Robust Artificial Intelligence Controller for Stabilization of Full-Bridge Converters Feeding Constant Power Loads

The dc/dc full-bridge (FB) converters are often utilized as the interface system in telecom power applications. The constant power loads (CPLs) in the telecom power systems demonstrate negative impedance which threatens the stability of the dc/dc converters. To address this issue, in this brief, a nonlinear controller is designed for the stabilization of the dc/dc full-bridge converter feeding CPLs. The soft actor-critic (SAC) algorithm with deep neural networks is adopted based on deep reinforcement learning (DRL) for optimal tuning of the controller parameters in the control law of an established nonlinear controller. According to the control requirements of the FB interface system, a reward signal is defined to train the neural networks of SAC. An efficient solution based on Hardware-in-the-Loop (HIL) is adopted for verifying and validating the feasibility of the proposed scheme using the OPAL-RT 5600.

Automated summary

This paper proposes a nonlinear controller for stabilizing the DC/DC full-bridge converter feeding negative impedance loads and adopts the soft actor-critic algorithm based on deep reinforcement learning for parameter tuning. A reward signal is defined to train neural networks. The feasibility of the proposed scheme is verified using a Hardware-in-the-Loop approach.