Optimal Modulation of Three-Phase Dual Active Bridge using Multidimensional Ripple Correlation and Artificial Neural Networks

The 3-phase Dual Active Bridge, DAB, converter has many advantages over the traditional single-phase DAB, such as higher efficiency and better current distribution. To achieve an optimal modulation, it is often necessary to modify the phase shift, as well as the duty cycles of the primary and secondary bridges. Due to the high complexity of the currents in the 3-phase Medium Frequency Transformer, MFT, and their complex relationship with the duty cycles it is necessary to utilize advanced optimization techniques. In this work it is proposed to employ Artificial Neural Network, ANN, trained from data generated using the Multidimensional Ripple Correlation Technique, RCC. Using ANN, instead the RCC, accelerates the system dynamics 10 times (from 10 ms in the case of RCC to less than 1 ms in the case of the proposed ANN), the RMS current compared to the traditional Single Phase Shift modulation, SPS, has decreased approximately 20% and this is reflected in a 30-40% reduction in winding and semiconductor losses for our application. The use of ANN ensures that the optimal modulation is obtained at each point, thanks to its ability to generalize, without demanding high computational cost and without introducing slow dynamics, improving one of the main problems of the RCC technique. Thanks to ANN offline training process, it is possible to obtain an optimal modulator offline (only using simulations) without the need to program complex control algorithms such as RCC in the converter controller.