Research On Comprehensive Energy Time Series Forecasting Method Considering Multi-dimensional Data Coupling

As a form of energy management that realizes the multi-energy complementary characteristics of the energy Internet,the integrated energy system(IES)has been rapidly developed at home and abroad during recent years.At the same time,the rapid development of distributed renewable energy has made the integrated energy system gradually develop into an energy system integrating renewable energy,electric energy,thermal energy and natural gas.The spatial coupling characteristics between users’ multi-energy loads in the integrated energy system and the randomness and variability in time series make accurate prediction of multi-energy loads more difficult.In this paper,the following researches are made on the comprehensive forecasting method of multi-energy load for mass users of integrated energy system.Aiming at the problem of analyzing the energy consumption behavior characteristics of massive multi-energy users in the integrated energy system,a cluster analysis method considering the coupling of multi-dimensional data is proposed.First,based on the Copula function,a quantitative analysis of the coupling characteristics between the electricity-heat-gas-light time series of multi-energy users is carried out,and the quantified value is incorporated into the multi-energy characteristic index,so as to extract the energy consumption behavior characteristics of multi-energy users.Then a hierarchical clustering framework for multi-energy users is established,and an adaptive k-means clustering algorithm is propo sed.Finally,the typical clustering results are obtained through simulation analysis,which provides a theoretical basis for the structure of the user’s multi-energy time series forecasting model data set and the multi-energy comprehensive forecast.Aiming at the problem of electricity-heat-gas-light time series forecasting for multi-energy users in the integrated energy system,a comprehensive forecasting method based on Capsule neural network(CapsNet)is proposed.Firstly,the coupling feature between the electricity-heat-gas-light time series of multi-energy users is extracted by convolution operation.Then,the multi-dimensional time capsule is used to encapsulate the time series information of the multi-energy coupling feature,and the dynamic routing mechanism is used for learning,so as to capture the nonlinear time correlation based on the extracted multi-energy coupling feature.Further,a fully connected linear regression layer is used to integrate the coupling-time features and generate the final prediction result.In addition,a two-layer iterative training method is used to adjust model parameters and speed up convergence.Finally,the user’s actual multi-energy data is used to compare and analyze the simulation results,which verifies the feasibility and effectiveness of the multi-energy comprehensive prediction method based on the capsule neural network proposed in this paper.Aiming at the multi-energy time series forecasting problem of the distributed heterogeneous user groups of the integrated energy system,a distributed forecasting method considering the spatiotemporal coupling of multi-users is proposed.First,a quantitative analysis of the spatio-temporal correlation of multi-user and multi-energy is carried out.Second,the multi-energy time series data from distributed users is converted into time-sequential images in three-dimensional space,and the implicit coupling-space in the image is extracted by convolution.Then,the dynamic routing mechanism is used to extract the time features of multi-user and multi-energy coupling-spatial features,and the regression layer activated by the Leaky ReLU function integrates the coupling-spatial features and generates the final prediction result.Finally,the comparison and analysis of simulation results verify the effectiveness and superiority of the multi-energy time series distributed forecasting method proposed in this paper.In summary,this article focuses on the forecasting of the massive heterogeneous multi-energy user electricity-heat-gas-light time series in the integrated energy system.Based on the capsule neural network,a comprehensive forecasting model considering the coupling of multi-user and multi-energy—temporal and spatial characteristics is proposed.It improves the accuracy of multi-user and multi-energy time series forecasting,and provides a theoretical basis for the energy management and planning of multi-energy users by the integrated energy system energy center.