Research On Multi-factor Short-term Load Forecasting Of Integrated Energy System Considering Demand Side Management

The integrated energy system(IES)is the future development trend of the energy industry and an important means to promote the supply side and demand side reforms in the energy field.At the present,the situation of energy supply and demand is unprecedented severe,and the relationship between supply and demand is unprecedented complex.Through the demand side management of the IES,the use of price signals and incentives to guide energy users to respond to demand is a powerful way to promote the balance of energy supply and demand,ensure the stability of energy supply,and improve the quality of energy supply.As the decision-making basis and data support of DSM,the forecasting performance of energy load forecasting model directly affects the demand response of users.Therefore,it is necessary to study the energy load forecasting model under the IES.This paper studies the energy load forecasting under the IES on three issues.First of all,the energy data under the IES is characterized by massive,heterogeneous and high-dimensional,which makes the energy data have different granularity,data loss,data anomaly and other problems in the process of collection,transmission,storage and use,thus leading to difficulties in matching the energy data with the prediction model.Secondly,the demand side users’ energy use behavior is affected by price signals and incentive measures to different degrees.The same form and intensity of demand side management is applied to users with high and low demand response potential,which is not conducive to accurate prediction of energy load,thus interfering with the formulation and implementation of demand response strategies.Finally,the machine learning model represented by deep learning has become an important guidance for energy load forecasting.The traditional energy load forecasting model cannot meet the requirements of enterprises for forecasting accuracy.The engineering application field urgently needs energy load forecasting models with higher allocation,better generalization and stronger forecasting performance.Based on this,this paper proposes a multi factor short-term load forecasting of IES taking into account DSM.The specific research contents and innovations are as follows:(1)Data driven abnormal load identification method.Aiming at the abnormal load of energy data,a recognition method based on Non-parametric Gaussian Kernel Density Estimation(NGKDE)is proposed,and the specific algorithm process and steps are given.In order to verify the effectiveness of this method,the Irish energy data set is used for an example analysis.(2)Data driven load feature recognition method.In order to identify energy users with high demand response potential and improve the accuracy of energy load forecasting,a feature recognition method based on energy user load feature curve and K-means clustering method is proposed.The extracted load feature curve of energy users is used for effective clustering to infer their demand potential.In order to verify the effectiveness of this method,the Irish energy data set is used for an example analysis.(3)Multi-factor feature selection method based on decomposition and correlation analysis.Aiming at the problem that the coupling between load and feature information is not tight,a multi factor feature selection method for energy load based on variational mode decomposition(VMD)and maximum information coefficient(MIC)is proposed.Through the correlation analysis of the decomposed energy load intrinsic mode function(IMF)components and the features of multiple factors,the input matrix is constructed and embedded into the prediction model,and the prediction results are obtained through weighted reconstruction.In order to verify the effectiveness of this method,the Arizona State University(ASU)Integrated Energy Data Set is used for an example analysis.(4)Load forecasting model considering price signal excitation.Considering the time lag of energy price signal guiding energy consumption of users,Hankel energy price matrix is established to strengthen the local space-time correlation between price vector and energy load.Based on the time series multitask Transformer deep learning model and Probsparse Attention Mechanism of Informer model,a multi factor short-term load forecasting model of IES with DSM is constructed,and the corresponding forecasting process and steps are given.Finally,the optimization performance of Ada Belief in IES load forecasting deep learning model is explored.In order to verify the prediction performance of the model,through comparative experiments and ablation experiments,a numerical example is analyzed on ASU Integrated Energy Data Set.