Short-term Wind Power Forecasting Method Based on Deep Recurrent Belief Network

The randomness and volatility of the wind energy seriously affect the forecasting accuracy of the wind power. The forecasting accuracy can be improved by digging the high-order characteristics between the different fluctuation degrees of the wind speed and the forecasting power. This paper firstly uses the swing window algorithm to identify the fluctuation process of the wind speed, and clusters the different fluctuation degrees by the generalized first search neighbor algorithm. Then, the different fluctuation processes are used as the classification training data of the deep recurrent belief network. The deep recurrent belief network is composed of two error feedback networks: the forward generating network and the horizontal and longitudinal error feedback networks, and the cross-entropy of error distribution is taken as the loss function to effectively control the direction of error iteration and the training scale of the model. The results of case studies indicate that the forecasting method proposed in this paper can improve the forecasting error in the fluctuation process. © 2021 Automation of Electric Power Systems Press.