Research On Very-short-term Prediction Methods Of Photovoltaic Generation Based On Data Driven

Solar photovoltaic(PV)power generation is considered an important replacement to traditional power generation due to its renewable and zero pollution characteristics.PV power generation forecasting technology can provide a period output power in advance for the PV system,providing a basis for power supply company and users,and it has great significance for power system and ensure the system operate steady and safety.Based on the historical data of PV power generation system output,this thesis analyzes the rule of PV power generation output under different weather conditions,proposes forecasting methods and verified the methods by a demonstration system.The main contents of this thesis are as follows:Firstly,the historical PV power generation output data are divided into three different types as sunny,cloudy,and rainy day based on the weather conditions.The divided output data is reconstructed in phase space and the chaotic characteristic of data is analyzed by Lyapunov exponent method.Secondly,a new method for very short-term forecasting based on the maximum Lyapunov exponent is proposed.Using the maximum Lyapunov exponent method,the PV generation under different historical weather conditions is forecasted and the forecasting accuracy is evaluated via the actual PV power generation data.Thirdly,considering the dramatic movements of clouds in cloudy days,a forecasting method toward cloudy days is proposed.This method is based on Hammerstein-Wiener(HW)and nonlinear autoregressive(NAR)models.Taking photos with an industrial camera,a HW model jointing the photo brightness and PV power is established,and the NAR method is used to forecast very short-term PV power generation with the identification model output.The actual PV power generation data is used to evaluated the forecast results.With comparing with the existing PV data processing method,this thesis combines the meteorological information provided by the weather forecast and Euclidean distance method to classify the PV data.The result is more scientific and theoretical.After the phase space reconstruction,the historical PV time series data based on different weather types show the chaotic characteristic as seemingly stable and unstable,confusion and not confusion,and this characteristic is proved by the Lyapunov exponent.The maximum Lyapunov exponent method is used to forecast the PV power in different weather conditions and the HW-NAR model is used to forecast the cloudy days as a supplement.The results suggest that those forecasting methods can provide higher accuracy and have a good guiding effect for the operation of PV system.