Abstract
Characteristic gas monitoring is an important approach for detecting incipient faults and supporting the intelligent operation and maintenance of high-voltage switchgear. However, existing methods primarily rely on concentration thresholds or steady-state features, making it difficult to simultaneously optimize sensor placement and accurately localize fault-gas sources. To address these limitations, this study proposes a virtual–physical sensing method for fault-gas source localization. A full-scale test platform and a 1:1 three-dimensional numerical model of 10 kV air-insulated switchgear (AIS) were constructed to obtain the time-series responses of CO, CO2, NO, and O3 for different fault-gas source locations. Candidate monitoring points were quantitatively evaluated using a combination of the Welch t-test, mutual information, and entropy difference. Multi-gas features from the selected monitoring points were then fused to identify nine fault-location classes. The results showed that the reported relative errors in final recorded concentrations did not exceed 9.09%; MP8, MP1, and MP2 were the three highest-ranked monitoring locations; under five-fold stratified cross-validation, the accuracy, macro-precision, macro-recall, and macro-F1 score of the random forest model all reached 100%. The proposed method provides a quantitative basis for reducing the number of gas sensors and localizing fault-gas sources in high-voltage switchgear.