Research On Spatial Load Clustering And Forecasting Method Of Distribution Network

Spatial load forecasting is the premise and basis of power distribution network planning.The load forecasting precision not only affects the investment and running of power grid,but also affects the rationality of urban planning.Based on the serious review and summary of domestic and foreign advanced theory and research method,on the basis of the big data era background,and with the integrated use of cluster analysis,neural network and intelligent optimization technology,the main research was carried out by the following three aspects to improve the precision of spatial load forecasting,practicability and applicability.First,the load classification model research based on daily load curves.A new method,Pattern Index Clustering(PIC),originated from clustering techniques in data mining,is proposed for daily load curve clustering in this paper.PIC method reduces dimensions of sequential load curves with six load pattern indexes,such as load factor and daily peak-valley ratio.Weights of the indexes are gradually corrected with clustering-validity-based Delphi method.Hence,the method clusters curves with weighted Euclidean distance as similarity measurement,resulting in a performance better than classic clustering algorithms.Numerical examples show that PIC method has shorter run time,stronger robustness and better clustered load curves.Therefore it reflects characteristics of typical load curves more directly.Second,the spatial load clustering and integrated forecasting method of distribution network considering regional difference.In view of the problem that the load density method based on intelligent algorithm has strong dependence on the sample and difficulty in practical application,a spatial load clustering and integrated forecasting method of distribution network considering regional difference is proposed.Firstly,power users and regional information of different cities and different types is obtained through mass investigation.Secondly,on the analysis of the deficiency of current load classification mode,a method based on typical daily load curves is put forward to check the sample load classification label and select the proper samples.Thirdly,a full sample space is constructed by the selected samples,which is further classified into hierarchical sub sample space according to a two-level partition of regional property and checked classification label.Finally,a SVM forecasting mode is built with the most similar sub sample space to the plot to be predicted,thus the spatial load distribution is obtained.An actual case shows the practicality and effectiveness of the proposed method,which meets the accuracy requirements.Third,the spatial load distribution based on clustering analysis and nonparametric kernel density estimation.The existing research focuses on the theoretical innovation and precision of the forecasting method,namely neglects the research on the load distribution.On the basis of clustering analysis and nonparametric kernel density estimation,a method of spatial load distribution is proposed.Taking Zhejiang power grid as an example,spatial load research samples are classified according to a two-level partition of urban development types and land use types,then the typical load density distribution characteristics are extracted in all kinds of samples with the proposed method.Finally,industrial,commercial,residential and other different kinds of spatial load distribution are analyzed,which provide reliable support for distribution network planning.