Application of preconditioned Generalized radial Basis Function Network to prediction of photovoltaic power generation

In this paper, an efficient method is proposed for short-time generation output prediction of PV systems. The prediction of time-series of PV generation output is too complicated to handle. The proposed method focuses on the improvement of the prediction model accuracy with a hybrid intelligent system. It consists of the precondition of input data and the predictor of multi-step ahead PV generation output. The former deals with clustering of input data to improve the performance of the predictor. It is very useful to classify data into some clusters and construct the prediction model at each cluster so that the prediction is improved due to the data similarity in the cluster. As the clustering method, DA (Deterministic Annealing) Clustering of global clustering is used due to the good performance. On the other hand, the latter makes use of an advanced GRBFN (Generalized Radial Basis Function Network) of ANN (Artificial Neural Network) as the predictor. As a result, the proposed method provides better results than the conventional ones. The effectiveness of the proposed method is demonstrated to real data of short time prediction of PV systems.