Research On Photovoltaic Output Range Prediction Technology Based On Combined Mode

The energy crisis is becoming increasingly severe,and in response to this challenge,China is gradually increasing the proportion of renewable energy such as photovoltaics integrated into the power grid,which has a significant impact on the safe and reliable operation of the power system,and places new demands on power grid planning and dispatch methods.Uncertainty prediction with high accuracy can more comprehensively reflect the volatility and randomness of photovoltaics,measure the future output size and fluctuation range through probability information,and help the power grid better accommodate related resources and ensure the safe and smooth operation of large-scale renewable energy integration.To obtain a method with better prediction accuracy for the uncertainty of the photovoltaic output interval,two approaches were studied based on quantile regression and probability density,respectively,after analyzing the relevant influencing factors.The main contents are as follows:Firstly,the data sources are introduced and the missing values and outliers are processed.The seasonal characteristics of PV output,the correlation analysis between each meteorological factor and PV output,the correlation analysis between meteorological factors and the autocorrelation characteristics of PV output are analyzed with the help of statistics and correlation coefficients.Secondly,based on the analysis of the seasonal characteristics of PV output,a photovoltaic output interval prediction model based on extreme learning machine quantile regression(QRELM)was constructed.Principal component analysis was used to reduce the dimensionality of meteorological factors,and models were separately trained for each season to improve prediction accuracy.To optimize prediction accuracy,an upper and lower bound quantile set combination structure was designed,and an improved differential evolution algorithm was used to solve the combination weights.Experiments showed that season-specific training models and upper and lower bound quantile set combination structures can effectively improve prediction accuracy.Finally,to obtain richer probability information,a photovoltaic output interval prediction model based on quadratic decomposition long short term memory networks was constructed.The photovoltaic output was decomposed twice using singular spectrum analysis and variational mode decomposition.After predicting each component combined with meteorological factors,point predictions were reassembled and error sets were constructed.Combining kernel density estimation to obtain the probability density of prediction errors and converting them to prediction intervals,a correction coefficient based on sparrow search algorithm was introduced to improve prediction interval accuracy.Experiments verified the effectiveness of quadratic decomposition and correction coefficients in improving prediction accuracy.