M. Lydia1* and G. Edwin Prem Kumar2
1Dept. of Mechatronics Engineering, Sri Krishna College of Engineering and Technology, Coimbatore, India
2Dept. of Information Technology, Sri Krishna College of Engineering and Technology, Coimbatore, India
Abstract
India, having the fourth largest installed capacity of wind power, is poised to grow in leaps and bounds in renewable energy utilization. The stochastic nature of wind has been a constant challenge in integration of wind power to the grid. According to the National Institute of Wind Energy (NIWE), the estimated wind potential of India at 120 m above ground level is around 695 GW. In order to effectively tap this power and to enhance wind power penetration in the grid, it is imperative that efficient wind speed and wind power forecasting models are in place. Forecasting of wind power aids in effective grid operations, planning of economic dispatch, estimation of candidate sites for wind farms and in scheduling operation and maintenance of wind farms. Deep learning models for wind forecasting have recently challenged the conventionally used forecasting models in terms of their accuracy, robust nature, and ability to handle huge volumes of data at a much lower computational cost. An exhaustive review of all the deep learning models used for wind speed/power forecasting is reviewed in this chapter. The research challenges faced and future research directions are also presented.
Keywords: Deep learning, forecasting, wind speed, wind power, accuracy
AR | Autoregressive |
ARMA | Autoregressive with moving average |
ARX | Autoregressive with exogenous input |
CEEMDAN | Complete ensemble empirical mode decomposition with adaptive noise |
CRPS | Continuous ranked probability score |
CNN | Convolutional neural network |
DAE | Denoising autoencoder |
DBN | Deep belief network |
DGF | Double Gaussian function |
DL | Deep learning |
DLNN | Deep learning neural networks |
DRNN | Deep recurrent neural networks |
ELM | Extreme learning machine |
EMD | Empirical mode decomposition |
ENN | Elman neural network |
ESN | Echo state network |
GD | Gradient descent |
GRU | Gated recurrent unit |
HBSA | Hybrid backtracking search algorithm |
IRAE | Independent recurrent autoencoder |
kNN | k-nearest neighbor |
LASSO | Least absolute shrinkage selector operator |
LSTM | Long short-term memory |
LUBE | Lower and upper bound estimation |
ML | Machine learning |
MAE | Mean absolute error |
MAPE | Mean absolute percentage error |
MLP | Multi-layer perceptron |
MTL | Multi-task learning |
MSE | Mean squared error |
NN | Neural network |
NWP | Numerical weather prediction |
PSO | Particle swarm optimization |
RBF | Radial basis function |
RF | Random forest |
RMSE | Root mean squared error |
RMSLE | Root mean squared log error |
RNN | Recurrent neural network |
R-NNs | Rough NN |
SAE | Stacked autoencoder |
SCADA | Supervisory control and data acquisition |
SDAE | Stacked denoising autoencoder |
SIRAE | Stacked independently recurrent autoencoder |
sMAPE | Symmetric mean absolute percentage error |
SSA | Single spectrum analysis |
STL | Single task learning |
SVM | Support vector machine |
SVR | Support vector regression |
TL | Transfer learning |
VMD | Variational mode decomposition |
WF | Wind forecasting |
WPF | Wind power forecasting |
WPRE | Wind power ramp event |
WSF | Wind speed forecasting |
WT | Wind turbines |
xGBoost | Extreme gradient boost |
Increasing use of renewable energy has become a mandatory alternative to mitigate the impact of environmental degradation caused by fossil fuels. Non-conventional energy resources like wind energy help in reduction of the carbon footprint globally, emission of green-house gases, and other pollutants. Accurate forecasting models are necessary for ensuring increased penetration of renewable energy in the power grid. Forecasting in offshore scenario is more challenging due to severe and unpredictable weather conditions that the wind turbines (WT) are exposed to.
Wind forecasting is classified into ultra-short-term, very short-term, short-term, and long-term wind forecasting based on the forecasting horizon. Ultra-short-term forecasting models spanning few seconds have been used to build accurate models for onshore and offshore WT models. Very short-term forecasting can span for few minutes and is very helpful in power prediction in intra-day and real-time markets. Short-term forecasting can span for few minutes to hours or days and is helpful in day-ahead markets, reserve setting, unit commitment, and economic dispatch. Long-term forecasts which usually span from days to weeks are useful for scheduling operation and maintenance activities.
