Wind-storage combined system based on just-in-time-learning prediction model with dynamic error compensation

Abstract

High-performance wind power prediction (WPP) models can decrease the uncertainty of the whole energy system to balance energy supply and demand more effectively. However wind power process is characterized by the complicated mechanism and strong nonlinearity, which makes it difficult to establish a high-performance WPP model. In this paper, wind-storage combined system (WSCS) based on just-in-time-learning (JITL) prediction model with dynamic error compensation is proposed for high-performance WPP. This method can describe the characteristics of local wind power process efficiently and accurately by approximating the nonlinear process into a composite process with dynamic linear autoregression and static nonlinear error compensation. What's more, a WSCS model is proposed to indirectly improve the prediction accuracy. The model employs energy storage systems (ESS) to complement wind farm output to improve the wind power fluctuation and the performance of WPP. Simulation results verify that the proposed model has high prediction performance and generalization ability. Specifically, the RMSE, MAE and MAPE are 58.8728KW, 33.1277KW, 4.5877 % for gale season in a month. The RMSE, MAE and MAPE are 52.2014KW 33.2680KW, 2.3628 % for breeze season in a month. And WSCS indirectly improved the accuracy of the day-ahead prediction that reduced the RMSE from 414.4409KW to 182.2397KW. Accordingly, the research results can improve the WPP technology and reduce the redundant energy of wind power through ESS to facilitate the storage of ESS and improve the wind power fluctuation.