Short-term Load Prediction And Energy Management Research Of Photovoltaic Changing Station

With the proposal of the "dual carbon" goal,the development of photovoltaic power stations and electric vehicles has been strongly supported by the state,and the integration construction of major electric vehicle photovoltaic charging stations has accelerated.Due to the random and mass charging behavior of electric vehicles,the optimal energy distribution solution cannot be accurately obtained when electric vehicles are connected to the regional power grid,resulting in a large number of disorderly charging loads affecting the regional power grid,easily aggravating the regional load peak-valley difference and reducing the economy of charging stations.To solve this problem,a forecasting model was established for users’ charging load,energy storage system and photovoltaic output,and on this basis,a multi-time scale energy management strategy of "day-day-real-time" was proposed to meet users’ charging needs,improve regional load peak and valley characteristics,and reduce the daily comprehensive operating cost of charging stations.First,the Stacking integration algorithm and its load prediction model are proposed.Key information such as the probability distribution characteristics of charging behavior of various electric vehicles and the initial state of SOC are used as the original data set.Monte Carlo model(MC)and the multi-variable residual modified gray prediction model(EMGM(1,6))are used as the base learner in the first phase.At the same time,it is trained by using the 50 fold cross-validation method,and the output result is taken as the training set and test set of the second stage.The Ridge regression prediction model(RR)is taken as the meta-learner of the second stage,and its output is the final prediction result.Comparative simulation experiments show that the absolute percentage errors of the single MC model and the EMGM(1,6)model are 5.62% and 5.84%,and the proposed integrated Stacking model is 4.79%.Secondly,in order to improve the prediction accuracy of the energy storage system,an IMM-dual-EKF strategy is proposed by combining the interactive multiple model(IMM)and dual-extended Kalman filter(Dual-EKF)models.First,the battery model corresponding to the actual state is matched by the interactive multiple algorithm(IMM).Combined with Dual filter strategy(dual-EKF)to determine its health state(SOH)and state of charge(SOC),the comparison experiment shows that the absolute percentage error of the Dual-EKF prediction model is 2.81%,and the absolute percentage error of the proposed IMM-Dual-EKF model is 0.77%.Thirdly,based on the prediction model of IMM-Dual-EKF energy storage unit and Stacking integrated load unit,a "day-day-real-time" multi-time scale energy management system for collaborative optimal scheduling of energy storage and load is proposed.In the day-ahead scheduling phase,the historical PV data and load data are used to output the day-ahead energy storage plan in advance.In the intraday stage,the energy scheduling plan of the energy storage unit is developed with the goal of minimizing the peak-valley difference by considering the minute-level source load prediction data.In the real-time stage,in order to minimize the deviation between the planned power and the actual power,an accurate SOC prediction model is introduced to realize the second-level control of the energy flow of the entire photovoltaic vehicle charging station.Finally,a 200 k W photovoltaic vehicle charging station in Guangyuan Road,Taiyuan City,Shanxi Province,was taken as an example for simulation analysis.The daily load peak-valley difference was 453.74 k W,and the daily comprehensive operating cost of the charging station was 2184.8 yuan.After applying this energy management model,the load peak-valley difference of the charging station was reduced to 186.23 k W.At the same time,the daily operating cost was saved by 6.13%.