Multi-step ahead wind speed forecasting using an improved wavelet neural network combining variational mode decomposition and phase space reconstruction

Accurate wind speed forecasting is crucial to reliable and secure power generation system. However, the intermittent and unstable nature of wind speed makes it very difficult to be predicted accurately. This paper proposes a novel hybrid model based on variational mode decomposition (VMD), phase space reconstruction (PSR) and wavelet neural network optimized by genetic algorithm (GAWNN) for multi-step ahead wind speed forecasting. In the proposed model, VMD is firstly applied to disassemble the original wind speed series into a number of components in order to improve the overall prediction accuracy. Then, the multi-step ahead forecasting for each component is conducted using GAWNN model in which the input-output sample pairs are determined by PSR technique. Finally, the ultimate forecast series of wind speed is obtained by aggregating the forecast result of each component. The proposed model is tested using two real-world wind speed series collected respectively in spring and autumn from a wind farm located in Xinjiang, China. The experimental results show that the proposed model outperforms all other comparison models including persistence method, PSR-BPNN, PSR-WNN, PSR-GAWNN and EEMD-PSR-GAWNN models adopted in this paper, which demonstrates that the proposed model has superior performances for multi-step ahead wind speed forecasting. (C) 2017 Elsevier Ltd. All rights reserved.