Data-driven Dynamic Baseline Calibration Method for Gas Sensors

Sammanfattning: Automatic Baseline Correction is the state-of-the-art calibration method of non-dispersive infrared CO2 sensing, which is the standard CO2 gas monitoring method. In this thesis, we improve it by introducing the dynamic baseline based on environmental data. The 96 data sets from 48 atmospheric stations verify the characteristics of the annual growth trend and seasonality of the baseline model. In order to improve the accuracy of the calibration, the k-means clustering method is used to identify different types of baselines. Then the localized dynamic baseline model is predicted by using the location information of the stations only, which provides an executable calibration implementation for dynamic baseline calibration without relying on historical CO2 data.