An ANN-Aided Parameter Design Method for CLLC-Type DAB Converters Considering Parameter Perturbation

The distributed nature of power electronic components parameters can affect the desired output voltage of the CLLC-type dual active bridge (DAB) converters, especially in mass production with limited budgets. To minimize inconsistency for CLLC-type DAB converters against manufacturing tolerance in large-scale applications, this article proposes a novel resonant component parameter design method based on artificial neural network (ANN). Moreover, an ANN-based data-driven model of the probability density function is first developed to portray the distribution of component parameters within the allowable tolerance range. Furthermore, to enhance data processing efficiency in the parametric design process, a batch-normalization method is proposed to convert the original dataset to the normalized one in batches automatically. The co-simulation method is implemented with Monte Carlo analysis by combining MATLAB with LTspice. To ensure the accuracy of the co-simulation method, experimental results for the limited parameter combinations are provided as the verification for the co-simulation method. Finally, Monte Carlo analysis is adopted to optimize the resonant components parameter with three quantitative evaluation indexes. The verification results show that the failure rate of the output voltage can be reduced to less than 5%.