Research On Power Load Pattern Recognition Method Based On Density Peaks Clustering Algorithm

The widespread application of smart meters has accumulated massive multi-source heterogeneous load data on the user side.In-depth analysis of power load characteristics,through unsupervised clustering algorithm to mine the hidden correlation feature information between load data,can effectively identify load power consumption patterns,master users’electricity consumption habits and energy consumption laws,realize refined user classification,and provide load forecasting,Specific application scenarios such as power consumption abnormality detection and demand response provide key information support to improve the reliability and economy of power grid operation.Density Peak Clustering(DPC)is a new type of unsupervised learning method.The clustering principle is simple,and it can find clusters of any shape,effectively overcoming the limitations of some traditional clustering algorithms.At the same time,the DPC algorithm also has some shortcomings,such as the difficulty in selecting the truncation distance dc,and the poor fault tolerance of the one-step allocation strategy.Aiming at massive high-dimensional and complex user load data,in order to improve the performance of the load clustering algorithm and realize the accurate identification of load power consumption patterns,this paper proposes a corresponding improvement scheme based on the existing density peak clustering algorithm.The main work is as follows:(1)In view of the abnormal data existing in the user’s daily load curve,this paper filters out the noise value data through Gaussian smoothing filtering method,and uses the cubic spline interpolation method to fill in the missing value data;in order to eliminate the dimensional difference of load energy consumption among different users,The load data is normalized,and the PCA dimensionality reduction process is performed on the high-dimensional complex load curve.(2)Aiming at the shortcomings of the original DPC algorithm,a density peak clustering algorithm(TSNN-DPC)based on a two-step assignment strategy and shared neighbor similarity is proposed.The improved algorithm introduces the basic idea of shared neighbors,fully considers the real distribution information and local structure information of the sample,and optimizes and improves the calculation of the local density p and relative distance δ.Two-step allocation strategy for possible slave points.Finally,the simulation verification is carried out in three application scenarios of UCI real data set,artificial data set and power grid measured data set.The results show that the proposed algorithm can effectively improve the lack of clustering performance on variable density and multi-scale data sets,and has strong robustness.(3)Based on the TSNN-DPC algorithm,a typical load pattern recognition and abnormal power consumption detection model is proposed.Firstly,the improved algorithm is used for cluster analysis of the load data set to be measured;then,the cluster centers and abnormal points are obtained based on the γ-diagram and ρ-δ decision diagram of the sample data set,respectively.Extract the electricity load pattern,and complete the abnormal electricity consumption pattern detection.Finally,a simulation example is constructed using 16 different types of commercial electricity consumption data obtained from the US Open Energy Information website(OpenEI).