Multiple Load Short-Term Prediction Of Integrated Energy System Based On Multi-Model Fusion

To realize the "dual carbon" goal,the Energy structure of our country is gradually developing in the direction of Integrated Energy System(IES).IES can improve the efficiency of energy utilization and reduce the loss of energy in the process of energy conversion between energy subsystems and in the process of production,transmission and distribution under the premise of meeting the demand of multiple loads.In order to establish intelligent,stable,economical and sustainable IES,it is necessary to analyze the multi-load demand.Therefore,the accuracy of IES short-term load forecasting is particularly important.Aiming at IES multi-load short-term forecasting,this paper studies from two time directions.VMD-SE-MLR-Forecast Net was used to forecast the day ahead load.First,the pearson correlation coefficient method is used to quantify the correlation coefficient,analyze the coupling between multiple loads and select meteorological factors with high correlation.Meanwhile,the autocorrelation analysis method is used to analyze the timing rule of load data.Secondly,the KNN algorithm and polynomial interpolation are used to detect and replace the abnormal values in the load data.Then the load is reorganized into each sub-sequence according to the timing law,so as to reduce the prediction span and improve the prediction accuracy.Then,VMD decomposition is used to decompose each sub-sequence into multiple sub-sequences with different frequencies.Through sample entropy analysis,high-frequency and low-frequency sequences with different frequencies are found.The low frequency series was predicted by multiple linear regression,and the high frequency series was predicted by the model with better effect on the test set.The day-ahead forecasting model finally selected in this paper is VMD-SE-MLR-Forecast Net.For IES pre-load forecasting,an error expansion coefficient was introduced to evaluate the forecasting performance of different models at different time points on the basis of pre-load forecasting,to alleviate the error expansion phenomenon of a single forecasting method in multi-step forecasting,so as to achieve multi-model fusion and determine the final pre-load forecasting model.By comparing with other models,the results show that the multi-model fusion prediction model proposed in this paper has higher prediction accuracy both ahead of the day and ahead of the week.