Data-Driven Optimal Control of Dual-Active Bridge Converter Using ANN-Based Triple-Phase Shift Modulation

Dual-Active Bridge (DAB) converters are integral to efficient bidirectional power conversion in demanding applications such as renewable energy systems and electric vehicle (EV) charging. Optimizing their performance using advanced modulation strategies like Triple-Phase Shift (TPS) across wide operating ranges, however, remains challenging. This paper presents a data-driven approach employing Artificial Neural Networks (ANN) for real-time optimization of TPS modulation in DAB converters. High-fidelity synthetic datasets, generated from accurate SiC MOSFET simulation models, were utilized to train an ANN capable of predicting optimal operating parameters <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(D_{1}, D_{2})$</tex> that minimize the sum of conduction and switching losses. The proposed methodology achieves significant loss reduction, demonstrating up to 15% improvement compared to the conventional SPS modulation technique, while consistently maintaining Zero Voltage Switching (ZVS) conditions across diverse operating points. The trained ANN exhibits excellent predictive reliability, achieving test correlation coefficients of 0.997 for <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$D_{1}$</tex> and 0.998 for <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$D_{2}$</tex>. Furthermore, the highly compact ANN architecture, comprising only 354 trainable parameters, ensures computational efficiency, rendering it suitable for real-time implementation on resource-constrained Digital Signal Processing (DSP) platforms.