Research On The Output Power Forecasting Of Large-scale Grid-connected Photovoltaic Station

At present, along with the large-scale of photovoltaic power generation system isapplied widely, more and more problems rise now. Solar irradiation have arelationship with seasons, day and night, cloudy and clear, so the output power ofPhotovoltaic(PV) power generation system have the inherent disadvantages such asintermittent and randomness, and energy storage technology immature now. Whenlots of PV power generation system connected to the electricity grid, which will bringchallenge to the safety and stabilization of power grid and then restrict the scale ofsolar power development speed and scale. So output power forecasting of PV powergeneration system has a very vital significance for power system operation. Thisthesis studied short-term solar power forecasting method. The main works are asfollows.Firstly, the statistics method is used to analyse the time series characteristic ofsolar irradiance. This thesis proposed the daily solar radiation forecasting model byuse of Recursive Least Squares (RLS) and Mallat algorithm. According to themeteorological data, the day with different conditions is classified to different daytypes, then built different forecasting model for different day types based on thehistory measurement data. The wavelet decomposition and reconstruction is used inthe solar irradiation, the solar irradiation time series with tendency are decomposedinto a low frequency component and several high frequency components. The highfrequency signals and the low frequency are forecasted with RLS. The forecastingresult of the original time series is the superposition of the respective forecasting, andforecasting results show that the forecasting model is not only effectiveness but alsopractical. Secondly, all sorts of meteorological factors are reasonably selected andprocessed, a method to select similar days of PV generated system is proposed, apower forecasting model based on similar day and Radial Basic Function (RBF)neural network is designed. The forecasting model is verified by the real data of acertain PV generating system located in northwest China, and forecasting results showthat the forecasting model is not only feasible but also practical.