Research On Power User Attribute Mining And Load Forecasting Technology

With the development of the smart grid,smart meters are widely used.Smart meters can easily collect energy consumption data and use these data to solve practical problems.The analysis of demand-side power consumption data can help us obtain power user behavior and accurately predict power load,which can provide auxiliary support for grid emergency dispatching,and also help grid companies provide users with personalized services,ultimately realizing the two-way interaction between users and the grid and promoting the development of smart power consumption.In this thesis,user pattern recognition,user characteristic recognition and load forecasting are analyzed firstly.Then,a model of social and economic information analysis of power users based on attention mechanism is introduced,which can effectively identify the social and economic information and analyze the relationship between users’ electricity consumption behavior and socio-economic data.Finally,a user load forecasting model is constructed by user clustering,which improved the accuracy of load forecasting of power users.The main contributions of this work are as follows:(1)The customers’ electricity consumption behavior and social-economic information are analyzed based on a public dataset.An attention-based model is constructed to explore the relationship between users’ behavior and their social-economic status.The model extracts the features of the high-dimensional electricity consumption curves,identifies ten users’ social-economic information through the classification network,and analyzes the correlation between different social-economic information and electricity consumption patterns.Finally,the validity of the proposed method is evaluated by comparing it with other classification models.(2)A cluster-based individual-level load forecasting method is proposed,which clusters the power consumers and builds a forecasting model for each group.The experimental results on different cases show that the proposed method reduces the time cost and effectively improves load forecasting performance.