Prediction interval of wind power using parameter optimized Beta distribution based LSTM model

Abstract

Prediction interval of wind power (PIWP) is crucial to assessing the economic and safe operation of the wind turbine and providing support for analysis of the stability of power systems. The hybrid model (Beta-PSO-LSTM) of long short-term memory (LSTM) neural network and Beta distribution function based particle swarm optimization (PSO) is put forward for prediction interval of wind power. In order to enhance the performance of the Beta-PSO-LSTM for PIWP in training process, wind power series are divided into different power intervals, and then the Beta-PSO-LSTM is used to estimate each power interval of the original wind power series. Furthermore, based on the analysis of the interval forecasting error information in wind power training data set, Beta distribution model is proposed to get better PIWP, and PSO is used to optimize the parameters of the model. Finally, the proposed Beta-PSO-LSTM model is compared with the Beta distribution optimized by PSO based the BP neural network (Beta-PSO-BP), the normal distribution based LSTM neural network (Norm-LSTM), Beta distribution based LSTM neural network (Beta-LSTM), and Beta distribution optimized by iterative method based LSTM neural network (Beta-IM-LSTM) for PIWP. The simulation results show that the PIWP obtained by the Beta-PSO-LSTM model has higher reliability and narrower interval bandwidth, which can provide decision support for the safe and stable operation of power systems.