Deep Learning-based Control of Multi-Port Solid State Transformer

This paper proposes a deep learning–based control strategy for a multi-port solid-state transformer (SST) to achieve accurate and efficient power flow regulation among its ports. A three-port SST architecture is developed along with its power electronic bridges. Then, the phase-shift angles between the bridges are systematically swept to span the full operating range of power transfer among the ports. The resulting dataset is used to train a deep neural network (DNN) that learns the nonlinear relationship between power flow and phase-shift modulation while taking into consideration different voltage levels, and passive parameters range. Once trained, the DNN directly maps the power reference signals to optimal phase-shift commands that minimize the tracking error between the measured and reference power. Moreover, the proposed approach eliminates the need for iterative optimization or analytical power flow models. This enables overcoming the challenges associated with the highly coupled and nonlinear dynamics of multi-port SSTs. In addition, the method significantly reduces the online computational burden while maintaining fast dynamic response and high control accuracy. Experimental results validate the effectiveness of the proposed deep learning–based controller under various operating conditions and power transfer scenarios.