Research On Load Clustering Analysis And Load Management Optimization

With the development of society and economy,the load of electric power is increasing rapidly.In recent years,many areas of our country have experienced power gap in the peak of electricity consumption.Only by increasing the investment of electricity generation is not economic.By tapping the demand side resources to ease the peak electricity has drawn more and more attention.At present,the power market of our country is not well,the demand side management mainly adopts the extensive means of orderly power utilization,with little consideration of the load form and lower user satisfaction.In this paper,the clustering algorithm in data mining is used to analyze the load of power system.In order to overcome the disadvantages of the existing demand side management methods,this paper proposes a load management method based on the user load pattern classification and the interaction of demand side resources.This paper mainly has the following contributions:1.For now there has been no evaluation method can compare the clustering results of load normalization method,this paper proposes a clustering algorithm evaluation index based on random forest algorithm.Load prior indexes and load shape indexes are the input for training the random forest.The performance of clustering are evaluated by out of bag test.Then the appropriate normalization are selected.Compared with the traditional evaluation method,this method is not affected by the clustering normalization method,and can be used to compare the clustering results under different normalization methods.2.To make up for the drawback that clustering method based on Euclidean Distance considering all dimensions of load curves is weak in load shape similarity,a two-step clustering algorithm combined with load shape indexes for power load profiles is proposed.First,clustering result is obtained by clustering method based on the Euclidean Distance considering all dimensions of load curves and clustering method and number are selected using clustering evaluation index.Second,load is reclassified using supervised learning algorithm based on load shape indexes.The proposed two-step clustering algorithm can make up for the weakness of traditional clustering algorithm in load shape similarity.3.A comprehensive week-ahead decision model for orderly power utilization is proposed.Four peak shifting and averting strategies based on users’ load profile and users’ participation willingness as well as load distribution and transmission congestionare considered.Network loss is approximately considered through the improved DC power flow method.A multi-objective optimization model aimed at minimizing the cost of peak shifting and averting strategies and network loss is built.The solving time is decreased by some solving strategies,for example,only considering the peak time constraints.The proposed method can make personalized power utilization scheme based on users’ characteristics.Orderly power utilization in both spatial and temporal dimensions can realize the optimal allocation of limited power resources,ease power shortage as well as reduce network loss and eliminate transmission congestion so that the security and economical efficiency of power network operation can be improved.