Ripple Minimization in Asymmetric Interleaved DC–DC Converters Using Neural Networks

Interleaving can be employed to reduce ripples in multiphase dc–dc converters although phases are operated under asymmetric conditions, such as different input voltages or loads. To allow ripple minimization under asymmetric conditions, phase shifts between the switch timings of phases have to be appropriately adjusted. This study presents a method based on artificial neural networks (ANNs) that can provide the required phase shifts to minimize ripple under asymmetric conditions. To obtain a machine learning dataset, the set of optimal phase-shift angles minimizing the common output capacitor current ripple is analytically obtained for two asymmetric interleaved boost converters. Then, a small-scale computationally efficient ANN is developed using the dataset to predict the optimal phase shift according to real-time operating conditions. The proposed method is experimentally validated. The proposed method predicts the optimal phase-shift angle in 28.4
$\mu \text{s}$
and the prediction is updated every 100
$\mu \text{s}$
to achieve ripple minimization.