Dynamic calibration of low-cost PM_2.5 sensors using trust-based consensus mechanisms.

Area of Science:

• Environmental science
• Sensor technology
• Data science

Background:

• Low-cost particulate matter (PM) sensors offer high-resolution urban air quality monitoring.
• These sensors face challenges including offsets, scaling mismatches, and drift, impacting data reliability.
• Existing calibration methods often require extensive data and frequent recalibration.

Purpose of the Study:

• To develop an adaptive, trust-based calibration framework for low-cost PM sensors.
• To dynamically adjust calibration models based on individual sensor reliability and performance.
• To improve the accuracy and scalability of urban air quality monitoring networks.

Main Methods:

• Proposed an adaptive trust-based calibration framework integrating accuracy, stability, responsiveness, and consensus alignment.
• Implemented a system where high-trust sensors receive minimal correction, while low-trust sensors utilize advanced wavelet-based features and deeper models.
• Validated the approach through extensive simulations and real-world deployment in Zurich, Switzerland.

Main Results:

• Achieved mean absolute error (MAE) reductions of up to 68% for poorly performing sensors and 35-38% for reliable sensors.
• Demonstrated superior performance compared to conventional calibration methods.
• Showcased reduced dependence on large training datasets and frequent re-calibrations through trust-weighted consensus.

Conclusions:

• The adaptive, trust-driven calibration framework significantly enhances the accuracy of low-cost sensor networks.
• The method proves effective in both controlled simulations and complex real-world urban environments.
• This approach ensures scalability and maintains data integrity for improved air quality management.