The Research On Substation Load Classification And Comprehensive Load Modeling

Load modeling is one of the most difficult problems in the field of power systemanalysis. Composite load model has a crucial impact in digital simulation of powersystems, so as to affect the power design, planning and power system operation.Overly conservative analysis result will bring unnecessary waste of resources, whiletoo optimistic analysis will take a risk to power system safety and stable operation.The paper sum up some existing load modeling theory and method, then asubstation load characteristics classification method is proposed based on fuzzyC-Means clustering algorithm(FCM) combined with mean shift algorithm(MS) inorder to prevent FCM from local optimization originating in its sensitivity to theinitial control parameters. Mean-shift algorithm is used to search for the mostintensive area of sample points in the feature space so as to obtain the number ofclusters and cluster centers for the initial parameters of fuzzy C-means algorithm. Themethod is applied to load characteristics clustering of220kV substation in Hunan.The example shows that the proposed method is less dependence on the initial controlparameters and the classification result is more reasonable than fuzzy C-meansalgorithm.Lack of data sources is an important reason for restricting the development ofload modeling. The paper analysis some data sources of load modeling, point out thatfault recorder data is feasibility and data pre-processing is necessity using in loadmodeling. And carry out a detailed analysis for fault recorder data preprocessingprocess and associated algorithm, then obtain the valuable data for load modeling bydata filtering, standardization of frequency, harmonic processing and the extraction offundamental positive sequence component.With the dynamic characteristics of power system load more and more complex,this paper put forward a generalized dynamic fuzzy neural network(GD-FNN) loadmodel suitable to describe the electric power load dynamic characteristics. Thelearning algorithm take system error and fuzzy-completeness as a sure standard toadjust parameters, the algorithm could make a comment on the fuzzy rules and theimportance of input variables, ensuring that the input variable width of every rule willbe auto-adapted adjustment according to the contribution to the system, then thesynchronous identification of load model structure and parameters could be achieved. The simulation from a substation measurement data showed that the generalizeddynamic fuzzy neural network load model have a better fitting degree and a stronglygeneralization capability, so the actual dynamic load characteristics can be describedcorrectly.