Artificial neural network–based phase-shift optimization for input ripple minimization in an interleaved boost–Z-source converter

Abstract This paper presents a neural-network-based phase-shift control strategy for minimizing input current ripple in an asymmetric interleaved Boost–Z-source converter. Unlike conventional fixed-phase interleaving, which provides ripple reduction only at specific operating points, the proposed method adapts the interphase shift according to the converter’s instantaneous duty cycles. A comprehensive ripple-surface mapping was first carried out over a wide range of operating conditions, generating the optimal phase shift required to minimize the RMS of ripple for each duty-cycle combination. These optimal operating points were then used to train a compact 2–10–1 feedforward artificial neural network (ANN) capable of predicting the ripple-minimizing phase shift in real time. The trained network is deployed on a low-cost STM32 microcontroller and integrated into the converter’s PWM generation system. Hardware experiments performed on a full-scale prototype validate the simulation findings and confirm the method’s practicality and robustness. The proposed approach combines the accuracy of data-driven modeling with the simplicity required for embedded implementation, offering a strong alternative to analytical ripple-minimization techniques for asymmetric or hybrid power converter architectures. The method demonstrates fast dynamic response to duty-cycle variations and strong agreement between simulation and hardware results.