DC/DC Power Converter Control-Based Deep Machine Learning Techniques: Real-Time Implementation

Abstract

The recent advances in power plants and energy resources have extended the applications of buck-boost converters in the context of dc microgrids (MGs). However, the implementation of such interface systems in the MG applications is seriously threatened with instability issues imposed by the constant power loads (CPLs). The objective is that without the accurate modeling information of a dc MG system, to develop a new adaptive control methodology for voltage stabilization of the dc–dc converters feeding CPLs with low ripples. To achieve this goal, in this letter, the deep reinforcement learning (DRL) technique with the Actor–Critic architecture is incorporated into an ultralocal model (ULM) control scheme to address the destabilization effect of the CPLs under the reference voltage variations. In the suggested control approach, the feedback controller gains of the ULM controller are considered as the adjustable controller coefficients, which will be adaptively designed by the DRL technique through online learning of its neural networks (NNs). It is proved that the suggested scheme will ensure the rigorous stability of the power electronic system, for simultaneous effects of CPL and reference voltage changes, by adaptively adjusting the ULM controller gains. To appraise the merits and usefulness of the suggested adaptive methodology, some dSPACE MicroLabBox outcomes on a real-time testbed of the dc–dc converter feeding a CPL are presented.