Multivariate Load Forecasting For Integrated Energy System Based On Multi-task Learning

The integrated energy system is a new type of energy system that integrates and utilizes various energy resources to improve energy utilization efficiency and reduce environmental pollution.In this kind of system,the generation and consumption of energy are coupled with each other,and it is necessary to predict the multivariate loads to ensure the stable operation of the system,Therefore,multi-load forecasting of integrated energy systems has become one of the research hotspots.The forecast results can provide strong support for the planning,scheduling and operation of energy systems,so as to achieve optimal allocation of energy resources and sustainable economic development.Aiming at the problem that the current load forecasting method of a single energy type is transferred to the multiple load forecasting of the integrated energy system,the implicit coupling information between the multiple loads will be ignored,and a multi-task learning-based multi-load forecasting method for the integrated energy system is proposed.Improve multivariate load forecasting accuracy.Multi-task learning needs to solve two problems of inter-task correlation analysis and sharing mechanism design.In terms of correlation analysis between tasks,this paper first uses the DBSCAN clustering algorithm to preprocess the data set with outliers to avoid the impact of outliers on the data analysis results.Then,the multi-time scale analysis of the multivariate load characteristics is carried out to provide a reliable basis for the correlation analysis of multivariate loads and the construction of forecasting models.At the same time,the dynamic correlation analysis method is introduced to analyze the correlation of multiple loads,which further improves the credibility of the analysis results.Finally,the correlation between influencing factors and multivariate loads was analyzed,and the data with significant correlation were screened out,which provided the basis and support for the input characteristics of the prediction model.In terms of sharing mechanism design,this paper considers the strong correlation between multiple loads,adopts the hard sharing method of multi-task learning parameters to predict the multiple loads of the integrated energy system,and uses the GRU network as the sharing layer of the model.GRU-MTL Multivariate Load Forecasting Model for Integrated Energy Systems.In order to further improve the prediction accuracy of the model,in view of the problem that the particle swarm optimization algorithm is easy to fall into local optimum,the inertia weight and learning factor are improved,and an improved inertia weight adjustment strategy and learning factor adaptive mechanism are proposed.It enables the algorithm to search more fully in the space,avoiding prematurely falling into the local optimal solution,and uses the improved particle swarm optimization algorithm to tune the hyperparameters in the model,and proposes a regional synthesis based on IPSO-GRU-MTL Multivariate Load Forecasting Models for Energy Systems.Finally,the effectiveness of the model proposed in this paper is verified by setting an example.The experimental results show that the model proposed in this paper can effectively predict multivariate loads,and has high prediction accuracy and small error rate.The load management of the energy system provides strong support and reference.