CHART: Intelligent Sensing-Based Fault Diagnosis Method for EV Charging Infrastructure

With the rapid global deployment of electric vehicle (EV) charging infrastructure, ensuring reliable operation and enabling early fault diagnosis have become critical technical challenges. We propose the cost-sensitive hybrid attention recurrent transformer (CHART) model, whose innovations include: integrating a time-series feature extraction module with an attention mechanism module to achieve joint modeling of local temporal features and global contextual dependencies; designing a weighted adaptive cost-sensitive (WACS) loss function that accounts for real operational and maintenance (O&M) costs to effectively reduce the risk of high-cost misclassifications; and introducing Optuna-based Bayesian hyperparameter optimization for automated architecture and parameter search. Unlike existing approaches that typically focus on a single aspect, CHART leverages real multisensor intelligent sensing data from a multilayer monitoring system in dc charging modules to deliver end-to-end, cost-oriented fault diagnosis. Experimental validation on a real-world dc charging infrastructure dataset demonstrates substantial performance gains over traditional methods, with average accuracy, precision, recall, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${F}1$ </tex-math></inline-formula>-score, and AUC increased by 16.67%, 13.41%, 0.76%, 7.79%, and 1.27%, respectively. Ablation studies confirm the independent and synergistic contributions of each core module, showing that removing the time-series feature extraction module or the attention mechanism module results in accuracy drops of 10.83% and 43.33%, respectively. These results indicate that CHART offers outstanding effectiveness, robustness, and economic efficiency in intelligent sensing driven EV charging infrastructure fault diagnosis. This research provides an efficient, data-driven solution for intelligent sensing and reliable operation of EV charging infrastructure.