Model-Free Predictive Control Based on Data-Driven for Fast and Accurate Power Allocation in Solid-State DC Transformers

Robustness is a crucial issue in model predictive control despite the excellent dynamic performance and effective management of multiple variables that this control strategy provides for converters. This challenge is particularly concerning in solid-state dc transformers (DCT) connected as input-series-output-parallel modules because this type of DCT not only regulates the output voltage/power but also balances the input voltage so that the power is averagely distributed among submodules. Static errors and limited dynamics pose a threat to their operation and even safety. To tackle this issue, this article proposes a novel model-free predictive control based on an autoregressive moving average structure. Herein, the physical model of DCTs is completely replaced by a data-driven model. Combined with the recursive least square method with the forgetting factor, the control system will no longer rely on the electrical parameters, guaranteeing strong robustness. Meanwhile, an adaptive balance controller is also proposed, which adjusts the gain of shift angles with reduced computational burden. These achieve accurate power/voltage allocation among all submodules while maintaining a fast dynamic response. The experimental comparisons with other schemes verify the effectiveness and superiority of the proposed method.