Analysis And Application Research Of Power Consumption Behavior Based On Data Mining

The popularity of smart grid has produced massive monitoring data,which is very important for grid planning,operation state analysis,load forecasting and user behavior analysis.User side data obtained by advanced measurement device is an important way to mine users’ habits of electricity usage and research user types.And it also provides scientific basis for power demand response and personalized service.The current research on the analysis of electricity users behavior is not mature.This paper uses a variety of data mining algorithms for data extraction,electricity characteristic analysis and application research on user behavior from macro and micro perspective.On the one hand it enriches the method of user behavior analysis.On the other hand,it verifies the application value of user behavior analysis.The traditional power user classification is mainly for the convenience of business management needs or based on the user’s actual label to be divided.This often does not take into account the user’s own real electricity features,and in the study of electricity law is still limited to the analysis of electricity load data,the lack of consideration of more factors.For this reason,this paper study commercial electricity consumption data sets and residential electricity consumption data sets.First,a semi-supervised classification model is set up for large power users.The k-Means algorithm is used to cluster users’ daily power curves,and the clustering result is used as the user’s category label.And then using the user electricity curve and category label as the training data to establish the classifier model,so as to realize the recognition classification of new users.Finally,for residents,the association analysis method is used to excavate the association rules of household appliances,which is introduced into the load forecasting model as a reference factor.Short-term load forecasting is carried out by using LSTM(Long-Short Term Memory)model.The experimental results show that compared with the traditional method,the proposed method significantly improves the prediction accuracy of 30.23%of user-level load data and achieves better experimental results.