One-day Ahead Forecast of PV Output Based on Deep Belief Network and Weather Classification

Due to reducing on the cost of photovoltaic (PV) panels and the rising awareness of environmental protection, PV power generations are growing rapidly in recent years. Because of the unstable and intermittent nature of PV power generation, it is normally considered as a risk to incorporate PV powers into the main grid. One solution is to predict the output of PV station in advance and adjust the output of energy storage resource accordingly. This paper proposes a method to realize one-day ahead forecast of PV power output based on weather classification and Deep Belief network (DBN). Firstly, different weather conditions are clustered into four weather types based on the average values of historical PV power outputs and K-means clustering algorithm. Then all historical PV output data of the station are divided into four groups according to the weather type. Different DBN models are built for each of the four weather types. The inputs of each DBN models are historical PV power output value of the similar day about forecasting day and the meteorological value of the forecasting day. Experimental results with a Baoding PV station prove the effectiveness of the proposed method.