使用支援向量機提升多個氣體感測模組量測之準確度

The enhancement of sensing characteristics in multiple sensing module by support vector machine algorithm

林冠州(Kuan-Chou Lin)

指導教授 ： 林致廷

國立臺灣大學/電機資訊學院/電子工程學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201701998

摘要

近幾年來物聯網的產值逐年攀升，各企業無非想成為物聯網技術的先驅，其技術主要可以分為三個層面，應用、網路和感測。在物聯網的大架構下，需要使用大量的感測器，故感測器的價格和功耗變成為重要的課題，市面上光學式感測器雖然擁有良好的精確度，但單顆感測器價格過於昂貴；金屬氧化物式感測器有著不錯的靈敏度，但要先升至高溫並操作在固定溫度下，功耗過高；有機感測器有著價格便宜和功耗低的優點，但選擇性不佳和生命期較短。金屬氧化物式氣體感測器和有機氣體感測器都會受到選擇性影響，簡單的說，當感測多種氣體混合時，量測結果會失真，而此失真結果並非全然線性組合。本論文使用一種機器學習的方法，叫支援向量機，將陣列所量測的資料用支援向量機建立分類器模型，新量測的資料經模型計算後，減少因選擇性不佳所造成的誤差。

關鍵字

氣體感測器 ； 機器學習 ； 支援向量機 ； 濃度計算

並列摘要

The output value of Internet-of-Things(IOT) has been increased in recent years, and all companies want to be the leader of it. The techniques of IOT can be divided into three parts, applications、Internet and sensing. Also, in the framework of IOT, need lots of sensors, so the price and the power consumption of sensors become a important issue. Commercial optical-sensors have a good sensing accuracy, but they are too expensive. Metal-oxide-sensors have a good sensitivity, but they have to operate at specific temperature, and it makes to much power consumption. Organic gas sensors are cheap and low power consumption, but the selectivity and life time is not good. Metal-oxide and organic gas sensors both suffer from the poor selectivity, that is, when sensing mixing gas, the measurement results will distort and may not be linear combination. This thesis uses one of machine learning, called Support Vector Machine (SVM), to build a classifier module with the measurement data from gas sensor array. New measurement data can be calibrated after calculating by module and reduce the error.

並列關鍵字

Gas Sensor ； Machine Learning ； Support Vector Machine ； Concentration Calculation

參考文獻

[27] MG-811 datasheet, HANWEI ELECTRONICS CO. , LTD

[2] Xiao Liu, Sitian Cheng, Hong Liu, Sha Hu, Daqiang Zhang and Huansheng, Ning, A Survey on Gas Sensing Technology,Sensors 2012, 12, 9635-9665

[3] H Sundgrent, F Wlnqulstt, I LukkarlS and I Lundstrom, Artificial neural networks and gas sensor arrays: quantification of individual components in a gas mixture, Meas. Sci. Technol. 2 (1991) 464-469.

[4] Walaa Khalaf , Calogero Pace and Manlio Gaudioso, Least Square Regression Method for Estimating Gas Concentration in an Electronic Nose System, Sensors 2009, 9, 1678-1691

[5] Chih-Wei Hsu, Chih-Chung Chang, and Chih-Jen Lin,A Practical Guide to Support Vector Classification