The wind speed or power data used to build forecasting models can be 10-minute averaged data or hourly averaged data or daily data, depending on the availability of data for the site under consideration. Recent literature includes the use of high-frequency SCADA data, sampled for every 1 second.
Forecasting models for wind speed or wind power have been essentially built using three different approaches, namely, physical, statistical, and learning models [1]. The physical models include the numerical weather prediction (NWP) models which include the mathematical description of physical processes taking place in the earth’s atmosphere (Figure 6.1). Statistical approach involves in quantitatively defining the relationship between meteorological predictions and historical data. Statistical modeling involves mathematical relationships between the dependent and independent variables. They include autoregressive (AR), auto regressive with moving average (ARMA), auto regressive with exogenous variables (ARX), and autoregressive integrated moving average (ARIMA). Learning approach include soft computing-based methods including artificial intelligence, fuzzy logic, support vector machine, and neural networks.
The methodologies used for wind forecasting can also be classified as model-driven, data-driven, and ensemble methods [2]. Model-based methods are built based on NWP models and meteorological data, whereas data-driven models attempt to learn relationships between the identified input variables. Data-driven models include statistical and artificial intelligence-based approaches like AR, ARMA, ANN, and ELM. A combination of model-based and data-driven models result in the ensemble methods. These methods incorporate signal processing techniques like wavelets and optimization algorithms like PSO, along with other prediction models like ANN and SVM.
Deep learning (DL) algorithms have proved to outperform traditional NNs and other conventional models with improved feature extraction and enhanced ability to learn complex non-linear relationships. According to literature, DL models for wind power forecasting includes hybrid ensemble deep reinforcement learning model, DL neural network, Bayesian DL model, convolutional gated recurrent unit vector, and support vector regression. The accuracy of these forecasting models can be ascertained using suitable performance metrics like root mean squared error, mean absolute error, and mean absolute percentage error.
Models for WPF are primarily based on the relationship of wind power and important features like wind speed at hub height, wind direction, wind shear, and other atmospheric parameters like relative humidity, atmospheric pressure, and temperature. Models for WSF have been realized using time series modeling and various others statistical and ML models. The various WPF models are outlined in Figure 6.2.
The persistence model works on the assumption that wind at a future instant will take the same value as the wind at current instant. These models have proven to work best for ultra-short-term horizons. However, their performance is relatively bad for longer time horizons.
Point forecasting gives a single value for the wind power forecast. Most of the statistical approaches, time series models, and learning approaches like MLP, SVM, RNN, and ELM perform best to provide point forecast of wind speed or power. Due to the intermittent nature of wind, forecast error occurs which hamper the operation cost and supply consistency [3]. However, the uncertainties in wind power can be effectively captured using probabilistic forecast methods. These methods generate pdfs, quantiles, or intervals of future wind speed or power.
Multi-objective optimization generally considers multiple criteria for the process of decision making. Zhou et al. proposed a multi-objective framework for wind power forecast intervals. They examined the relationship between the average width of forecast interval and its approximation error [3].
Morena et al. [4] proposed a DLNN for prediction of wind power ramp events (WPREs). WPREs refer to sudden increase or decrease in wind power that may affect the WT operation. WPREs are spatially related events and their prediction aids in protection of WTs and in penetration of wind power to the grid.
Wind speed interval prediction plays a significant role in WSF and WPF. Wind being a stochastic resource, needs nonlinear temporal parameters for prediction of intervals. Gaussian process, fuzzy inference, and beta distribution function are the conventional methods used for interval forecasting. Zhou et al. developed an LSTM based model for interval forecast. The prediction intervals were computed using the LUBE method. Gan et al. proposed the use of temporal CNN for interval prediction [5]. GD optimization–based DL model was developed by Li et al., short-term interval forecasting of wind power [6].
