A dual-scale deep learning model based on ELM-BiLSTM and improved reptile search algorithm for wind power prediction

Abstract

Accurate wind power forecast is critical to the efficient and safe running of power systems. A hybrid model that combines complementary ensemble empirical mode decomposition (CEEMD), sample entropy (SE), random forest (RF), improved reptile search algorithm (IRSA), bidirectional long short-term memory (BiLSTM) network and extreme learning machine (ELM) is proposed for wind power prediction in this paper. Firstly, the CEEMD decomposes the non-stationary original wind power sequence into comparatively stationary modal components, and sample entropy aggregation is used to decrease the computational complexity. Secondly, redundant features are further eliminated through random forest feature selection. Thirdly, the BiLSTM model and the ELM model are applied to forecast high and low frequency components, respectively. IRSA is used to optimize the model's parameters. Finally, the predicted value of each component is summed to arrive at the final predicted value of wind power. By comparing with ten other models, the results show that the dual-scale ensemble model of BiLSTM and ELM can obtain better prediction accuracy. The RMSE of the model proposed in this study is reduced by more than 10% compared with other benchmark models, which demonstrates that the proposed model can better fit the wind power data and achieve better prediction results.