Learning-Based Optimal Large-Signal Stabilization for DC/DC Boost Converters Feeding CPLs via Deep Reinforcement Learning

For the dc/dc boost converter feeding constant power loads (CPLs), improving its dynamic characteristics and guaranteeing the output voltage stability have aroused much attention from the power electronics society in recent years. This article proposes an optimal output regulation strategy by fusing a robust stabilization strategy with deep reinforcement learning (DRL). First, by employing the higher-order sliding mode observer (HOSMO) to estimate the uncertainties of the system, then we are able to realize a fast performance recovery ability via feedforward compensation loops. Second, the deep deterministic policy gradient (DDPG) is trained to adaptively adjust the control coefficients. Third, the stability analysis is given to guarantee the large-signal stability of the control system. In order to evaluate the strength and effectiveness of the proposed adaptive method, experiments are conducted in this article on a dc microgrid platform.