CO concentration prediction in mid-infrared TDLAS systems via an EWT-SE-WTD enhanced BPNN model

In a tunable diode laser absorption spectroscopy (TDLAS) gas detection system, nonlinear noise and background interference can significantly degrade the characteristic features of the second harmonic signal, thereby affecting the accuracy of the detection system. To address this issue, this paper proposes a method that combines the EWT-SE-WTD denoising algorithm with a concentration prediction model based on a back-propagation neural network (BPNN) to improve system performance. The denoising process first applies empirical wavelet transform (EWT) to adaptively decompose the noisy signal, then utilizes spectral entropy (SE) in conjunction with wavelet threshold denoising (WTD) to reconstruct and smooth the signal. Experimental results demonstrate that the algorithm increases the signal-to-noise ratio (SNR) of the second harmonic signal from 95.48 to 583.56, significantly enhancing the signal quality. Incorporating this method into the BPNN model yielded a fitting correlation coefficient (R²) of 0.9999 between the predicted means and the standard values within the concentration range of 10 ppm to 100 ppm. Furthermore, the system limit of detection (LOD) is 171 ppb, verifying the effectiveness and stability of the method in low-concentration, high-precision gas detection scenarios.