Discrimination and Prediction of SF₆ Decomposition Gas Mixtures Based on Distribution of Relaxation Time Analysis

Gas sensor array based on MOS sensors is widely used for gas discrimination and concentration prediction in mixed gases, but their performance is often limited by cross-sensitivity, making it difficult to effectively distinguish the impact of each gas component using traditional DC resistance measurement. In contrast, Distribution of Relaxation Time (DRT) provides more comprehensive electrochemical information about the interactions between mixed gases and the sensor, holding the potential to enhance the capability of gas discrimination and concentration prediction for mixed gases. This study utilizes Distribution of Relaxation Time (DRT) analysis to investigate the EIS data of four types of sensors exposed to H₂S, CO, SO₂, and their binary mixtures, revealing the competitive processes between different gases as well as the changing patterns of relaxation time and impedance of the sensor with different gas concentration ratios. Based on the features extracted from DRT analysis, the Transformer-Encoder model achieved the best performance for mixed gas prediction, with the RMSE reaching as low as 0.94 ppm. In conclusion, our findings provide a novel approach for gas discrimination and concentration prediction for mixed gases.