Research On Forecasting Error Modeling Of Grid-connected Photovoltaic Power Generation

With the concept of the Energy Internet proposed,scholars from all over the world attach great importance to build an open and shared Energy Internet with large-scale distributed renewable energy.However,the primary problem facing the realization of electric energy replacement and cleaning substitute is the optimal dispatching,electric energy storage and integration problem caused by grid-connected renewable energy generation with stochastic fluctuation characteristics.Forecasting error is regarded as the representative variable of renewable energy output uncertainty.The accuracy of the forecasting error model affects directly the stochastic optimization scheduling and decision-making scheme of the power system containing the high proportion of renewable energy.Research on wind power forecasting error can be obtained,but there is little literature on the prediction error of photovoltaic power generation.In this paper,the forecasting error of photovoltaic power generation is taken as the main research object.The distribution characteristics and distribution model of prediction error that may be brought about by the pre-day and intra-day prediction are analyzed.The influence of different numerical conditions on the distribution characteristics of prediction error is studied,and the error is clustered accordingly.The conditional probability model is established to describe the error distribution more accurately.The specific research content is as follows.The analysis of numerical characteristics was carried out by means of Analysis of Variance,fuzzy C-means(FCM)clustering and numerical chromaticity graphs.The results show that the overall level of pre-day PV output prediction error is related to weather type,temperature difference and temperature and the error value has a correlation with the predicted output amplitude and power variation at that time.There is a strong temporal correlation between intra-day PV output forecasting errors within one hour,and the magnitude of the time error is related to the numerical characteristics of the predicted power output at the certain time.According to the characteristics of pre-day power prediction,this paper comprehensively considers the influence of meteorological factors and predictive output numerical characteristics on forecasting error.The whole level of the forecast errors will be classified through the FCM algorithm,and the following classification on the output forecast will be conducted according to its numerical characteristics.Then,the model for estimating the error distribution is established.For the characteristics of intra-day power prediction,considering the influence of time correlation and predictive output numerical characteristics,according to the time correlation of the forecasting errors and FCM algorithm,the first clustering is conducted on the time errors,and the second clustering is performed based on the numerical characteristics of the predicted output.Finally,the model suitable for estimating the intra-day PV output forecasting error distribution is established.The distribution of the forecasting error after clustering shows skew,kurtosis characteristic,and multi-peak phenomenon that may occur.A General Gauss Mixed Model(GGMM)is established in this paper.The superiority and effectiveness of this model compared with other models are verified by historical data examples of several photovoltaic power plants at home and abroad.And the analysis method is free from the effects of seasonal,meteorological factors,prediction algorithms and geographic information of photovoltaic power plants.