Research On Power User Feature Extraction And Classification Recognition Based On Data Driven

Due to unauthorized changes in electricity consumption by power users and mistakes by employees of power supply enterprises,there are some abnormal classified users in marketing system of power supply enterprises,which affects the accurate implementation of electricity pricing policies.The traditional mode of power user census results in a large search range and low work efficiency.Utilizing big data analysis can effectively improve work quality and efficiency.In recent years,power supply enterprises developed digital transformation,built marketing business application systems and power consumption information acquisition systems,accumulated massive power consumption data of users,which laid a foundation for accurate identification of user classification.Therefore,the use of data-driven methods for feature extraction and classification recognition of power users is expected to fully explore the potential information in electricity consumption data through interdisciplinary intersection.The research results can provide strong support for electricity marketing and demand side response.The main research content and achievements are as follows:Firstly,in response to the current research on electricity anomaly detection and user classification,which is often based on user electricity data and lacks user marketing profile data,and the low efficiency of algorithm operation,a user profile anomaly mining model based on outlier algorithm is proposed.It can identify users who are far away from the group by setting thresholds of the number of nearest neighbors and the number of outliers in advance.This model combined the production and electricity consumption characteristics of various users,and extracts peak valley difference and average electricity consumption of working and non-working days from electricity consumption data to reduce data redundancy during algorithm operation and improve computational efficiency.Using the open dataset released by the Irish Energy Regulatory Commission,we conducted experiments and analysis showed that the model can accurately and quickly select users with abnormal classification.Furthermore,based on the k-means algorithm,we proposed an improved k-means algorithm to address the issue of setting the clustering center or cluster number parameters in advance,and to fine screen the outlier classification users after coarse screening.Based on measuring similarity methods,we added convergence conditions to improve k-means algorithm.After improving,the algorithm can adaptively determine the optimal number of clusters k,reduce algorithm iterations,and improve operational efficiency.The effectiveness of the improved k-means algorithm was verified using both artificial and real datasets,indicating that the improved k-means algorithm can efficiently and correctly classify and identify abnormal users.Finally,we verified the effectiveness of user classification recognition methods based on outlier detection algorithm and improved k-means algorithm by real datasets.It indicated that the classification strategy of coarse screening and fine screening proposed in this article can improve classification efficiency,which can be widely applied to user file management,electricity consumption inspection,electricity price execution inspection,and marketing inspection applications in various power supply enterprises.