Research on on-line detection of CO and CO2 mixed gas concentration based on improved extreme learning machine and TDLAS
Yinsong Wang, Qingmei Kong, Jianqiang Gao, Shiman Chen
Abstract
In order to improve the detection accuracy of CO and CO 2 mixtures in the industrial field, and to solve the problems that the existing gas concentration detection models are vulnerable to environmental interference, lack generalization ability, and cannot be updated online, a concentration detection method based on an improved online extreme learning machine is proposed. Based on tunable semiconductor laser absorption spectroscopy (TDLAS), a deep extreme learning machine was used to detect gas mixtures online. Firstly, a TDLAS system with a wavelength near 1583 nm was used to analyze mixed gases of CO and CO 2 at different concentrations, and the initial gas detection model was established using an offline database. Then, a new sample number is obtained in real time during the detection process to update the model parameters online, and a dynamic forgetting factor is introduced to adjust the weight of the new and old samples to improve the detection accuracy and adaptive ability of the algorithm. Finally, the experimental results show that the algorithm can update the model parameters online when the concentration changes, and the RMSE of CO and CO 2 are 0.01243 % and 0.11856 %, respectively, which achieve high detection accuracy and have certain engineering application value.