Enhanced trace CO2 detection sensor for gas production monitoring using QCL absorption spectroscopy with CPO-BiLSTM model

Abstract

Carbon dioxide (CO₂) frequently manifests as a contaminant within a range of pure gases. A trace CO₂ detection sensor designed based on tunable diode laser absorption spectroscopy (TDLAS) and wavelength modulation spectroscopy (WMS) technology is proposed. This sensor employs a quantum cascade laser (QCL), a multi-pass gas cell (MPGC) and a HgCdTe photodiode for photoelectric conversion. The extracted spectral signals are smoothed and denoised by wavelet transform optimized by empirical mode decomposition (EMD). The signal-to-noise ratio (SNR) of the spectra has been elevated from 20.22 dB to 29.60 dB, resulting in a significant reduction of the noise component. The crested porcupine optimizer (CPO)-bidirectional long short-term memory (BiLSTM) model was used to convert the concentration. Comparison with back-propagation neural network (BPNN) and least squares support vector machine (LSSVM), the experiment shows that the CPO-BiLSTM model outperforms the other two with a root mean square error (RMSE) of 0.0078. Allan analysis of the sensor yielded a minimum theoretical limit of detection of 1.56 ppb. This sensor can be used for long term monitoring of the CO₂ concentration in pure gases.