Self-learning Controller Design for DC-DC Power Converters with Enhanced Dynamic Performance

This article presents a promising self-learning-based robust control for output voltage tracking in DC-DC buck power converters, particularly for applications demanding high precision performance in face of large load uncertainties. The design involves a computationally simple online single hidden layer neural network, to rapidly estimate the unanticipated load changes and exogenous disturbances over a wide range. The controller is designed within a backstepping framework and utilizes the learnt uncertainty from the neural network for subsequent compensation, to eventually ensure an asymptotic stability of the tracking error dynamics. The results obtained feature a significant improvement of dynamic and steady-state performance concurrently for both output voltage and inductor current in contrast to other competent control strategies lately proposed in the literature for similar applications. Extensive numerical simulations and experimentation on a developed laboratory prototype are carried out to justify the practical applicability and feasibility of the proposed controller. Experimental results substantiate the claims of fast dynamic performance in terms of 94% reduction in the settling time, besides an accurate steady-state tracking for both output voltage and inductor current. Moreover, the close resemblance between computational and experimental results is noteworthy and unveils the immense potential of the proposed control system for technology transfer.