The dynamic characteristic of wind power generated in a wind farm is very successfully captured by multi-step forecasting models. These models are based on three approaches, namely recursive approach, direct approach, and multi-input and multi-output approach [2]. The DL methods used for multi-step forecasting have been tabulated in Table 6.1.
Table 6.1 Models for multi-step wind forecasting.
Author | DL models |
Chen et al. [7] | Predictive stacked autoencoders |
Liu et al. [8] | Variational mode decomposition, SSA, LSTM, and ELM |
Zhang et al. [9] | Deep Boltzmann machine |
Xiang et al. [10] | Secondary decomposition, phase space reconstruction-bidirectional GRU, and chicken swarm optimization |
Yan et al. [11] | Improved singular spectrum decomposition, LSTM, and grasshopper optimization algorithm–based deep belief network |
DL is a subset of machine learning which, in turn, is a subset of artificial intelligence. DL has the immense ability to learn without human intervention, extract significant features, and can handle unstructured and unlabelled data as well. The various paradigms of DL are depicted in Figure 6.3.
Batch represents a group of training samples used in one iteration. In batch learning, the batch size can either be equal to the total dataset or slightly lesser than the dataset.
Sequential learning is a paradigm of DL, where one part of a task is learnt before the next. Niu et al. proposed a sequence-to-sequence prediction model based on GRU. This kind of learning guaranteed stability in multistep forecasting and ensured reliable power system operation [2]. Zhang et al. developed a sequence-to-sequence model for WPF [12]. They used NWP data for building the model and compared its performance with DBN and RF.
In this kind of learning, the knowledge of the existing model is continuously updated based on the input data. Hence, the learning process happens at the instant when a new input emerges.
Yu et al. proposed the concept of scene learning, by embedding the WTs into the grid space [13]. Spatio-temporal features were extracted and given as input to deep CNN for WPF.
Transfer learning (TL) is about transferring the knowledge gained in one problem to another associated problem. Hu et al. implemented TK for WSF, by transferring knowledge from older farms rich in data to newer farms [14]. WF errors were found to be reduced significantly. Qureshi et al. proposed a WPF method based on DLNN and meta regression TL [15]. They found that TL significantly reduced the computational time of the DL algorithms.
Neural structured learning is about training NN using structured signals along with other inputs. Structure can either be in the form of graph or induced by perturbation. Mi and Zhao proposed a WSF model based on SSA and neural structural learning [16].
This kind of learning is believed to have widespread applicability to real world problems. It involves multiple related problems and exploits the data collected for all the tasks. Hence, it is proven to enhance generalization. A multi-task learning (MTL) for WPRE forecasting has been developed in [4]. MTL for simultaneous forecast of wind energy and demand was developed by Qin et al., using a single NN method [17]. When compared with STL methods, the MTL methods resulted in more authentic results due to the presence of joint layers in its configuration.
A DLNN for forecasting wind power in a 7-MW offshore WT was proposed by Lin and Liu. They used 1-s sampled SCADA data for building the model. They used eleven features including wind speed at four different heights, pitch angles of each blade, mean pitch angle, temperature, yaw error, and orientation of nacelle [18]. A CNN-based wind speed forecasting model was proposed by Hong and Satriani for an offshore WT near Taiwan [19]. The model was used for day-ahead forecasting and was based on historic spatial and temporal data. Hong and Rioflorida proposed a hybrid CNN-RBF NN for day-ahead WSF [1]. Lin et al. developed a DLNN model for WPF in offshore WTs. The hostile effect of outliers was reduced using isolation forest algorithms [20]. A short-term WSF algorithm based on graph convolution DLNN was developed by Khodayar and Wang [21]. The proposed spatio-temporal model used the application of DL, graph theory, and rough set theory. The spatio-temporal features were robust and could tolerate uncertainties and noise and the proposed model outperformed DBNs and SAE networks (Figure 6.4).
