Multi-parameter optimization and design of self-triggered low voltage hybrid DC circuit breaker based on machine learning

The low-voltage direct current (LVDC) hybrid circuit breaker (HCB), with advantages of low conduction loss and high breaking performance is better to meet the application needs of photovoltaic system. Among these, self-triggered hybrid circuit breakers (STHCBs) based on natural commutation have become an important development direction for LVDC HCBs due to their simple structure and selective protection. However, existing design methods do not adequately consider the influence of multiple parameters, resulting in unreliable current commutation during interruption. To address these issues, this paper proposes a machine learning-based multi-parameter optimization method. The method employed Long Short-Term Memory (LSTM) algorithm to predict the interruption waveforms of mechanical switches under different short-circuit fault conditions as an input to the optimization. In addition, an interruption model of HCBs was developed, which comprehensively considered power electronic device characteristics, arc behavior and drive circuit. Based on the theoretical analysis and predicted waveforms, we used Genetic Algorithm (GA) to determine the optimal design parameters for the corresponding interruption waveforms. Experiment results confirm that this method can effectively enhance the interruption reliability of STHCBs while also demonstrates adaptability. The proposed method provides technical reference for the design of LVDC HCBs.