Short-term photovoltaic power forecasting method based on K-means algorithm and spiking neural networks

Abstract

In view of the adverse impacts on the power grid dispatching, energy management and other aspects caused by the volatility, intermittence and uncertainty of photovoltaic (PV) power generation, this paper proposes a novel short-term PV power forecasting method based on K-means algorithm and spiking neural networks (SNN). Considering meteorological factors and seasonal characteristics, this approach utilizes the K-means algorithm to cluster different short-term PV power generation characteristics and implements SNN training with different kinds of PV power generation data obtained from clustering results. In addition, an improved particle swarm optimization (PSO) algorithm is employed to globally optimize the synaptic weights of SNN, which enhances the pattern recognition ability and numerical stability of the forecasting model. We assess the performance of the proposed method by using actual PV power generation data in a certain area for simulation analysis. The simulation results illustrate that the proposed method can achieve a better prediction accuracy than the conventional BP neural network and support vector machine (SVM), and performs excellent engineering application potential.