Long Short-Term Memory Network based on Neighborhood Gates for processing complex causality in wind speed prediction

Abstract

Abstract Obtaining high-precision wind speed prediction results is very beneficial to the utilization of wind energy and the operation of the power system. The purpose of this study is to develop a novel model for wind speed causality processing and short-term wind speed forecasting. In this study, the hybrid model combining causality processing strategy called “decomposition- virtual nodes-pruning” and Long Short-Term Memory Network based on Neighborhood Gates is proposed to obtain high-precision wind speed predictions. First, Pearson Correlation Coefficient, Maximal Information Coefficient and Granger causality test are used to explore the correlation and causality between wind speed and meteorological factors. Then, the causality is divided into five categories: center, chained, ring, tree and network causality, according to the topological structure of causality. Next, all types of causality can be unified into an equivalent tree causality by the causality processing strategy. Afterward, Long Short-Term Memory Network based on Neighborhood Gates is proposed to dynamically adjust the network structure according to the specific equivalent tree causality. Finally, the performance of the proposed model is verified by eight models from three aspects with different features, different methods and different equivalent trees in the case in Fuyun meteorological station, Xinjiang province, China. The evaluation metrics of the prediction results obtained by the proposed model are optimal among the eight models. The experimental results show that the proposed model is very competitive and very suitable for processing complex causality in wind speed prediction.