Energy Consumption Prediction And Operation Optimization Of Integrated Energy System Considering User Behavior

China has been the largest energy consumption country.The energy structure of China is extremely unreasonable.Fossil energy still occupies a major position in the energy supply system.Through the overall planning and collaborative management of the supply,conversion,transmission,and utilization links,the integrated energy system can realize various energy complementation and meet the diversified energy use demands of users.It is an ideal way to improve the renewable energy uptake rate and energy utilization rate,which has become an inevitable choice to build a clean and low-carbon energy structure and realize the carbon peaking and carbon neutrality goals.But,the system operation effect depends on the optimal operation strategy.With the deepening of user involvement in the system,the interaction between supply and demand becomes more and more frequent.The user behavior is not easy to describe,the load random fluctuation is large,and the supply and demand are difficult to match,which makes the problem of system optimization particularly complex.To this end,we give more consideration to the characteristics of user behavior,focusing on solving the problems of low accuracy of load prediction and poor system operation effect in the integrated energy system.This paper conducts research from four aspects:user behavior analysis,multiscale load prediction,system operation optimization,and energy management platform development.The main contributions are as follows:(1)Aiming at the problem that the energy use behavior is difficult to quantify,the‘clustering-classification’method for energy use behavior pattern analysis is proposed.Firstly,a self-optimizing improved k-means++clustering algorithm is proposed to cluster the daily load curve divided by energy flow type to obtain user behavior patterns.Then,a variety of external factors are selected and combined with the clustering results to train the future behavior pattern classifier and obtain the key external factors affecting energy behavior.Mining energy uses behavior preference behind massive energy consumption data is an effective measure to reduce the complexity of load forecasting and improve forecasting performance.It is also the basis for customized and differentiated operation strategies to improve user participation and response efficiency.(2)In the face of the contradiction between the amount of training data and the prediction accuracy in practical engineering,the ’decomposing-merging’ prediction method is proposed based on the temporal characteristics of energy consumption.Firstly,the load variation is decomposed into linear components and nonlinear components by considering both periodic and random characteristics.Then,Holt-Winters is used to construct a linear prediction model.The extreme learning machine uses linear prediction results,nonlinear residuals,and original data as input to build a nonlinear prediction model.The convergence of the model is guaranteed,the cumulative error is reduced,and the ultra-short-term load is accurately predicted under the limited training dataset.(3)Aiming at the problems of low prediction accuracy of high-resolution day-ahead load forecasting model,the‘temporal-behavioral’prediction method is proposed.Firstly,the energy behavior analysis method and data decomposition technique are used to construct input vectors of behavior characteristics and time characteristics to reduce the negative impact of invalid data on its performance.Then,based on the deep learning network,a parallel analytical prediction model for behavior and time features processing multidimensional input is constructed.The experimental results show that the model can not only effectively improve the precision of highresolution day-ahead load prediction,but also has a strong ability to resist the inaccuracy of input data,so it is more suitable for engineering applications.(4)Aiming at the problem of complex user behavior leading to the operation efficiency of the integrated energy system,the ’source-storage-load’ collaborative optimization method is proposed based on energy use behavior analysis and load prediction.Firstly,the future energy behavior pattern is obtained through energy use behavior analysis.The user satisfaction parameters are introduced to quantify the degree of compatibility between scheduling results and energy use behavior,which is included in the optimization objectives.Then,the equipment model including energy supply equipment,energy use equipment,and energy storage unit is constructed.Finally,an integrated reinforcement learning algorithm is proposed to realize the cooperative scheduling of supply and demand,which not only meets the needs of users but also guarantees the operating efficiency of the system.(5)Considering the interaction between users and the system operation,based on the essential characteristics of multiple heterogeneous data and diverse equipment types of the integrated energy system,a user-oriented integrated energy system energy management platform was developed based on the "cloud-edge-end" framework,and some research results such as load prediction and energy use behavior analysis were verified in the platform.To sum up,this paper studies the difficulties of describing energy behavior,predicting high-resolution load,and taking into account user preferences in the operation process of the integrated energy system.An operational framework of energy behavior information guidance,load forecasting data support,and ’supply-demand’ collaborative scheduling decision-making is formed.While improving the operating efficiency of the system,it also takes into account the individual energy needs of users and provides theoretical and technical support for the development and construction of the comprehensive energy system.