Power Electronic Circuit Parameter Identification Based On Hybrid System Theory And KELM

With the development of technology,power electronic technology has been widely used in electric drives,smart grid,aerospace,ships and other areas which play crucial roles in national economic development and national security.Power electronic devices play the role of power conversion,motor drive and so on.The entire system cannot work properly with devices' failures.Fault diagnosis and prediction of power electronic circuits are of great significance which can improve reliability of power electronic equipments and reduce fault loss.Parameter identification is the basis of parametric fault diagnosis,and it provides theoretical basis and support for circuits' fault diagnosis.This paper mainly studies parameter identification methods of power electronic circuits,the research includes:(1)Failure modes and failure characteristic parameters of power electronic devices are analyzed and equivalent circuits of these devices are determined based on their characteristic parameters.(2)Parameter identification method based on hybrid sysytem model of power electronic circuits is studied.Taking the node voltage and the output voltage of the circuit as new state variables,the hybrid system model of power electronic circuit is established,and devices' characteristic parameters are identified using gravitational search algorithm.Taking the closed-loop BUCK circuit as an example,modeling,simulation and experimental analysis are carried out.Results show that this method can effectively identify characteristic parameter values of all the devices in the circuit and have a good identification performance.(3)Power electronic circuit parameter identification based on data driven is studied.With the help of kernel extreme learning machine,characteristics of circuit test signals are taken as the input of the kernel extreme learning machine model,and the corresponding parameter values can be identified,which is used as the output of the model.Taking the closed-loop BUCK circuit as an example,characteristic parameters are respectively identified by BP neural network,support vector regression,extreme learning machine and kernel extreme learning machine,and the performance of these algorithms are compared.Results show that proposed methods can identify the characteristic parameters of all the devices and the KELM has best performance and generality.