Research On User Classification Based On Power Load Adaptive Clustering

With the increasing development of smart grid technology,the widespread entry of smart meters into user-side has brought massive amounts of user data,providing the possibility to optimize the efficiency and economics of grid operation by relying on data mining technology in the new era.Today,when data technology is advancing by leaps and bounds,power grid problems relying on data analysis have also been developed quickly.Many practical application scenarios,such as demand-side management,personalized electricity prices,and load peak shaving,etc,all put high requirements on the refinement and accuracy of user classification.Therefore,it is necessary to carry out the clustering study on the load,and the following work is mainly done in this article:Firstly,the basic concepts and basic theories of clustering are introduced,and the existing load clustering methods are reviewed and summarized.It is found that the extraction of load characteristics has a decisive influence on the final clustering results,and the current methods generally exist two problems: the number of clusters is difficult to determine,and the clustering results may fall into local optimality.Therefore,the work of this paper aims to construct a set of algorithm framework that can effectively extract load characteristics and perform adaptive clustering.Then in the data standardization stage,the kernel smoothing method is used to smooth the original load data to eliminate local anomalies,and the smoothed data is extracted with discrete gear level features.For the obtained characteristic variables,it further combining the ideas of K-modes algorithm,K-means++ algorithm and hierarchical K-means algorithm to constructed a hierarchical K-modes algorithm with optimized initial conditions to overcome the traditional algorithms’ drawbacks:difficulty in determination of the number of clusters and local convergence.Finally,the experimental analysis on simulated data and actual data verifies the robustness and effectiveness of the proposed clustering algorithm framework,which proves the usability of the work in this paper in actual operation,and can guide the powerdepartment conducts more in-depth user-side strategy formulation by the user classification results achieved by the work in this paper.