Research On Electricity Consumption Behaviors Based On Unsupervised Learning Methods

With the popular use of smart meters,it is more efficient and reliable to collect constomers' power consumption data.The data mining of constomers' power consumption data and the analysis of the power consumption behaviors behind the data guide the power supply companies to provide differentiated power services to different constomers more pertinently,and enable power supply companies to achieve greater value from massive amounts of data.To this end,this paper attempts to analyze the constomers' power consumption behaviors from two aspects,one is the anomaly detection of the constomers' power consumption behaviors,and the other is the clustering analysis of the constomers' power consumption behaviors.In the aspect of anomaly detection of power consumption behaviors,the traditional adjacency-based anomaly detection method only determines the abnormal degree of samples from Euclidean distance.This paper proposes a novel detection algorithm based on local matrix reconstruction(LMR).This method calculates the abnormal degree of samples from two angles of the Euclidean distance between the samples and the data distribution of samples.Firstly,the daily load characteristics of each power customer are extracted to characterize their typical power consumption behaviors.Then,the anomaly detection algorithm based on local matrix reconstruction proposed in this paper is used to calculate the anomaly degree of each customer's power consumption behaviors.Finally,different threshold values are set to classify power customers into two categories: abnormal class and normal class.In this paper,the performance of the algorithm is analyzed by using the Receiver Operating Characteristic curve(ROC curve)and the Precision-Recall curve(PR curve),and compared with the traditional adjacency-based anomaly detection method.The experimental results show that the proposed method has higher detection accuracy and parameter stability.Aiming at the problems of data loss and misclassification caused by not considering the longitudinal randomness of load,a novel consumption behavior clustering method based on Earth Mover's Distance(EMD)is proposed,which takes into account both the transverse and longitudinal differences of power customers' behaviors.This method collects the load distribution of a power customer at the same moment for a few days,and characterizes the customer's power consumption behavior from both transverse and longitudinal perspectives.Finally,a group of household electricity load as a benchmark example is used to analyze the proposed method.The results show that in the case of the customers with similar transverse characteristics,the proposed method can capture the longitudinal characteristics of the users.From both the quantity and quality perspective,this method is more accurate and more feasible for clustering the customers' load.