An electric vehicle charging pile fault diagnosis system using Borderline-SMOTE and LightGBM

Electric vehicle charging pile fault diagnosis (CPFD) technology has been rapidly developed and successfully employed in the field of electric vehicle charging piles. However, in real life, failure data is very difficult to be obtained, as a result, it will cause data samples to be imbalanced seriously and make CPFD more and more challenging. To solve this problem, a novel Borderline-SMOTE-based imbalance correction for CPFD is proposed in this paper. With regard to the imbalance correction, Borderline-SMOTE over-sampling technology is utilized to solve the problem of unbalanced samples. For CPFD implementation, the (Light Gradient Boosting Machine) LightGBM ensemble learning combined with a grid search cross-validation algorithm is designed to build a fault detection model. Related experiments have proven the proposed methods can achieve the highest diagnostic accuracy, which is superior to other popular methods.