Accurate hazardous gas detection under uncertain disturbing environment with multi-fusion model combined with artificial olfactory system

• A novel precise hazardous gas detection based on artificial olfactory system (AOS) was proposed. • Deep learning AOS model has better performance for gas quantitate identification under disturbing environment than common regression method. • The multi-fusion exhibits an improved quantitative identification accuracy of gases in the presence of interfering components. • The transfer learning model and multi-fusion model can serve as precise detection methods for monitoring leaking gases in complex environments. The rapid expansion of urban gas infrastructure has led to a significant increase in the frequency of gas pipeline maintenance, renovation, and emergency repairs. However, this has also resulted in an increasing number of safety accidents. Conventional gas detection methods, such as metal oxide semiconductor (MOS) gas sensors, have a natural fault of cross-response, resulting in low detection accuracy and a high false alarm rate. To address these challenges, a novel gas precise identification technology based on artificial olfactory systems (AOS) under uncertain disturbing environment was investigated with the primary objective of achieving accurate quantitative detection amidst interfering gases. First, a conventional univariate linear regression was employed to forecast gas concentration with the MOS sensor. To enhance the accuracy of gas leak detection under a disturbing environment, gas prediction approaches were proposed and developed based on different shallow machine learning and deep learning models combined with AOS response map. The results showed that due to the properties in time-frequency feature extraction of AOS response map data, the transferred ResNet18 model achieved an impressive coefficient of determination of 0.8368 compared with the average coefficient of 0.6653 for the common univariate linear regression model. Furthermore, to further enhance the model’s recognition accuracy in the presence of interfering gases, a multi-fusion model was established based on artificial olfactory data to enhance the features of the main component in the interfered atmosphere. The results demonstrate that the multi-fusion model exhibits improved quantitative identification accuracy of gases in the presence of interfering components, achieving a superior average coefficient of determination of 0.9417 for pure components and a remarkable over 240% improvement over traditional methods under interference, as well as similar computation efficiency of the single shallow model. Consequently, the transfer learning model, along with the multi-fusion model based on AOS, can serve as precise detection methods for monitoring leaking gases in complex environments. The results obtained in this research can be applied to identify the leaking or emission gases accurately under conditions with complex disturbing components, which can be utilized in gas pipeline monitoring and extended to other work scenarios, such as oil and gas production stations and long-distance pipeline repairmen, etc.