A Dual-Path Deep Learning Model for Low-Cost Air Quality Sensor Calibration

The low-cost electrochemical sensors have a wide range of applications, including air quality monitoring. However, these sensors are particularly susceptible to significant drift when exposed to unstable temperatures, an issue that becomes critical in air quality monitoring scenarios. Temperature fluctuations over the year can induce considerable sensor drift. While many researchers have proposed algorithms to calibrate these low-cost sensors, few of them treated them as a crucial variable and developed an independent mechanism to address sensor drift during the training of their machine learning or deep learning models; instead, it is classified as a variable that is the same as gas sensors for the input data of the model. In this study, we introduce a calibration algorithm that builds upon the 1-D CNN model by incorporating an environment recognition pathway to compensate for sensor drift induced by temperature fluctuations. The core concept is to enable the model to understand that temperature can cause nonlinear effects on sensor readings. The model needs to learn how temperature influences drift by applying an additive bias to the final output of the CNN model. In this research, we analyzed a dataset, which was collected over a year. The proposed network demonstrates strong resilience to temperature-induced drift throughout a yearlong experiment.