A dual sensor for SO2 concentration and temperature based on ultraviolet differential optical absorption spectroscopy combined with convolutional neural network

Sulfur dioxide (SO2) is an unavoidable pollution product of fossil fuel combustion, and its concentration and temperature levels are considered key indicators of combustion efficiency and energy utilization. Currently, SO2 concentration is commonly measured by ultraviolet differential optical absorption spectroscopy (UV-DOAS), but simultaneous detection of SO2 concentration and temperature has not yet been realized because the effects of temperature and concentration on spectral intensity cannot be distinguished. To address this issue, a dual sensor is reported in the study for simultaneous detection of SO2 concentration and temperature by combining UVDOAS with convolutional neural network (CNN) model. Firstly, we elucidate temperature-induced redshifts and intensity variations in ultraviolet differential absorption spectra over the range of 298.15-873.15 K. Then we establish the law of variation for spectral intensity at each wavelength. On this basis, a dual-parameter data expansion method is introduced to obtain 31,500 groups of spectral data at various concentrations and temperatures. Finally, the CNN model is used to achieve simultaneous detection of SO2 concentration and temperature. The experimental results of this sensor showed that the mean relative errors (MREs) of SO2 concentration and temperature are 1.5 % and 1.3 % in the concentration range of 5.01-600.06 parts per million (ppm) and temperature range of 298.15-873.15 K, respectively. This demonstrates the potential of the sensor in optimizing combustion efficiency.