Ultra-short-term Photovoltaic Power Forecasting Considering Meteorological Factors

Influenced by various meteorological factors,the output value of photovoltaic power plants has strong fluctuation and uncertainty,which brings harm to power generation planning,grid-connected and economic dispatch of power system,and hinders the further promotion and application of photovoltaic power plants.Therefore,the grid-connected operation of large-scale photovoltaic power plants requires accurate photovoltaic power forecasting.In view of the fact that the output of photovoltaic power station is difficult to be predicted because of the change of meteorological factors.This paper presents a method of ultra-short-term photovoltaic power forecasting considering meteorological factors.The main research contents are as follows:In this paper,different meteorological factors affecting the output of photovoltaic power are analyzed.Firstly,the correlation between solar irradiation,temperature,wind speed,relative humidity and atmospheric pressure and photovoltaic output is analyzed.Then,the correlation between photovoltaic output and different meteorological factors under different weather types was analyzed in sunny,cloudy and rainy days.Through the analysis of meteorological factors in typical days and different weather types,the complex and non-linear coupling relationship between photovoltaic output value and different meteorological factors is obtained.Pearson correlation coefficient,mutual information and conditional mutual information are introduced to make the correlation between photovoltaic output and different meteorological factors more intuitive and reliable.Firstly,the original feature set is constructed.The original feature set includes historical photovoltaic power output data and meteorological factors that influence photovoltaic power output data.Then,the method of conditional mutual information is used to quantitatively analyze the feature correlation of the original feature set.The higher the correlation of features,the more important it is to forecast the output of photovoltaic power.By comparing different importance analysis methods,the effect of the relevance and redundancy of features is taken in the results obtained by conditional mutual information,which provides a theoretical basis for subsequent construction of low redundancy and high accuracy optimal feature subset.According to the analysis results of feature importance,a method of ultra-short-term photovoltaic power forecasting based on conditional mutual information feature selection is proposed.Firstly,the deep gated recurrent unit neural network is optimized and used as the prediction model of feature selection.Then,according to the analysis results of feature importance such as conditional mutual information,the features are arranged in descending order according to the feature importance,and the feature selection of forward sequence is carried out according to weather types,and the feature subset with minimum error is selected as the optimal feature subset.Finally,the optimal feature subset is combined with the optimized depth neural network model to carry out the o ultra-short-term photovoltaic power forecasting.The results show that the optimal prediction model constructed after feature selection and the prediction model constructed without feature selection have higher prediction accuracy;the optimal feature subset obtained by conditional mutual information feature selection has fewer feature dimensions,while ensuring that the prediction model has higher prediction accuracy.