Method for accurate gas concentration retrieval in an in situ environment with particulate interference

Tunable diode laser absorption spectroscopy (TDLAS) technology is widely applied in trace gas detection due to its high sensitivity, fast response, and compact setup. However, within a complex in situ environment, the existence of particulate matter can attenuate the transmitted spectral signals and degrade the signal-to-noise ratio (SNR), thereby severely compromising gas concentration measurement accuracy. Existing denoising methods for TDLAS fail to adequately address the combined effects of particle extinction and noise interference on the transmitted spectrum. Moreover, their applicability under high-concentration particulate interference is further restricted by the difficulty of accurately fitting the baseline from the absorption signal. To solve the problem above, a transmitted spectral processing method based on variational Bayesian adaptive extended Kalman filtering (VB-AEKF) is proposed in this paper. By incorporating Mie scattering theory into the Beer-Lambert law, the state vector composed of concentration value and extinction coefficient is formulated, and the observation model for the TDLAS system is developed. Then, a dual-layer iterative structure is employed to achieve state estimation under unknown baseline conditions, ensuring reliable gas concentration retrieval even under severe particle scattering interference. Compared with traditional filtering methods, the VB-AEKF algorithm has an enhanced denoising effect for the simulated signals. The error of concentration retrieval results is less than 1%, and the calculation accuracy and stability of the extinction coefficient are superior to those of the robust linear regression method. Additionally, in situ methane concentration measurement experiments under various particle sizes and mass concentrations verified the effectiveness of the VB-AEKF algorithm, demonstrating significant advantages in baseline estimation, signal reconstruction, and concentration retrieval. The results showed that the mean relative error of measured gas concentration obtained by the proposed method is 0.46%, which is substantially lower than that of traditional methods (4.69%), and the standard deviation of concentration is reduced by 61.7%.