PM2.5 Sensor Drift Compensation: Real-Time Hardware Deployment Dataset
Description
This repository contains a real-time hardware validation dataset collected using a Raspberry Pi 5–based adaptive PM2.5 sensing platform deployed in an HVAC-conditioned indoor laboratory. The monitoring campaign spans approximately 5.98 days of continuous unattended operation and contains 103,059 synchronized observations recorded at approximately 5-second sampling intervals. The deployed framework executes 7,305 estimator resets (RL_Action = 17.1% of timesteps) while maintaining drift RMSE = 0.464 μg/m³ below the measurement noise floor (σᵥ = 1.128 μg/m³) on all six deployment days. The sensing platform integrates a Plantower PMSA003I low-cost optical particulate matter sensor and a BME688 environmental sensor. The deployed adaptive sensing framework combines humidity-bias correction, Kalman filter–based state estimation, and a reinforcement learning supervisory controller to perform adaptive estimator resets during continuous operation. The dataset contains timestamped measurements of PM2.5 concentration, temperature, relative humidity, atmospheric pressure, gas resistance, estimated PM2.5 baseline, humidity-bias estimate, Kalman filter residual, estimated drift state, Kalman gain, posterior error covariance, reinforcement learning state, reinforcement learning action, cumulative estimator-reset count, corrected PM2.5 concentration, and spike-detection status. These runtime variables represent the complete output of the deployed adaptive sensing framework and support reproducible hardware validation of the proposed methodology. Unlike the companion calibration dataset, this repository contains data collected exclusively during independent real-time hardware deployment. The calibration and model-development data used for signal characterization, stochastic digital twin parameterization, Kalman filter design, and reinforcement learning training are available separately in the companion dataset, Longitudinal Indoor Air Quality Dataset Collected Using a Low-Cost Multi-Sensor IoT Monitoring Platform (Mendeley Data, Version 2, DOI: 10.17632/b5jvs7kykn.2). This dataset is designed to facilitate research in low-cost air quality sensing, adaptive sensor management, sensor drift compensation, Kalman filtering, reinforcement learning, digital twins, edge computing, Internet of Things (IoT) applications, indoor environmental monitoring, and intelligent sensing systems. It may also serve as a benchmark dataset for the development and evaluation of adaptive calibration algorithms, state estimation methods, uncertainty-aware sensing frameworks, anomaly detection techniques, and hardware-in-the-loop edge AI applications. Future versions will incorporate hardware configuration documentation, sampling protocol details, additional deployment campaigns, source code, archived software releases, and links to related publications.
Files
Institutions
- The University of Texas at El PasoTexas, El Paso