Balanced Distribution Adaptation for Metal Oxide Semiconductor Gas Sensor Array Drift Compensation.

Area of Science:

• Sensor Technology
• Machine Learning
• Data Science

Background:

• Drift compensation is crucial for metal oxide semiconductor (MOS) gas sensor arrays.
• Traditional machine learning methods demand frequent calibration and extensive labeled gas data, which is costly and impractical.
• Recalibration and obtaining labeled gas data pose significant challenges in real-world applications.

Purpose of the Study:

• To propose a novel drift compensation method for MOS gas sensor arrays.
• To reduce domain mismatch between sensor data using a weight balance factor.
• To enhance drift compensation accuracy through optimization techniques.

Main Methods:

• A novel balanced distribution adaptation (BDA) method is introduced.
• BDA adjusts conditional and marginal distributions using a weight balance factor to minimize domain mismatch.
• Particle swarm optimization is employed to determine optimal weight balance factor parameters for improved accuracy.

Main Results:

• The BDA method demonstrated superior performance compared to joint distribution adaptation (JDA) and K-Nearest Neighbor (KNN).
• Recognition accuracy improved by 4.54% and 1.62% over JDA in two experimental settings.
• Accuracy gains of 12.23% and 15.83% were observed over KNN.

Conclusions:

• The proposed BDA method offers effective drift compensation for MOS gas sensors without requiring target domain labeled data.
• BDA significantly outperforms existing standard drift compensation techniques.
• The integration of particle swarm optimization further refines the accuracy of drift compensation.