Research On Correlation Analysis And Load Prediction Method Of Energy Consumption Behavior Based On Comprehensive Energy Big Data

With the development of energy utilization technology,to improve the efficiency of energy supply and reduce the uncertainty of users’ energy consumption behavior,the integrated energy system has attracted more and more attention.In addition,with the development of sensors and industrial Internet of things technology,there are more and more measurement types and higher sampling frequency in the integrated energy system.With the accumulation of big data,mining data value from users’ energy consumption data has gradually become a hot topic.Under the background of big data of integrated energy system,the analysis and prediction of energy consumption behavior is conducive to the stable and economic operation.In this paper,the data of integrated energy system are cleaned firstly,and the subsequent energy consumption behavior correlation analysis and load prediction are carried out based on the clean data.The main works are as follows:（1）In order to solve the problem of abnormal data in the integrated energy system,this paper proposed a joint cleaning method for multi load data of the integrated energy system based on seasonal decomposition regression and MICE multiple filling model.Firstly,the real data set of building energy consumption was analyzed for the data missing and anomaly,and then an outlier detection algorithm of multivariate data based on isolated forest was proposed.The external data was processed by STL,and the multi data was cleaned by MICE model.The proposed data cleaning method takes into account the coupling characteristics of multiple loads,which helps to improve the accuracy of data recovery.（2）Aiming at the problem of insufficient analysis of users’ multiple energy demand and consumption behavior,a load level correlation analysis method of integrated energy system based on FP-growth algorithm was proposed.Firstly,k-means was used to discretize the data of cooling load,heating load and electric load to obtain the classification method of multi energy load level;Secondly,the discretization method based on entropy was used to discretize the meteorological data;Finally,according to the proportion distribution of different load levels,the minimum support and minimum confidence were determined,and the correlation analysis of comprehensive energy load levels was carried out.The proposed method can mine the relationship between users’energy consumption behavior and external factors,and provide a reference for understanding users’ energy consumption behavior.（3）Aiming at the problem of energy consumption load forecasting in integrated energy system,this paper proposed an ultra-short-term multi-energy load forecasting method based on MMoE multi-task learning strategy and attention mechanism.MMoE model was used to fully consider the strong correlation and weak correlation among the three types of loads.Combined with GRU network,a load forecasting model was established,and attention mechanism was added to improve the forecasting performance of the model.The proposed method can fully consider the coupling characteristics of cooling,heating and electric loads in the integrated energy system,improve the prediction accuracy of multi energy load and reduce the modeling complexity.