Multi-step wind speed forecasting using EWT decomposition, LSTM principal computing, RELM subordinate computing and IEWT reconstruction. (1st July 2018)

• Record Type: Journal Article
• Authors: Li, Yanfei; Wu, Haiping; Liu, Hui
• Is Part Of: Energy conversion and management. Volume 167(2018)
• Page Start: 203
• Page End: 219
• Publication Date: 2018-07-01
• Subjects: Wind speed forecasting -- Hybrid model -- Empirical wavelet transform -- Long short term memory network -- Regularized extreme learning machine -- Inverse empirical wavelet transform; Direct energy conversion -- Periodicals; Energy storage -- Periodicals; Energy transfer -- Periodicals; Énergie -- Conversion directe -- Périodiques; Direct energy conversion; Periodicals; 621.3105
• DOI: 10.1016/j.enconman.2018.04.082
• Languages: English
• ISSNs: 0196-8904

Abstract

Highlights: A new deep learning based method is proposed for the wind speed forecasting. The EWT is adopted to decompose the original wind speed data. The LSTM network is executed as the principal predictor for the decomposed data. The RELM network is utilized as the subordinate predictor for the error data. The IEWT is employed to reconstruct the all final forecasting results. Abstract: The wind speed forecasting is crucial for the wind power conversion and management. In this study, a new hybrid model consisting of the EWT ( Empirical Wavelet Transform ) decomposition, the LSTM ( Long Short Term Memory ) network, the RELM ( Regularized Extreme Learning Machine ) network and the IEWT ( Inverse Empirical Wavelet Transform ) reconstruction is proposed. The hybrid model are carried out as follows: the EWT is employed to decompose the raw wind speed series into several sub-layers; the LSTM network is executed as the principal predictor of each sub-layer; the RELM network is utilized as the subordinate predictor to model the forecasting error series of each sub-layer; the IEWT is adopted to construct the final forecasting series and filter the outliers. To validate the forecasting capacity of the proposed hybrid EWT-LSTM-RELM-IEWT model, seven different forecasting models are implemented on five wind speed time series. The experimental results demonstrate that: (1) the single LSTM model cannot provide satisfactory wind speed forecasting results in the involved wind speed time series; (2) the EWT can promote the wind speed forecasting accuracy and stability of the LSTM network significantly; (3) the RELM network based error modeling method improves the performance of the proposed EWT-LSTM-IEWT forecasting structure significantly; (4) the IEWT based outlier correction method is effective in promoting the wind speed forecasting accuracy and stability in the proposed EWT-LSTM-RELM structure; and (5) among all the involved models, the proposed hybrid model has the best performance in one-step to five-step predictions.