Abstract Rapid growth of electric bikes demands fast-charging systems that are efficient, stable, and capable of operating under varying load and voltage conditions. However, existing multi-level DC-DC converters and control strategies often suffer from high current ripple, voltage imbalance, and limited dynamic performance, restricting their suitability for large-scale fast-charging deployment. In this research, optimized deep learning-based multilevel DC-DC converters has been proposed for fast charging of electric bikes. The proposed system employs Harmony Search Algorithm (HSA) for tuning PI controller to finds the best PI gains globally which minimizes the steady state error. However, PI controller is inherently linear, it cannot fully handle the nonlinear and dynamic behaviour of a multilevel DC–DC converter. To address this limitation, Levenberg–Marquardt (LM) based Deep Neural Network (DNN) is used to learn nonlinear relationship between the instantaneous voltage error, the change in voltage error, the PI control signal, actual load voltage and generates optimal modulation index while minimizing the mean square error. The proposed method outperforms existing converters in terms of voltage ripple, charging time, and efficiency, according to simulation in MATLAB/Simulink and experimental validation. The proposed DNN-HSA converter consistently achieves 2.5%, 6.7%, 10.9% and 14.8% higher efficiency compared to existing PI-MPC, MPC, VP-MPC, and multi-level DC/DC converters.
Marimuthu et al. (Sun,) studied this question.