Optimal Design Methods Of Filter Parameters Of Full-Bridge Current Doubler Rectifier Converter

DC/DC converter is widely used in aircraft power system,renewable energy system,automobile and other fields.It is an important part of electrical system.The optimal design of the converter is helpful to solve the problems of large error and less consideration in the traditional empirical method,and achieve the overall design goal.The DC / DC converter studied in this paper is a ful l bridge current doubler rectifier converter.The modeling,design and multi-objective optimization of the filter are studied.The objective of optimization is quality and power loss,and the constraints include ripple requirement,temperature,RMS current and so on.Firstly,based on the topology of full bridge current doubler rectifier and phase shifted full bridge ZVS soft switching control mode,the circuit modes are analyzed,and the soft switching conditions and duty cycle loss are analyzed.On this b asis,the circuit model is established.Building component model and planning and design process are the basis of optimization,which can improve the accuracy of modeling and make the optimization results more accurate.Then,according to the manufacturer’s component manual,the data sheet of material and model of inductance core and the alternative data sheet of capacitance are established to select the minimum model that meets the limited conditions in the design;At the same time,the internal resistance of inductor and capacitor is calculated iteratively in the filter design process combined with circuit simulation to solve the problem that the internal resistance of filter in circuit simulation is not unified with that in design.Then,in order to achieve multi-objective optimization,this paper uses machine learning method to train the support vector machine and artificial neural network model,which are used to judge the feasibility of design parameters,and replace the mapping relationship between design variables and the two optimization objectives of quality and loss.With the trained model,a large number of parameters of feasible design schemes can be obtained quickly,so as to find the optimal filter design scheme.Finally,the optimization method based on machine learning is compared with the traditional optimization method based on genetic algorithm(GA).The results of the two optimization methods are consistent,which shows the feasibility of the proposed method.At the same time,the total tim e of obtaining training data,realizing model training,predicting target value and optimizing in machine learning method is one tenth of that of genetic algorithm,which shows the advantages of this method in short implementation time and less computing r esources.Finally,the experimental results show that the optimized filter parameters have the advantages of lower quality and loss.