PV Prediction based on PSO-GS-SVM Hybrid Model

Abstract

Photovoltaic power generation is affected by many factors, with volatility and intermittent characteristics. Large-scale photovoltaic access to the power grid poses great challenges to the safety and stability of power systems. Therefore, accurate prediction of photovoltaic power generation helps dispatchers adjust scheduling schedules in a timely manner, effectively reducing the adverse impact of photovoltaic power generation access on the power grid. This paper proposes a hybrid PV power prediction model based on PSO-GS-SVM. The particle swarm optimization (PSO) method is used to optimize the large step size of the support vector machine (SVM), and the parameter optimization range is obtained. GridSearch Method (GS) refined parameters optimization of PSO-SVM, and obtained PSO-GS-SVM hybrid model. The model is used to train and predict the normalized and dimensional sunny and non-clear working conditions data sets, and compared with BP neural network, SVM and PSO-SVM models. The results show that the PSO-GS-SVM hybrid model has better generalization ability and higher fitting effect.