A K-means cluster division of regional photovoltaic power stations considering the consistency of photovoltaic output

Abstract

The uncertainty and volatility of photovoltaics seriously impact the grid's power quality. Short-term photovoltaic(PV) forecasts have a positive effect on the stable operation of the power system. The accuracy of cluster division is a key factor in the output prediction of regional PV power stations. This paper proposes a cluster division method, including a novel feature selection technique and an optimized cluster algorithm based on K-means. The proposed method performs feature analysis and parameter optimization of the division of regional photovoltaic plant clusters, analyzes the clustering dimension of photovoltaic output consistency, and establishes a K-means clustering model of photovoltaic power plants that considers time, space, and inherent characteristics of power plants first. Then, a prediction model based on Long Short-Term Memory (LSTM) is established for each cluster to realize the prediction of regional cluster photovoltaic output. The simulation results demonstrate that the Mean Absolute Percentage Error (MAPE) of the proposed method is 18.27 % and Root Mean Square Error (RMSE) is 45.79 %, which verifies the superiority of the proposed method over comparison models. It shows that the proposed method can effectively solve the problem of low prediction accuracy caused by weak output consistency of power stations in regional photovoltaic clusters.