Qin et al. developed a fusion model of forecasting, using LSTM and DLNN [17]. Using CNN which exploited the spatial features of the wind field and LSTM that trained on the dynamic features, the prediction of wind signal was effectively made. Liu et al. proposed a hybrid combination of SDAELSTM for forecasting wind speed. This novel combination built on big data, outperformed the conventional MLP network [22]. The suitability of optimal LSTM networks for forecasting wind in long-term was ascertained by Pujari et al. [23]. A short-term WSF model based on LSTM was proposed based on four modules [24]. It included the application of crow search algorithm, wavelet transform, feature selection and LSTM based DL time series prediction. A hybrid WSF methodology was developed by Liu et al. incorporating the functionalities of wavelet transforms, LSTM, and ENN [25]. The efficacy of the model for multi-step prediction was also tested and was found to be excellent.
Time series modeling utilizes time stamped data, to model the future values of the series as a function of present and past values. Yang and Chen developed a fusion model based on the combination of EMD, SAE, and ELM for accurate wind speed forecasting [26]. They also evaluated the impact of shared-hidden-layer style, which was responsible for enriching data-poor sources. Tahmasebifar et al. proposed hybrid models for point and probabilistic forecasting of wind power based on ELM [27]. The other techniques used in the fusion include, mutual information, bootstrap approach, and PSO.
A GRU is similar to LSTM with a forget gate but has lesser parameters than LSTM. Niu et al. developed an attention-based GRU for WPF [2]. The attention mechanism aids in identifying the most significant input variable. Kisvari et al. proposed a GRU based WPF method [28]. They used the isolation forest algorithm for filtering options. The results obtained outperformed the LSTM model. Liu et al. proposed a novel fusion methodology for WPF by combining SSA, CNN, GRU and SVR method [29]. The proposed multi-step prediction model outperformed conventional forecasting models.
Autoencoders belong to a family of NN, whose output is same as the input. They output is reconstructed by compressing the input on to a latent-space representation. Wang et al. developed SIRAE for efficient ultra-short-term WPF [30]. Variable mode decomposition technique was used to decompose the original sequence into sub-sequences, which were then used as input variables. Jahangir et al. proposed a novel WPF algorithm, in which DAE were used for denoising the inputs and R-NNs were used for forecasting [31]. A neuron-pair known as upper and lower bound neurons comprise rough neuron.
The research works carried out in developing ensemble models used for wind forecasting has been tabulated in Table 6.2.
The research works that are carried out using various other DL algorithms have been described in Table 6.3.
Forecasting with increased accuracy with an aim to reduce the operational cost of WT is required. Forecasting algorithms exclusively aimed at off-shore environmental conditions will be a definite boost to the budding off-shore wind energy generation.
Table 6.2 Ensemble models for WPF.
Author | Proposed model | Algorithms used |
H. H. H. Aly [32] | Intelligent clustered hybrid model | Recurrent Kalman filter, Fourier series, wavelet neural network, ANN |
Peng et al. [33] | DL ensemble model | Wavelet soft threshold denoising, GRU |
Wang et al. [34] | DL based ensemble model | Wavelet transform, deep CNN, ensemble technique |
Liu et al. [35] | Time-variant multi-resolution ensemble model | Outlier robust ELM, multiobjective multi-verse optimizer algorithm, clustering autoencoder |
Liu et al. [36] | Deep reinforcement learning model | Wavelet transform, LSTM, DBN, ESN, reinforcement learning |
Chen et al. [37] | Nonlinear-learning ensemble | LSTM, SVR machine, external optimization algorithm |
Jiajun et al. [38] | Ultra-short-term wind prediction | Wavelet transform, DBN, Light gradient boosting machine, random forest |
Liu et al. [39] | Smart DL ensemble model | Wavelet packet decomposition, CNN, convolutional LSTM network |
Development of DL models for prediction interval forecasting for wind farm clusters is identified as an area of future research. Development of top-quality prediction models for WPRE forecasting for wind farm clusters in a particular geographical area is an area yet to be explored. Autotuning of DRNN is an essential area of research. Wind forecasting algorithms with an ultimate aim of aiding effective charging of Plug-in electric vehicle, energy storage, and integration to the power grid should be built, taking into consideration the need and specifications of these applications.
