Research On Parameter Identifiability And Identification Method Of Synchronous Generator Excitation System

Computational modeling is an essential tool for operation planning and stability analysis in power system.Therefore,studying the operation characteristics of large grid,especially the dynamic performance under the disturbance always simulate by modeling in most cases.Among those models,the accuracy of generator excitation system parameters is crucial for obtaining the realiable results when the structure is determined.These parameter values can be obtained by parameter identification.Choosing the appropriate algorithms can promote the efficiency and precision of estimates.Frequently,considering the models limitations,some parameters of models are often non-identifiable.Aiming at that the identification results of excitation system parameters are often unstable because partial parameters are associated under limited measurement,we construct the matrix based on trajectory sensitivity method,then the singular value decomposition approach for subsets selection is adopted to divide the excitation system parameters into associated set and non-associated.By calculating the discriminant coefficient,the associated parameters are allocated in several independent sets further,and then by means of assigning typical value to representative parameters,the dependency can be removed.Finally,the improve genetic algorithm is used to estimate the parameters due to the nonlinearity of excitation system.In addition,taking account into the measure error in field test,the Bayesian theory is used to infer the joint posterior probability distribution of parameters,thus evaluating their practically identifiability.The prior knowledge of parameters can avoid the local optima.Since the excitation system is quite complicate,the likelihood functions are computationally intractable,a likelihood-free technique,approximate Bayesian computation(ABC)is employed here to bypass the likelihood,and the total joint posterior distribution over all parameters can be obtained efficiently based on the sequential Monte Carlo(SMC)approach.Finally,compared with the sensitivity-based method,test studies can reveal the reason of parameters that are practically unidentifiable,and prove the validity of this Bayesian method.