Neural Network-based control optimization of a Quadruple Active Bridge with respect to conduction losses

This paper proposes a control optimization of a Quadruple Active Bridge (QAB) converter using a neural network. The QAB is a four-port converter capable of handling DC sources and loads at different voltages. Its central node is a transformer, which provides galvanic isolation between the ports. It will be shown that for a common switching frequency, there are seven degrees of freedom, making its optimal control particularly complex. This paper proposes a novel neural network-based approach to develop a controller which improves the performance on the basis of a non-simplified model of the transformer. The introduced optimization leads to significantly reduce conduction losses, enabling the converter to operate efficiently with variable voltages and powers over its entire operating range. The results are verified by simulation.