Research On Methodology Of Short-term Photovoltaic Power Forecasting

With the development of related technology,the unit cost of photovoltaic has been declining in recent years,bringing a rapid development of PV power generation.Solar energy,as a kind of clean and renewable energy,can be used to develop PV power generation,which can greatly alleviate energy crisis and reduce environmental problems caused by fossil energy burning.However,due to the intermittence of surface solar radiation,PV power output tends to be volatile and intermittent.Consequently,the stability of power system will be affected when the installed capacity of gridconnect PV plant becomes large.Therefore,it is important to perform an accurate prediction on PV power output,so as to make proper power planning and scheduling.Based on the data of two real grid-connected PV power plants,the author analyzes and summarizes the research progress in related fields in recent years,and conducts a detailed study on short-term and ultra-short-term PV power forecasting.This paper mainly discusses the following aspects:(1)The author designs a calculator of extraterrestrial radiation based on related theory and a database for PV power prediction through abnormal point monitoring,effective time interval determination,missing data interpreting and data normalization.(2)A short-term PV power forecasting model based on ELM-SVM is proposed in this paper.Firstly,the author divides weather into three typical types: sunny,cloudy and rainy based on the correlation coefficient between PV power output and extraterrestrial radiation provided by daily weather forecast,and sets up three sub forecasting models.Then targeted parameters in each sub-model are selected according to the feature of each typical weather type by using Pearson correlation coefficient.Finally,a method called score competition is proposed to select regression model,in which ELM is selected of sunny,and SVM in cloudy and rainy days.It can be concluded that the ELMSVM hybrid model shows stronger adaptability and better prediction accuracy compared with single forecasting models because it can combine the advantages of different regression models.(3)Historical power output data are used as the input of the model.The author proposes two kinds of ultra-short-term PV power forecasting models based on ELM.Compared with BP neural network,ELM brings better prediction accuracy.Finally,the author establishes sub-models according to the error distribution in different time intervals.The result shows that time-segment forecasting model based on ELM has better performance especially in unstable weather conditions.