Adaptive Ripple Reduction in DC-DC Boost-Buck Converters Using Decision Trees

Voltage and current ripple in DC-DC boost-buck converters (DCDC-BBC) can significantly impact the performance, efficiency, and lifespan of electric vehicle (EV) power systems. High ripple levels cause power losses, increased thermal stress, and instability, making effective ripple suppression crucial for reliable operation. This paper proposes a machine learning (ML)-based adaptive LC filter selection method to dynamically reduce ripple under varying EV load conditions. A decision tree algorithm, trained on experimental data, predicts the optimal LC values based on real-time ripple measurements. The system continuously adjusts LC parameters to suppress fluctuations during acceleration, deceleration, and cruising. Simulink simulations show a ripple reduction of 87.5% in the boost converter and 95.33% in the buck converter, while hardware validation confirms improvements of 95.45% and 94.51% in boost and buck converters, respectively. Additionally, efficiency improvements of 5.6% to 7.7% were observed, demonstrating better power conversion and system stability. These results confirm that ML-based adaptive filtering outperforms conventional fixed LC filters, making it a promising solution for improving DCDC-BBC efficiency and stability in EV applications.