Hybrid model with secondary decomposition, randomforest algorithm, clustering analysis and long short memory network principal computing for short-term wind power forecasting on multiple scales

As the first prerequisite to carve out the increased exploration of the wind power generation and developments, accurate wind power prediction is sufficiently reliable to eliminate the dilemma caused by its intrinsic irregularity, intermittence and non-stationary. Therefore, the paper proposes the hybrid model composed of secondary decomposition, preliminary forecasting and error analysis, which can capture the fluctuation of the wind power series better, but also guarantee the forecasting stability simultaneously. More specifically, the secondary decomposition is developed to grasp the primary trend of a wind power series; Next, random forest algorithm, kmeans clustering and Long short term memory(LSTM) network are successfully employed to infer the latent characteristics of the decomposed modes as much as possible; For the sake of estimating the uncertainty associated with the preliminary results, the process based on LSTM network models the error sequences, of which the inherent information could be further mined. Then, the final predicted values are obtained by integrating the error sequences and preliminary results. Finally, the properties of the developed model are illustrated through wind power data from two wind farms. Besides, compared with the contrastive models, the proposed model presents 88.06%,96.35% improvements in terms of Mean Relative Error(MRE), Root Mean Square Error(RMSE) at most in the two cases, which demonstrates the superiority of the proposed model. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This paper proposes a hybrid model for wind power prediction, which can better capture the fluctuations of wind power data and ensure the stability of forecasting through secondary decomposition, preliminary forecasting, and error analysis.