Dual-Component Gas Sensor Based on Light-Induced Thermoelastic Spectroscopy and Deep Learning

Abstract

In this paper, an acetylene–carbon dioxide dual-component gas sensor based on light-induced thermoelastic spectroscopy and deep learning is reported for the first time. Two lasers with wavelengths of 1530 and 1577 nm were coupled by a wavelength division multiplexer to excite the two gas molecules. The sensor combined four algorithms, namely, sparrow search algorithm (SSA), convolutional neural network (CNN), bidirectional gated recurrent unit neural network (BiGRU), and attention mechanism (Attention), to achieve two-component gas concentration inversion in three cases, in which the overlap of the two gas spectral lines is different. The advantage of the combination of the SSA-CNN-BiGRU-Attention model is that the weight can be assigned according to the characteristics of the second harmonic (2f) signal itself, and the optimal parameters can be automatically found. The SSA-CNN-BiGRU-Attention model effectively improved the accuracy of concentration inversion and significantly reduced the mean relative error (MRE). The experimental results show that the R-square values of the test set are all greater than 0.99, and the MRE is less than 1.2%, showing a high concentration inversion accuracy. This work provides instructions for the absorption line overlap in dual-component gas detection and is expected to be applied to the study of concentration inversion of more gas components in the future.