Research On Detection Method Of Outlier Value Of Power Data Based On Fast Density Peak Clustering And LOF

With the construction of smart grid and the speed-up of electric power system informatization,electric power enterprises have accumulated huge amount of data,these data mainly come from five major links,such as the generation,transmission,change,distribution and power consumption of the power grid.The analysis and research of power data is of great significance to the digital and intelligent development of China's power grid.The main causes of outlier are missing attribute value,abnormal power consumption behavior,power equipment failure and so on.The accuracy of the data analysis will be greatly affected if the dirty data is not processed in advance before the analysis of the electric power data.At the same time,outlier detection can find the abnormal data and hidden information,which has an important reference significance for the safe operation of power grid.Through the method of data driving,the outlier detection of power data can be realized,the abnormal state of electric energy can be monitored,the reaction speed of detecting abnormal phenomena can be accelerated,the stability of power grid can be improved,and the operation cost can be reduced,reduce the economic losses of power grid enterprises at the same time.Outlier detection of electric power data is the basic work of power data research.At present,the traditional power data analysis method cannot deal with such complex and large volume of data.It is necessary to find a method to deal with the current power system data.This dissertation compares the advantages and disadvantages of K-Means,DBSCAN,and Fast Density Peak Clustering algorithms.The Fast Density Peak Clustering algorithm has obvious advantages in the processing of power data,so this dissertation chooses this algorithm to study the power data.In this dissertation,a Fast Peak Density Clustering algorithm based on LOF is proposed.In view of the local characteristics of the Fast Peak Density Clustering algorithm for outlier detection and its strong dependence on truncation distance,the relative density and relative distance are redefined by using the idea of Local Outlier Factor.The rule of determining outliers is given at the same time.The improved algorithm takes full account of the data characteristics of power data and reduces the dependence on truncation distance.Based on the simulation of real power data,the effectiveness of the algorithm is verified.The results show that the algorithm can better describe the data characteristics of power data,and make the characteristics of outliers and clustering centers more obvious.Load curve clustering analysis plays an important role in power distribution in power grid companies.Accurate and rapid clustering of load data can speed up the efficiency of power data analysis,which is load forecasting and time-sharing pricing.The basis of research work such as user power consumption behavior analysis In this dissertation,on the basis of improving the definition of relative density and relative distance in the Fast Density Peak Algorithm by using the idea of LOF,this dissertation aims at the deficiency that the original algorithm depends on the possible clustering center in artificial identification decision-making graph.A normalized product based on relative density and relative distance is used to distinguish the critical points of cluster centers,so as to achieve the purpose of automatic selection of cluster centers.Experimental results show that the proposed algorithm is practical and effective.