Abstract: Since the limitation of the primary energy, the increasing energy consumption and more concern on environmental protection, much attention has been devoted to explore and utilize the renewable energy. Photovoltaic output power forecasting is the foundation of the planning and operation of photovoltaic system. In this paper, a novel short-term PV power forecasting integrating knowledge mining method and intelligent algorithm has been proposed. The main idea of this method is that firstly though knowledge mining it analyzes the important factor impacting the photovoltaic array output. Then Fuzzy c-mean clustering is adopted to classify the history data and the meteorological data on forecasting day. The selected subset including high similarity days would improve the quality of the training samples. Later, Bayesian neural network is built to mapping the complex relationship among the input data and the PV power, and the parameters of the network is optimized according to Bayesian theory to improve the generalization of the model. At last, to valid the effectiveness and accuracy of the proposed method, simulation is carried out. The forecasting result shows the goodness of this method by comparing with traditional BP network.

Key words: photovoltaic array, power forecast, similar day, fuzzy c-mean clustering, bayesian neural network