A wind power prediction model with meteorological time delay and power characteristic

Wind energy, as a widely distributed, pollution-free energy, is strongly supported by the government. Accurate wind power forecasting technology ensures the balance of the power system and enhances the security of the system. In this paper, a wind power prediction model with the improved long short-term memory (LSTM) network and Adaboost algorithm was constructed based on the mismatch of data and power climb. This method was based on mutual information (MI) and power division (PD), named MI-PD-AdaBoost-LSTM. MI was used for quantifying the time delay between variables and power. Furthermore, to solve the relationship between wind speed and power in different weather fluctuation processes, the method of power fluctuation process division was proposed. Moreover, the asymmetric loss function of AdaBoost-LSTM was constructed to deal with the asymmetric characteristics of wind power. An improved artificial bee colony (ABC) algorithm, which overcame the local optimal problem, was used to optimize the asymmetric loss function parameters. Finally, the performance of different deep learning prediction models and the proposed prediction model was analyzed in the experiment. Numerical simulations showed that the proposed algorithm effectively improves the power prediction accuracy with different time scales and seasons. The designed model provides guidance for wind farm power prediction.