Intelligent Three-Phase-Shift Strategy Based on Normalized Models and Neural Networks for Dual Active Bridge Converters

Conventional three-phase shift (TPS) modulation strategies for dual-active bridge (DAB) converters rely on complex theoretical derivations and cumbersome calculations. To overcome these limitations, this article proposes an intelligent TPS control strategy that combines normalized modeling with feedforward neural network (FNN) learning. A harmonic model of the DAB converter is established, and analytical relationships are derived among the output power, the root-mean-square (RMS) current of the series inductor, and the modulation variables. Current and power normalization is implemented to decouple the model from hardware parameters and voltage levels. On this basis, a unified optimization model is constructed to minimize the normalized RMS current under output power constraints. The gradient descent algorithm is employed to solve the optimization model over a wide operating range, generating a normalized dataset of optimal TPS variables. An FNN model is then trained to establish an end-to-end nonlinear mapping from voltage gain and normalized power to the optimal modulation variables. Experimental results show that this intelligent TPS control strategy can be directly applied to different DAB converters while guaranteeing voltage stability and dynamic response performance.