The application of various DL paradigms in WF should be exploited to get more precise forecasting, suitable for various applications. A significant improvement in the efficiency of captured energy from wind will go a long way in transforming wind farms to wind power plants.
Table 6.3 Miscellaneous DL models for WF.
Proposed model | Algorithms used | Metric for forecast accuracy |
Two-stage DL WSF [40] | Wavelet packet decomposition, CNN, bivariate Dirichlet process mixture model | MAE, MAPE, RMSE, coverage width-based criterion |
Time series prediction of wind power generation [41] | Deep feed forward network, deep CNN, RNN, attention mechanism, LSTM | MSE, RMSE, RMSLE |
Forecasting energy consumption and wind power generation [42] | Deep ESN | MAE, RMSE, MAPE, sMAPE |
Interval DGNN for WSF [43] | ELM-HBSA | RMSE, MAPE |
Short-term WSF [44] | CEEMDAN-LSTM, CEEMDAN-error-VMD-LSTM | RMSE, MAPE, MAE |
WPF [45] | LASSO, kNN, xGBoost, SVR, RF | R2, RMSE, MAE |
Cascaded DL WPF [46] | EMD-VMD-CNN-LSTM | RMSE, MAE |
Short-term WSF [47] | Rough deep neural architecture with SAE, SDAE | RMSE, MAE |
Probabilistic spatiotemporal WSF [48] | Convolutional GRU, 3D-CNN, variational Bayesian DL | RMSE, CRPS |
This chapter presents an exhaustive review on DL models for WF. The need for WPF models, the algorithms used to model, their applications, and their classification were described. The various DL paradigms were also presented with their application in WPF and WSF. An overview of research works undertaken in WF, based on DLNN, LSTM, ELM, GRU, SAE, ensemble models, and other assorted models was presented in detail. The research challenges that are yet to be explored in this area have also been included to serve as inputs for future researchers.
1. Hong, Y.Y. and Rioflorido, C.L.P.P., A hybrid deep learning-based neural network for 24-h ahead wind power forecasting. Appl. Energy, 250, 530–539, 2019.
2. Niu, Z., Yu, Z., Tang, W., Wu, Q., Reformat, M., Wind power forecasting using attention-based gated recurrent unit network. Energy, 196, 117081, 2020.
3. Zhou, M., Wang, B., Guo, S., Watada, J., Multi-objective prediction intervals for wind power forecast based on deep neural networks. Inf. Sci., 550, 207–220, 2021.
4. Moreno, M.D., Navarin, N., Gutierrez, P.A., Prieto, L., Sperduti, A., Sanz, S.S., Martinez, C.H., Multi-task learning for the prediction of wind power ramp events with deep neural networks. Neural Networks, 123, 401–411, 2020.
5. Gan, Z., Li, C., Zhou, J., Tang, G., Temporal convolutional networks interval prediction model for wind speed forecasting. Electr. Power Syst. Res., 191, 106865, 2021.
6. Li, C., Tang, G., Xue, X., Chen, X., Wang, R., Zhang, C., The short-term interval prediction of wind power using the deep learning model with gradient descend optimization. Renewable Energy, 155, 197–211, 2020.
7. Chen, J., Zhu, Q., Li, H., Zhu, L., Shi, D., Li, Y., Duan, X., Liu, Y., Learning heterogeneous features jointly: A deep end-to-end framework for multi-step short-term wind power prediction. IEEE Trans. Sustainable Energy, 11, 3, 2020.
8. Liu, H., Mi, X., Li, Y., Smart multi-step deep learning model for wind speed forecasting based on variational mode decomposition, singular spectrum analysis, LSTM network and ELM. Energy Convers. Manage., 159, 54–64, 2018.
9. Zhang, C., Chen, C.L.P., Gan, M., Chen, L., Predictive Deep Boltzmann Machine for multiperiod wind speed forecasting. IEEE Trans. Sustainable Energy, 6, 4, 2015.
10. Xiang, L., Li, J., Hu, A., Zhang, Y., Deterministic and probabilistic multi-step forecasting for short-term wind speed based on secondary decomposition and a deep learning method. Energy Convers. Manage., 220, 113098, 2020.
