Control Strategy for DC to DC Converter using MLP embedded Actor-Critic Architecture

Buck-boost DC to DC converters find a huge application in the industry for acquiring a constant DC output voltage. Due to the instability issues caused by the constant power loads, there has always been a requirement for adaptive control strategies that shall control and hence stabilize the voltage. In this paper, we have proposed an enhanced multilayered perceptron embedded in the actor-critic architecture to control the voltage variations. A PPO-based tuner consisting of the MLP has been used which produces coefficients that can be fed directly to our closed-loop feedback controller for the self-adaptive control system. The input and output sets have been analyzed along with appropriate hyperparameter tuning of the neural network within. Considering the seasonality of the data distribution of the input voltage, an appropriate training method has been used, ensuring the optimal real-time performance of the controller.