Forecasting power output for grid-connected photovoltaic power system without using solar radiation measurement

Abstract

With the increasing penetration of photovoltaic (PV) power system into the utility network, the issue caused by the fluctuation and intermittence of PV power output draws more attention. In order to predict the hourly power output for a PV system without any complex meteorological instrumentation, a NARX network-based forecasting model is proposed in this paper. Hottel's radiation model is applied to calculate clear-sky radiation incident on any inclined surface at any time. Furthermore, weather forecast data from public websites are taken as the variables that depict the cloud status in future days. By giving enough historical data, the network can be well trained. While forecasting power output, the NARX network can also adjust its weights according to the forecast error. Due to its adaptation to time-varying input, the proposed method can predict PV power output accurately even if the weather changes suddenly. Forecasting results of a roof-top PV power system verify the effectiveness of the proposed model with high precision and high efficiency.