Fast Amplitude Determination of Switching Overvoltage in Black-Start Plans Based on Gas Turbine Distributed Energy Supply System

Abstract

Gas turbine distributed energy supply system (DESS) is a kind of important black-start unit in the future power system. This paper proposes a method of using Support Vector Machine (SVM) model for fast amplitude determination of transmission line switching overvoltage in the black-start plans based on Gas turbine distributed energy supply system. Black-start is the last line of defense for ensuring the reliability of power system. Hence black-start plays an important role both in the process of system recovery to ensure system security. During the process of making black-start plans of power system, it is necessary to verify the rationality of some technical issues by repeated modeling and simulation of different black-start plans, thus costing a lot of manpower and time. In recent years, distributed integrated energy supply system is greatly supported by government because of high efficiency and less pollution. Especially, gas turbine integrated energy supply system has excellent self-start and flexible adjustment ability, which can be considered as suitable black start unit. In this paper, firstly, the black-start scenarios are classified by the function and the type of the black-start units. Secondly, transmission line switching overvoltage involved in the process of black-start are modeled through PSCAD/EMTDC simulation software and analyzed by a large number of simulations. Thirdly, a support vector machine (SVM) model is established for fast amplitude determination of overvoltage in a black-start scenario. In this model, the selection of characteristic inputs in SVM method is analyzed in detail under the influence of important technical problems and the features of Gas turbine distributed energy supply system, and then the characteristic inputs are selected by orthogonal decomposition method. In the study case, artificial neural network (ANN) and support vector machine method are used for comparison, 200 samples are used in training set and more than 1400 samples are used in testing set, the error analysis shows that the support vector machine method is more effective than the artificial neural network method in the case of small training sample size. At last, an actual example analysis which considered the Guangzhou Higher Education Mega Center distributed energy station as black-start unit shows that the fast amplitude determination of switching overvoltage model can effectively reduce manpower and time.