A Hybrid Method for Hour-ahead PV Output Forecast with Historical Data Clustering

Abstract

An accurate forecast of the photovoltaic (PV) output can indispensably enhance the stability of power supply. However, the effectiveness of prediction results highly depends on many factors and the data preparations are also an essential process to be considered. The focus of this paper is to classify PV data types by density-based spatial clustering of applications with noise (DBSCAN). Then, the procedure for historical data clustering and for hour-ahead PV output forecast implemented by back propagation neural network (BPNN) model is present. Forecasted results show that data clustering provided by DBSCAN can efficiently classify the PV data types for input to BPNN to achieve better accuracy.