode decomposition) and EWT are carried out. The decomposition results show that the EWT can correctly decompose the settlement price and avoid redundant components. To better measure the viability and efficiency of the proposed method, experiments are on three financial data sets from Tushare before predicting Nickel price. The various implementations are illustrated that the EWT-GBDT can achieve the best results no matter on which data set. Therefore, we have reason to believe that the EWT-GBDT is a useful algorithm to predict LME Nickel settlement prices.
Article
Spatio-temporal wind speed prediction is of great significance to the grid-connected operation of multiple wind farms in smart grid. This paper proposes a spatio-temporal wind speed prediction method based on capsule network (CapsNet) for geographically dispersed wind farms over a region. In the proposed method, the historical wind speed data from the wind farms are originally converted into chronological images in a 3D space, and the spatial features implicit in the images are extracted by the convolutional operation. Then, the temporal information of wind speed spatial properties is encapsulated in multi-dimensional time-capsules and learned by the dynamic routing mechanism, thus capturing the nonlinear temporal dependencies based on the extracted spatial features. A regression layer activated by the leaky rectified linear unit (Leaky ReLU) function integrates the spatio-temporal features and generates the final prediction results. Furthermore, a two-layer iterative training approach is employed to well-tune the model parameters and accelerate the convergence speed. Finally, the real data of multiple wind farms from Ohio are collected in the case studies to demonstrate the superior performance of the proposed method compared with other forecasting methods.
Article
Wind speed forecasting is essential for the dispatch, controllability, and stability of power grids, and its accuracy is vital to the effective use of wind resources. In this study, a novel hybrid wind speed forecasting system is developed based on the data area division (DAD) method and a deep learning neural network model. The system consists of three modules: extraction module, data preprocessing module, and forecasting module. In the data extraction module, a large amount of valid historical data is extracted, filtered, and classified from the forecast location and the surrounding locations. In the data preprocessing module, complementary ensemble empirical mode decomposition is used to decompose the wind speed data. In the forecasting module, a long short-term memory network optimized by using a genetic algorithm is used to forecast the decomposed wind speed data and integrate them into the final forecast results. Numerical simulation results show that (a) the forecast system maintains RMSE in the range of 0.2–0.6 m/s and MAPE in the range of 3.0–7.0% for short-term wind speed forecasts at different locations for different time periods, showing good stability. (b) For wind speed forecasting at different time intervals, the accuracy of wind speed forecasting at 10-minute and 30-minute intervals is better, while the accuracy of forecasting at a 60-minute interval needs to be improved, but overall, the forecasting system shows good generalizability. (c) The forecast system improves the forecast accuracy of short-term wind speed forecasting more effectively than other conventional methods, and the improvement of RMSE and MAPE remains in the range of 14–39% and 13–27% even compared with the hybrid forecast model that has better forecast accuracy. (d) For area-wide short-term wind power forecasting, the forecast deviation value of this forecasting system remains below 6% throughout the year, showing good practicality.
Article
Accurate and reliable wind speed forecasting is important for the dispatch and management of wind power generation systems. However, existing forecasting models based on the data decomposition approach only perform time–frequency analysis of wind speed series, while ignoring the coupling relationship between other meteorological variables and wind speed in the time and frequency domains. Therefore, a novel hybrid wind speed forecasting model based on multivariate data secondary decomposition approach and deep learning algorithm with an attention mechanism is proposed in this study. Specifically, singular spectrum analysis is used to decrease the noise of the original multivariate series. Multivariate empirical mode decomposition is used to decompose the denoised series into their respective intrinsic mode functions and residuals. Further, a new hybrid deep learning algorithm, which combines a convolutional neural network optimized via an attention mechanism and a bidirectional long short-term memory network, is proposed to extract spatiotemporal correlation features between all intrinsic mode functions and residuals and to perform final wind speed forecasting. Finally, three experiments were performed to evaluate the performance of the proposed model comprehensively. The experimental results indicate that the proposed model is superior to other baseline models in terms of both accuracy and effectiveness.