Machine learning-based intelligent mode selection for adaptive-turns LLC resonant converters to maximize performance across wide voltage and load ranges

Purpose The purpose of this study is to develop an intelligent control strategy for operating mode selection in an LLC resonant DC–DC converter with an adaptive transformer turns ratio. By replacing conventional threshold-based mode switching with a machine learning–driven approach, the aim is to enhance efficiency, stability and adaptability under dynamic and uncertain operating conditions. Design/methodology/approach A two-mode LLC converter topology using magnetic flux manipulation for transformer turns-ratio modulation is investigated, enabling operation in common mode and differential mode. A supervised machine learning framework – using support vector machine (SVM) and random forest (RF) classifiers – is trained on combined simulation and experimental data sets. Input features include real-time parameters such as input voltage, load power, switching frequency and historical converter states. The trained classifiers replace the fixed hysteresis-based logic to enable adaptive, data-driven operating mode decisions. A 300-W experimental prototype is built to validate the proposed method. Findings Compared with conventional hysteresis-based mode selection, the proposed ML-driven controller achieves up to 1.6% higher average efficiency and more than 20% reduction in switching-frequency variation across the tested operating range. The two classifiers – SVM and RF – consistently maintain soft switching and stable regulation under dynamic load conditions. These results confirm that data-driven mode selection enhances both performance and robustness relative to traditional threshold-based methods. Originality/value To the best of the authors’ knowledge, this work is among the first to apply supervised machine learning for real-time mode selection in an LLC resonant converter with adaptive transformer turns ratio. The approach eliminates the need for manually tuned voltage thresholds and hysteresis windows, enabling robust performance under variable and uncertain conditions. The results contribute to the development of intelligent, self-optimizing power electronics systems and open new avenues for integrating data-driven control into high-frequency converter design.