A Machine-Learning-Algorithm Enhanced Multi-Functional Gas Sensor for Self-Humidity Compensation and Partial Discharge Detection

Gas-Insulated switchgear (GIS) is prone to partial discharges (PDs) in high electric field environments, and the concentration of generated NO2 is an essential indicator for determining the PD types and severity of faults. Notably, environmental humidity greatly influences the insulation performance of gas-insulated switchgear and the signals of NO2 gas sensors. Thus, the simultaneous detection of humidity and NO2 and the decoupling of signals has practical importance. Herein, a groundbreaking sensor is developed to achieve self-calibrated sensing of humidity and NO2 gas, which is realized by a multifunctional WS2/ZnO sensitive material with an innovative self-humidity compensation algorithm of DF-MT1DCL. This synergistic system delivers dynamic, real-time humidity adaptive calibration and also enables precise recognition of partial discharge types. The sensor exhibited simultaneous response and a wide detection range (100 ppb-10 ppm of NO2, 10.8-94.3% RH) exposed to NO2 and humidity at room temperature. As a result, simultaneous monitoring and decoupling of signals can be realized. Further, a multitask deep learning model DF-MT1DCL combined 1D-CNN with LSTM was proposed to complete the humidity adaptive calibration based on a single WS2/ZnO sensor, which realizes the simultaneous prediction of humidity and NO2 concentration, with R2 values of 99.1% and 93.5% respectively. The WS2/ZnO sensor with excellent humidity and NO2 sensing performance and the DF-MT1DCL algorithm assistance was applied to partial discharge monitoring in a simulated gas-insulated switchgear, and high-precision classification of partial discharge types was achieved with 100% classification accuracy. Therefore, the constructed WS2/ZnO multifunctional sensor combined with the DF-MT1DCL algorithms improves the resistance to humidity interference of NO2 detection and also accurately recognizes the partial discharge type, which provides a new perspective for the intelligent sensing technology for health monitoring of electric power equipment.