Research On The Bicomponent Gas Sensing System Based On Neural Network Spectroscopic Decoupling Algorithm

One challenge that remains to be addressed for the simultaneous detection of multiple gas components using spectroscopy is the issue of interference between the spectra of different gas components.Although mathematical methods such as Bayesian estimation(BE)and principal component analysis(PCA)have been used to mitigate cross-interference,these methods often rely on strict prior conditions and assumptions that are difficult to satisfy in practical applications,resulting in reduced predictive accuracy.Least squares(LS)and partial least squares(PLS)are also effective in mitigating cross-interference,but they are unable to identify the specific types of gas mixtures present.By contrast,deep learning has emerged as a more promising option in recent years,owing to its strong performance and integrability with other scientific methods.Various neural network-based deep learning algorithms have yielded impressive results in gas concentration retrieval,spectral filtering,and related areas.Further research is needed to fully explore the potential of deep learning algorithms in addressing the challenges associated with detecting multiple gas components using spectroscopy.The main points of this article are as follows:1.In this study,we conducted a detailed examination of the theoretical basis of gas spectroscopy and the basic principles of deep learning neural network algorithms.Based on a thorough understanding of the relevant principles,we developed a methane and water dual-component gas sensor that integrates a neural network-based spectroscopic decoupling algorithm.This overcomes the common problem of water interference in methane detection using broadband mid-infrared absorption spectroscopy.In contrast to traditional gas detection methods that rely on gas molecule specificity for detecting multiple components,the proposed detection algorithm can directly address the issue of cross-interference in multi-component gas detection,achieve decoupling the absorption spectra of mixed gases,and realize fast and accurate detection of multiple gas components without any additional equipment.2.The proposed system was tested in terms of prediction error,long-term stability,and noise adaptability.To improve concentration inversion accuracy,we introduced a mechanism for selecting relevant absorption peaks instead of relying on a single feature.We conducted an hour-long experiment in which we sampled a mixture of 5 ppm methane and 1500 ppm water vapor every minute,and the system showed superior stability with average predicted concentrations of 5.0014 ppm for methane and 1499.998 ppm for water vapor.Furthermore,the system demonstrated accurate predictions even with deteriorated data quality,indicating its robustness.Overall,our system has strong accuracy and robustness,making it suitable for various practical scenarios.