Methanol Online Detection and Methane Interference Dynamic Compensation Method Based on EAWEL-BPNN

Accurate detection of methane and methanol is crucial for industrial safety and sustainable energy systems, yet conventional non-dispersive infrared (NDIR) sensors exhibit cross-sensitivity in the 3–4μm band due to spectral overlap between these compounds. While current mitigation strategies combine multi-band spectroscopy or machine learning algorithms (e.g. partial least squares regression, convolutional networks), they encounter practical limitations including high implementation costs and computational inefficiency. This study develops a dynamic compensation framework integrating multi-band spectral analysis with an error-adaptive weighted ensemble learning-enhanced BP neural network (EAWEL-BPNN). The system synergizes a 3.3 μm NDIR methanol sensor and 1684 nm tunable diode laser absorption spectroscopy (TDLAS) methane detector, achieving real-time interference decoupling through multi-wavelength datas fusion. Experimental validation demonstrates methanol detection accuracy improvement from ±40.6%FS to ±1%FS under 0–2500 ppm methanol and 0–2000 ppm methane concentrations. This hardware-algorithm co-design establishes a computationally efficient paradigm for multi-component gas analysis in industrial monitoring scenarios.