Quantifying and comparing the effects of human and organizational factors in electric maloperation accidents with HFACS–CatBoost and SHAP
Chuan Lin, Qifeng Xu, Yifan Huang
Abstract
Abstract The proportion of electric maloperation accidents (EMAs) in substations caused by human and organizational factors (HOFs) has gradually increased. Although there has been some research into the factors affecting EMAs in substations, the available results are insufficient to support the interpretation of HOFs in EMAs. This article explores the relationships between the HOFs and EMAs using Human Factors Analysis and Classification System‐gradient boosting with categorical features support (HFACS–CatBoost) and Shapley Additive exPlanation (SHAP) methods. First, the HFACS framework was introduced to identify 135 EMAs in the Southern Power Grid risk causation. CatBoost was used to construct an accident classification model to analyze the important relationship between accidents and HOFs and to compare and analyze with the extreme gradient boosting (XGBoost) and the binary logistic regression (BLR) to verify the superiority of CatBoost. Finally, to solve the problem of inadequate interpretation of the CatBoost black‐box model, the SHAP value plot was applied to express the contribution degree relationship between accidents and HOFs. The results show that the above method can explore and explain the importance and contribution of HOFs in EMAs. And from this, it is concluded that poor psychological state, poor communication and coordination, inadequate supervision, and inadequate training and education are highly correlated with the occurrence of EMAs. The findings will help substation operations and maintenance staff to develop safety measures to address the confusion of HOFs in substations and prevent the occurrence of EMAs.