A baseline correction and noise suppression method based on fitting neural network for CH4/C2H6 dual gas sensing system

Abstract

We designed a methane/ethane (CH4/C2H6) dual-gas sensor on the basis of direct absorption spectroscopy (DAS) technology and explored the potential for low power consumption and miniaturization of the detection system using the effective optical range is only 25 cm for absorption cell and an uncooled photodetector. The problem of baseline drift introduced by uncooled photodetectors has been improved by using of a fitting neural network approach, and absorbance spectrum denoising has been realized. We established the mathematical relationship between absorbance and concentration for each gas of CH4/C2H6 through experimental detection, which shows strong linearity. And the correlation coefficients (R2) reached 0.99995 and 0.99998, demonstrating the robustness of the sensor. Through 3 h of measured and evaluated system, we obtained the minimum detectable column densities of 0.588 ppm⋅m and 0.128 ppm⋅m for CH4 and C2H6, respectively, which proved that the sensor has high sensitivity, strong stability, and has broad application prospects.