11. Yan, X., Liu, Y., Xu, Y., Jia, M., Multistep forecasting for diurnal wind speed based on hybrid deep learning model with improved singular spectrum decomposition. Energy Convers. Manage., 225, 113456, 2020.
12. Zhang, Y., Li, Y., Zhang, G., Short-term wind power forecasting approach based on Seq2Seq model using NWP data. Energy, 213, 118371, 2020.
13. Yu, R., Liu, Z., Li, X., Lu, W., Ma, D., Yu, M., Wang, J., Li, B., Scene learning: Deep convolutional networks for wind power prediction by embedding turbines into grid space. Appl. Energy, 238, 249–257, 2019.
14. Hu, Q., Zhang, R., Zhou, Y., Transfer learning for short-term wind speed prediction with deep neural networks. Renewable Energy, 85, 83–95, 2016.
15. Qureshi, A.S., Khan, A., Zameer, A., Usman, A., Wind power prediction using deep neural network based meta regression and transfer learning. Appl. Soft Comput., 58, 742–755, 2017.
16. Mi, X. and Zhao, S., Wind speed prediction based on singular spectrum analysis and neural network structural learning. Energy Convers. Manage., 216, 112956, 2020.
17. Qin, Y., Li, K., Liang, Z., Lee, B., Zhang, F., Gu, Y., Zhang, L., Wu, F., Rodriguez, D., Hybrid forecasting model based on long short-term memory network and deep learning neural network for wind signal. Appl. Energy, 236, 262–272, 2019.
18. Lin, Z. and Liu, X., Wind power forecasting of an offshore wind turbine based on high frequency SCADA data and deep learning neural network. Energy, 201, 117693, 2020.
19. Hong, Y.Y. and Satriani, T.R.A., Day-ahead spatiotemporal wind speed forecasting using robust design-based deep learning neural network. Energy, 209, 118441, 2020.
20. Lin, Z., Liu, X., Collu, M., Wind power prediction based on high-frequency SCADA data along with isolation forest and deep learning neural networks. Electr. Power Energy Syst., 118, 105835, 2020.
21. Khodayar, M. and Wang, J., Spatio-temporal Graph Deep Neural Network for Short-term Wind Speed Forecasting. IEEE Trans. Sustainable Energy, 10, 2, 2019.
22. Liu, X., Zhang, H., Kong, X., Lee, K.Y., Wind speed forecasting using deep neural network with feature selection. Neurocomputing, 397, 393–403, 2020.
23. Pujari, K.N., Miriyala, S.S., Mittal, P., Mitra, K., Optimal Long Short-Term Memory Networks for long-term forecasting for real wind characteristics. IFAC PapersOnLine, 53-1, 648–653, 2020.
24. Memarzadeh, G. and Keynia, F., A new short-term wind speed forecasting method based on fine-tuned LSTM neural network and optimal input sets. Energy Convers. Manage., 213, 112824, 2020.
25. Liu, H., Mi, X., Li, Y., Wind speed forecasting method based on deep learning strategy using empirical wavelet transform, long short-term memory neural network and Elman neural network. Energy Convers. Manage., 156, 498–514, 2018.
26. Yang, H.F. and Chen, Y.P., Representation learning with extreme learning machines and empirical mode decomposition for wind speed forecasting methods. Artif. Intell., 277, 103176, 2019.
27. Tahmasebifar, R., Moghaddam, M.P., Sheikh-El-Eslami, M.K., Kheirollahi, R., A new hybrid model for point and probabilistic forecasting of wind power. Energy, 211, 119016, 2020.
28. Kisvari, A., Lin, Z., Liu, X., Wind power forecasting – A data-driven method along with gated recurrent neural network. Renewable Energy, 163, 1895–1909, 2021.
29. Liu, H., Mi, X., Li, Y., Duan, Z., Xu, Y., Smart wind speed deep learning based multi-step forecasting model using singular spectrum analysis, convolutional Gated Recurrent Unit network and Support Vector Regression. Renewable Energy, 143, 842–854, 2019.
