Multiobjective Optimization Considering PET's Vibration Suppression of Dual Active Bridge Converter Based on BP-NSGA-II

The article proposes a multiobjective optimization procedure for a dual active bridge (DAB) converter based on nondominated sorting genetic algorithm with back propagation (BP) neural network embedded in back propagation nondominated sorting genetic algorithm (BP-NSGA-II), where the BP neural network was used to predict the vibration of power electronic transformer (PET). Experimental results demonstrate the high accuracy of this neural network in vibration fitting of PET. Based on the offline trained neural network, the phase-shift angles corresponding to minimum vibration is found in the triple phase-shift mode of DAB. Compared with common phase-shift modes at the same transmission power level, the optimized method could effectively reduce the vibration amplitude of PET at the key frequency with more than 20 dB. Meanwhile, the vibration amplitude at the concerned frequency of the optimized method is smaller than that of all the training sets. Furthermore, in BP-NSGA-II, the current stress, reflow power, and vibration of PET are used as objective functions to optimize the operating state of DAB further. The experimental results show that the proposed multiobjective optimization method, based on BP-NSGA-II, can improve the performance of the DAB converter by reducing its current stress and reflow power while suppressing the vibration amplitude of PET.