Research On Key Technology Of Distribution Network State Estimation Based On Load Classification

With the rapid development of the modern economic society, as well as the increasing scale and more complex structure of power system, both safety and economy have to be achieved when power system running, which requires the power grid dispatcher to keep a clear understanding of the current running risk level of the power grid in order to deal with the emergencies that may occur in the grid. Power system state estimation is the basis of advanced application of Energy Management System (EMS)and Distribution Management System (DMS), while high quality state estimation is an important technical support for reliable running of power grid, and is also the basis of other advanced application analysis software. Due to the existence of three-phase unbalance in distribution network and the features of usually a radially oriented network structure, great system dimension, higher R/X ratio and less measurement value, the state estimation algorithm of transmission network usually cannot be directly used in distribution network. In view of the above problems, based on the traditional state estimation algorithm, this paper studies the state estimation according to the characteristics of less real-time measurement of distribution network.The main contents of this paper are as follows:Aiming at the problem that single phase model of transmission network cannot be directly applied to distribution network, the main equipments in distribution system,including lines, transformers, loads, capacitors, switches and gates, as well as different connection modes of the equipment are modeled for three-phase in detail. On this basis,the topology analysis method of radial distribution network is given by the node injection model.In order to solve the problem of less measurement information and discontinuous load data of distribution network, a new method for generating pseudo measurement of distribution network based on nonparametric estimation is proposed in this paper to supplement the measurement data. In this method, the weighted Nadaraya-Watson kernel estimation method is used to analyze the existing load data and obtain the probability distribution level, and an effective historical prediction curve is generated through the regression equivalence model. An efficient second order differential Stirling isometric interpolation method is adopted to excavate the load value at different times to solve the problem of discontinuity of pseudo-measurement data. This method is characterized by good prediction accuracy as that the accuracy of the pseudo measurement load of the non-measuring points can reach or close to the actual measured values, and the continuity, smoothness and timeliness of the load data can be maintained, thereby improving the accuracy of the state estimation of the distribution network.In view of the fact that the current distribution network has the problems such as lack of real-time data and even no measurement in partial region, which resulted in convergence problems and poor accuracy of the state estimation, a state estimation (SE)method for power distribution network, based on a load feature extraction and K-means clustering technology of Cuckoo's search algorithm, is presented to conduct a typical classification of different types of historical load data. The load characteristics are extracted by using an improved cuckoo algorithm (adaptive step size), and the K mean clustering method is employed to classify the load data which has been defined by the load characteristics in order to generate the pseudo measurement data of the area with similar load but not enough measurement. At the same time, according to the problems that the distribution network is unobservable and the computational complexity is increased because of the lack of measurement, a state estimation method based on network clipping is proposed, by using the equivalent cutting method of the power grid,the part without real time data measurement points in the power grid is set equivalent to the electrical Island. By minimizing the objective function, the computable equivalent network model can be used to reduce the computational complexity, reduce significantly the amount of calculation and improve the computational accuracy and performance of the state estimation.This paper presents a loop analysis method in order to solve the problem that traditional algorithm to deal with loop network is not easy to be convergent, discusses the road matrix and loop branch correlation matrix of the loop analysis method, and comes up with the power flow calculation method of distribution network based on the branch circuit by the analysis of the circuit equation of the loop analysis method. On the basis of this, the convergence conditions of power flow calculation for distribution network without grounding are discussed. And then for the calculation results based on the requirements of geographic information system (GIS) visualization application of high concurrency, a device oriented memory database, vector graphics services and client-side rendering technology is presented, and it is effectively applied to distribution network advanced decision analysis of real time GIS system.Finally, based on the research results of this paper, the intelligent decision support system of urban and rural distribution network and its field application are introduced.