30. Wang, L., Tao, R., Hu, H., Zeng, Y., Effective wind power prediction using novel deep learning network: Stacked independently recurrent autoencoder. Renewable Energy, 164, 642–655, 2021.
31. Jahangir, H., Golkar, M.A., Alhameli, F., Mazouz, A., Ahmadian, A., Elkamel, A., Short-term wind speed forecasting framework based on stacked denoising auto-encoders with rough ANN. Sustain. Energy Technol. Assess., 38, 100601, 2020.
32. Aly, H.H.H., A novel deep learning intelligent clustered hybrid models for wind speed and power forecasting. Energy, 213, 118773, 2020.
33. Peng, Z., Peng, S., Fu, L., Lu, B., Tang, J., Wang, K., Li, W., A novel deep learning ensemble model with data denoising for short-term wind speed forecasting. Energy Convers. Manage., 207, 112524, 2020.
34. Wang, H., Li, G., Wang, G., Peng, J., Jiang, H., Liu, Y., Deep learning-based ensemble approach for probabilistic wind power forecasting. Appl. Energy, 188, 56–70, 2017.
35. Liu, H., Duan, Z., Chen, C., Wind speed big data forecasting using time-variant multi-resolution ensemble model with clustering auto-encoder. Appl. Energy, 280, 115975, 2020.
36. Liu, H., Yu, C., Wu, H., Duan, Z., Yan, G., A new hybrid ensemble deep reinforcement learning model for wind speed short term forecasting. Energy, 202, 117794, 2020.
37. Chen, J., Zeng, G., Zhou, W., Du, W., Lu, K., Wind speed forecasting using nonlinear-learning ensemble of deep learning time series prediction and extremal optimization. Energy Convers. Manage., 165, 681–695, 2018.
38. Jiajun, H., Chuanjin, Y., Yongle, L., Huoyue, X., Ultra-short-term wind prediction with wavelet transform, deep belief network and ensemble learning. Energy Convers. Manage., 205, 112418, 2020.
39. Liu, H., Mi, X., Li, Y., Smart deep learning-based wind speed prediction model using wavelet packet decomposition, convolutional neural network and convolutional long short-term memory network. Energy Convers. Manage., 166, 120–131, 2018.
40. Liu, H., Duan, Z., Chen, C., Wu, H., A novel two-stage deep learning wind speed forecasting method with adaptive multiple error corrections and bivariate Dirichlet process mixture model. Energy Convers. Manage., 199, 11975, 2019.
41. Mishra, S., Bordin, C., Taharaguchi, K., Palu, I., Comparison of deep learning models for multivariate prediction of time series wind power generation and temperature. Energy Rep., 6, 273–286, 2020.
42. Hu, H., Wang, L., Lv, S., Forecasting energy consumption and wind power generation using deep echo state network. Renewable Energy, 154, 598–613, 2020.
43. Khodayar, M., Wang, J., Manthouri, M., Interval deep generative neural network for wind speed forecasting. IEEE Trans. Smart Grid, 10, 4, 2019.
44. Ma, Z., Chen, H., Wang, J., Yang, X., Yan, R., Jia, J., Xu, W., Application of hybrid model based on double decomposition, error correction and deep learning in short-term wind speed prediction. Energy Convers. Manage., 205, 112345, 2020.
45. Demolli, H., Dokuz, A.S., Ecemis, A., Gokcek, M., Wind power forecasting based on daily wind speed data using machine learning algorithms. Energy Convers. Manage., 198, 111823, 2020.
46. Yin, H., Ou, Z., Huang, S., Meng, A., A cascaded deep learning wind power prediction approach based on a two-layer of mode decomposition. Energy, 189, 116316, 2019.
47. Khodayar, M., Kaynak, O., Khodayar, M.E., Rough deep neural architecture for short-term wind speed forecasting. IEEE Trans. Industr. Inform., 13, 6, 2017.
48. Liu, Y., Qin, H., Zhang, Z., Pei, S., Jiang, Z., Feng, Z., Zhou, J., Probabilistic spatiotemporal wind speed forecasting based on a variational Bayesian deep learning model. Appl. Energy, 260, 114259, 2020.
18.216.168.162