Study On PV Power Prediction Method Based On Multiple Linear Regression With Time-series And Stochastic Error Correction

Nowadays,with the installed capacity of PV power station in our country increasing continuously,the uncertainty of photovoltaic power generation which may lead to the instability of power grid cannot be ignored any more.“Solar Curtailment” is the main method to solve that problem,but it brings great waste of Photovoltaic resources.PV power prediction can improve the utilization of the photovoltaic resources to a large extent.Therefore,it is essential to study a precise PV power prediction method which can improve photovoltaic utilization and maintain the stability of power grid.In this paper,A Multiple Linear Regression PV Power Prediction with Time-series and Stochastic Error Correction(MPTSEC)was proposed.Firstly,it studied the relationship between PV power output and five meteorological factors like temperature,relative humidity,wind speed,dew-point temperature and cloud cover,respectively,while calculated Pearson correlation coefficient of each relationship and chose the meteorological factors with bigger coefficient as inputs of the forecasting model.Then,it built Multiple Linear Regression(MLR)periodic power prediction model based on the linear relationship between meteorological factors and PV power output.Moreover,it analyzed the residual sequence of MLR prediction models to get residual prediction result,and used residual prediction result for correcting the MLR prediction result to get Multiple Linear Regression PV Power Prediction with Time-series Error Correction(MPTEC)periodic power prediction result which verified the effect of residual prediction on improving periodic power prediction accuracy.Finally,it used chance constraint programming to predict random power output,and used the random power prediction result for correcting MPTEC prediction result to get MPTSEC prediction result.MPTSEC prediction result verified the effect of random power prediction on improving the power prediction accuracy,and it also showed a relatively accuracy,compared with traditional methods like BP and SVM.