A Methane Sensor Based on Serial-Reference Physics-Guided Inference under Mixed-Alkane Absorption Interference

Accurate methane quantification by absorption spectroscopy remains challenging in mixed-alkane backgrounds. Broadband alkane absorption induces optical-power attenuation and matrix-dependent wavelength modulation spectroscopy (WMS) harmonic distortion, causing the measured signal to deviate from a simple one-to-one relationship with methane concentration. As a result, conventional direct retrieval based on methane absorbance or harmonic amplitude becomes less reliable without gas separation or effective interference correction. To address this issue, a methane sensor based on a serial-reference physics-guided inference (SRPGI) method is proposed. A serial-reference optical configuration is designed to provide shared reference information within the same scanning cycle, and a physics-guided residual inference framework is developed by integrating reference-assisted features, physics-based priors, and data-driven residual correction. The proposed method enables direct methane quantification without pretreatment or chromatographic separation. Experimental results show that, in the ablation study, the concentration error under interference is reduced from 20.99 ppmv to 0.32 ppmv at 150 ppmv CH4, corresponding to an approximately 65.6-fold reduction in retrieval error. With increasing mixed-gas interference strength, the raw error rises substantially, whereas the SRPGI relative error remains below 0.4%. Across different alkane background gases, the SRPGI relative error remains below 0.25%. The lightweight SRPGI model also preserves the sensor response speed, with the response time (T10%–90%) remaining nearly unchanged from 31.0 s to 31.2 s. These results demonstrate that the proposed sensor provides improved accuracy, stability, and matrix adaptability for direct methane quantification in mixed-alkane background conditions.