Research On Consumer Electricity Behavior Analysis Based On Data Mining

As the goal of building a new generation of power system is put forward,improving the comprehensive service capability of the power grid is a very important part of the process of realizing this goal.The improvement of comprehensive service capability requires that the power grid terminal can analyze the power consumption behavior of power grid users in a timely manner,find out the problems existing at the user terminal,mine the needs of the user terminal and provide corresponding services.With the construction of smart grids and the popularization of smart meters,the data collection speed has been improved and the data quality has become more reliable.As an advanced technical means based on data,data mining can use the massive data collected by smart meters to efficiently,quickly and accurately dig out the hidden behavioral rules in the consumer electricity behavior,so it is very necessary to carry out the research of consumer electricity behavior analysis based on data mining.At present,the research direction of consumer electricity behavior analysis is mainly focused on abnormal power consumption detection and clustering of power consumption behavior.Anomaly detection of power consumers is conducive to discovering electricity stealing behaviors and reducing economic losses.After screening out abnormal electricity users,clustering the behaviors of normal electricity users can facilitate the power grid to provide personalized services to different users.However,traditional power user anomaly detection requires manual extraction of data features,which is difficult and inefficient,and is suitable for detecting a large amount of power data with a long time span,which cannot detect abnormal power users in a short time.In addition,the consumer electricity behavior is not static,and the traditional electricity consumption behavior clustering method cannot identify the emerging types of electricity consumption behaviors.Therefore,the paper develops the following contents:1.For anomaly detection of a large number of power data,firstly,clean the original power data and fill in the missing data,then use nonlinear factors of deep learning to automatically extract data features.In this paper,an abnormal power detection algorithm based on convolutional neural network(CNN)is designed,and the effects of the double Dropout strategy,Batch Norm,and Mish activation function on anomaly detection are studied.The real electricity consumption data are used to evaluate the abnormal detection effect,and the accuracy,precision rate,recall rate and F1 score values are selected as evaluation indexes to compare with the abnormal detection results of Alex Net,support vector machine and decision tree algorithm.The experimental results show that the proposed algorithm has advantages in anomaly detection.2.For anomaly detection of a small amount of power data,the Cut Mix method is selected to reasonably amplify the original power consumption data,and then the data features,extracted through the CNN,are classified by the model fusion stacking method.The stacking method adopts a hierarchical structure.In the first layer,there are three primary learners that need to be optimized for parameters respectively,including random forest,gradient boosting decision tree and extreme gradient boosting.In the second layer,random forest is the secondary learner,and the original features of the data are also added to the input of the second layer as a guide for optimization of the fusion model.The anomaly detection is carried out using the real power consumption data,and the ROC curve,AUC value and P-R curve are used as evaluation indicators.The experimental results show that the model fusion algorithm has a good anomaly detection effect.3.In order to realize the incremental clustering of electricity consumer behavior types,a method based on BIRCH algorithm is used to conduct incremental clustering research.Firstly,in the original clustering stage,the user electricity consumption data is extracted through the CNN to obtain feature clusters,and then in the incremental clustering stage,the CNN feature extraction is performed on the user electricity consumption data of the new electricity consumption type.Finally,after integrating the clustering feature clusters extracted from the original clustering stage and the incremental clustering stage,a variety of sample sets are used for verification,and the DBI and purity values are used as evaluation indicators to test the effect of the incremental clustering algorithm.The experimental results show that,the method can identify and cluster the new electricity consumption types,and improve the accuracy of the overall clustering of consumption behaviors.