Modeling and Simulation-Based Evaluation of a Flyback Converter Using Classical and Learning-Based Control Methods Under Avionics-Oriented Power Conditions

This paper compares a proportional-integral-derivative (PID) controller, an adaptive artificial neural network (ANN) controller, and a compact deep reinforcement learning (DRL) controller for a flyback-derived 28 V to 12 V converter. An averaged model and an approximate small-signal duty-to-output transfer function are used for controller design and interpretation of the results. The methods are evaluated under startup, line-voltage disturbance, load-step, and light-load conditions near the continuous-conduction-mode/discontinuous-conduction-mode (CCM/DCM) boundary using overshoot, settling time, steady-state error, output ripple, and integral of absolute error.