Prediction of Photovoltaic Power Generation Based on PSO-RNN and SVR Model

Abstract

Considering the photovoltaic volatility, intermittency and random, the accurate prediction of photovoltaic power output is very important for grid dispatching and energy management. In order to improve the accuracy of photovoltaic system short-term power prediction, this paper analyzes the relationship between the power output and the environment factors. The principal component analysis (PCA) based particle group-ridge wave neural network model and support vector machine regression (SVR) for short-term prediction model are developed. In this paper, the PCA is used to reduce the number of input environment factors and extract the main components. The ridge wave neural network parameters are selected by particle swarm optimization (PSO). The SVR model is used to optimize the network structure for a better model performance. The correlation and reliability of the prediction results are discussed. The results show that, excluding the influence of weather interference factors, SVR has higher precision and accuracy in prediction model, smaller mean variance, and better prediction effect in the prediction mode.