Ultra-short-term prediction of photovoltaic output based on an LSTM-ARMA combined model driven by EEMD

Abstract

A new method is proposed for ultra-short-term prediction of photovoltaic (PV) output, based on an LSTM (long short-term memory)-ARMA (autoregressive moving average) combined model driven by ensemble empirical mode decomposition (EEMD) and aiming to reduce the intermittency and uncertainty of PV power generation. Considering the superposition of the overall trend and local fluctuations contained in the PV output data, an EEMD adaptive decomposition criterion based on continuous mean square error is proposed to extract the various scale components of the PV output data in the time-frequency domain; an ARMA (autoregressive moving average) model suitable for short correlation analysis is constructed for the intrinsic mode function components that characterize local fluctuations of PV output. Environmental parameters such as solar radiation, temperature, and humidity are introduced to construct a LSTM prediction model with autocorrelation capability and environmental characteristics for the EEMD residual that characterizes the overall trend of PV output. Finally, the overall trend and the local fluctuation forecast results are fused to realize an ultra-short-term forecast of PV output. The training set and test set were randomly selected from the PV microgrid system of Hangzhou Dianzi University and used for PV output prediction according to different seasons and weather types. The maximum MAPE on sunny, cloudy, and rainy days was 23.43%, 32.34%, and 33.10%, respectively. The minimum MAPE on sunny, cloudy, and rainy days was 5.53%, 6.47%, and 19.19%, respectively. The results show that the prediction performance of this method is better than traditional models. The ultra-short-term forecasting method for PV output proposed in this paper can help us to improve the safety, flexibility, and robustness of PV power systems.