Research on Atmospheric Pollution Gas Concentration Detection System Based on Hybrid Learning Model

Due to the limited number and uneven distribution of air quality monitoring equipment, the use of sensors to detect the concentration of polluting gases has become an alternative. However, the gas sensor is prone to aging and is easily affected by environmental variables, resulting in sensor drift. Firstly, in order to find a suitable calibration model to reduce the number of periodic calibration, a hybrid model combining Random Forest (RF), XGBoost and Bi-directional long-term and short-term memory network (Bi-LSTM) is proposed to calibrate and verify the sensor response value. Secondly, the computational fluid dynamics (CFD) simulation was used to optimize the structure of the sensor chamber, aiming at the difference of sensor response value caused by airflow interference and uneven airflow distribution. Finally, an air pollution gas concentration detection system based on RF-XGB-Bilstm hybrid model is developed to measure the concentrations of CO, SO 2 , NO 2 and O 3 . According to the evaluation index results of R ² , RMSE and MAE, it can be concluded that the atmospheric pollution gas concentration detection system developed in this paper has been proved to be feasible and practical.