Design and Validation of an Embedded Baseline Compensation Model with Wireless Sensor Network for Monitoring Total Volatile Organic Compounds

Portable and intelligent sensor systems are required for real-time air quality monitoring, and among them, the metal-oxide-semiconductor gas sensors are widely used in many scenarios. However, the inconsistency, baseline drifts, and complex off-board algorithms hinder their further applications. In this study, a lightweight artificial neural network-based baseline compensation model is proposed, which is integrated within a minimized air quality monitoring system for real-time total volatile organic compound (TVOC) sensing. The model is trained and applied to correct baseline drift across six additional VOC data sets. Evaluated with a one-dimensional convolutional neural network, the compensation method yields a 31.4% increase in R2 score in concentration prediction for a specific VOC data set. The model is then deployed onto the microcontroller of each sensor node in the system, which autonomously corrects intraexperimental baseline drift for an individual sensor and harmonizes the initial baselines across multiple sensors. After normalization, the maximum initial baseline variation among sensor nodes is reduced by four times. The system is deployed in a dormitory via a ZigBee mesh, providing intuitive air quality readings for different indoor environments. This work lays the groundwork for a broad implementation of compact, efficient sensor networks that enable precise, real-time indoor air quality assessment.