Research on CO concentration detection based on deep learning and TDLAS technology

Abstract

In order to improve the accuracy of detecting CO and other gases in the industrial field, a CO concentration detection scheme based on tunable diode laser absorption spectroscopy (TDLAS) and the CO absorption spectral line at 1566 nm was proposed. At the outset, after removing the outliers from the original data, the harmonic signal is filtered using a moving average filtering method to enhance the signal-to-noise ratio. Subsequently, a one-dimensional convolutional neural network (1D-CNN) is employed to extract the peak and valley values, as well as the mean values of each segment of harmonic data. In the end, the Long short term memory network (LSTM) model establishes a nonlinear fitting relationship between gas characteristics and concentration, and the attention layer is added to further optimize the network weight parameters. The experimental results indicate that the R2 for the test set is 0.9884, and the RMSE is 156.869. Through Allan deviation analysis, the detection limit reaches 1.9605 ppm at an integration time of 60 s. Compared to the traditional regression model, the accuracy and stability are significantly improved, making it highly suitable for gas detection in the industrial circle.