Piecewise Clustering Prediction Model of Distributed Photovoltaic Power Based on Principal Component Analysis

Photovoltaic power generation is an important field of renewable energy development. However, when photovoltaic power generation is connected to the grid, the random fluctuation and intermittence of solar photovoltaic power generation caused by various environmental factors should be fully considered. Reliable prediction of photovoltaic power is a solution. Based on the analysis of two data analysis theories, which are cluster analysis and principal component analysis, this paper proposes a piecewise clustering method to improve the accuracy of RBF photovoltaic power prediction model. Firstly, the historical operational data after cleaning process are systematically clustered, and the corresponding target similar sample sets are obtained by two clustering methods. The RBF neural network algorithm is used to verify the prediction of the output power, which can effectively improve the stability and accuracy of the prediction results and simplify the number of input samples. Secondly, principal component analysis is carried out based on historical data, and the extracted principal component is taken as the input of RBF neural network photovoltaic power prediction model. It is proved by a case study that the proposed method can effectively improve the accuracy of prediction results and reduce the dimension of input samples, thus simplifying the network structure and shortening the learning time of neural networks. The effectiveness and reliability of the piecewise clustering method are verified by the case study.