Use Support Vector Machine (SVM) to estimate gas concentration in mixture condition

摘要

In many gas sensors, the selectivity is still a big issue, which makes the real concentration distortion. We introduce a way to estimate its concentration in gas mixture condition. The system consists of three gas sensors, carbon dioxide sensor, carbon monoxide sensor and humidity sensor in environment-controlled chamber, which give lots of different concentration combinations. To estimate its concentration, we use machine learning techniques called Support Vector Machine (SVM). First, the measurement data are labeled according to its concentration combination. Second, the measurement data are trained as many classifier via SVM and then make new measurement data (unlabeled) to be classified. After classified, each label will get different votes. Finally, weighted averages concentration according to the votes. We show that the analysis results are close to the real values.