Application of bi-directional long short-term memory in separating NO and SO₂ ultraviolet differential absorption spectrum signals

To safeguard the environment, it is crucial to monitor the emissions of nitrogen oxide (NO) and sulfur dioxide (SO 2 ), harmful pollutants generated during fossil fuel combustion in industries. However, accurately measuring ultra-low concentrations of SO 2 and NO remains a challenge. In this study, we developed an optical measurement system based on ultraviolet differential optical absorption spectroscopy (UV-DOAS) to address this issue. The 200 - 230 nm cross-sensitivity band was chosen for SO 2 and NO. Experimental data with a mixed gas concentration range of 1 - 25 ppm for SO 2 and NO was utilized. We proposed a fast algorithm based on Bi-directional Long Short-Term Memory (Bi-LSTM) to extract the differential optical density, overcoming the mutual interference between SO 2 and NO. A nonlinear calibration model was employed to invert the separated differential absorption spectra and determine the gas concentrations. The results demonstrated a detection limit (DL) of 0.27 ppm and a full-scale error of 3.15 % for SO 2 , while for NO, the DL was 0.32 ppm and the full-scale error was 2.81 %. The uncertainties in SO 2 and NO detection were calculated as 1.73 % and 1.96 %, respectively.