ResNetTCA-Assisted Synergetic Laser-Electronic Sensor Array for Multiple Gas Detection for Underground Pipeline Leakage

Cross-sensitivity has long posed a major challenge in multiple gas detection, requiring the construction of sensor arrays. However, the simultaneous physical adsorption and chemical reactions from different types of gas molecules on the sensor surface are the primary causes, leading to signal overlap that is difficult to separate. Herein, we propose a synergetic laser-electronic sensor array (SLESA) capable of simultaneously capturing the physical and chemical properties of target gases. The detection accuracy of the sensor array has been significantly improved by introducing the ResNet enhanced with a temporal attention mechanism [residual network with temporal convolutional attention (ResNetTCA)]. The fivefold cross-validation results indicate that the ResNetTCA model achieved regression accuracies of 98.7%, 97%, and 96.1% for methane (CH₄), carbon monoxide (CO), and hydrogen sulfide (H₂S) concentrations, respectively. The proposed ResNetTCA model demonstrates superior performance over conventional neural network architectures, such as convolutional neural network (CNN), ResNet-18, ResNet-34, long short-term memory (LSTM), and transformer, achieving an improvement in regression accuracy ranging from 0.8% to 5.7%. The combination of SLESA and the ResNetTCA algorithm enables effective suppression of high-humidity interference in underground gas pipelines, leading to highly accurate gas detection. This study not only addresses cross-sensitivity in multiple gas detection but also provides reliable strategy for underground gas pipeline leakage detection with accuracy and fast alerts.