Research On Application Of Machine Learning In Electric Consumption Analysis Of Customers

With the deepening of research of artificial intelligence in China,machine learning algorithms are rapidly developing as one of important research contents.Among them,the anomaly point detection algorithms are an important research branch in machine learning algorithms.Based on the actual customer electric consumption data of the power company,this thesis analyzes the customer’s behavior about electric consumption by using the anomaly point detection algorithms,and achieves an aim which anomaly electric behavior can be efficiently detected.The main research work is as follows:a.Summarize the relevant work experience of electric power workers,and propose a set of evaluation index system of abnormal electric consumption,and apply it to the feature construction stage of data preprocessing using the original,and verify the practical application value of the system.b.The six types of anomaly detection algorithms are collected,and three kinds of algorithms with broad application prospects are selected.The performance of the algorithm is analyzed and compared,and the IF algorithm is selected because of its better performance.Research experiments show that the IF algorithm still has the disadvantage of unable to identify local anomalies.Therefore,the CBIF algorithm is proposed based isolation forest algorithm,which combined with the K-Means clustering algorithm,and can effectively recognize anomaly points and uniformly convert the anomaly points into global anomaly points,thus achieved an aim which can improve algorithm accuracy.c.Using the CBIF algorithm to design the customer’s anomaly electric utilization detection model,and apply it to the real customer electric utilization data to identify the anomaly customers which had electric utilization behavior.Experiments show that the model can identify anomaly electric customers more efficiently and accurately,and can provide reference for this algorithm applies other fields.