LightGBM-SHAP-Based Quality Tracing and Prediction of Electrical Equipment

Electrical equipment is a crucial infrastructure in the power system, susceptible to quality issues, and posing threats to the power system. Tracing quality problem causes helps prevention while predicting quality status based on these causes enables swift action. Most studies target quality problem identification, neglecting cause tracing. This paper introduces the concept of quality tracing, prediction, and early warning using LightGBM-SHAP. Firstly, LightGBM is employed to model the relationship between quality-related factors and target variables. Subsequently, the SHAP method is applied to analyze the significant importance of each influencing factor and deeply assess its impact pattern. This facilitates the effective identification and tracing of quality-influencing factors within electrical equipment. Finally, pivotal influencing factors are selected based on their characteristic importance, and LightGBM is utilized to predict the quality issues. Illustrating with transformers as an example, the predictive and early warning capabilities of the LightGBM-SHAP model are proposed to outperform the comparison model. Furthermore, it effectively dissects the causes of quality problems and offers decision support for quality tracing, prediction, and early warning in the realm of electrical equipment.