基于单传感器动态检测和神经网络的气体监测  

Monitoring of gases based on single sensor’s dynamic detection and neural network

作　　者：黄家锐[1] 李广义[2] 徐伟宏[1] 黄行九[1] 刘锦淮[2] 

机构地区：[1]中国科学院智能机械研究所,安徽合肥230031 [2]中国科学技术大学化学系,安徽合肥230026

出　　处：《传感器技术》2005年第7期72-75,共4页Journal of Transducer Technology

基　　金：国家自然科学基金资助项目(60374049)

摘　　要：研究了气体传感器动态检测结合神经网络识别空气中有机气体的新方法。这种方法利用单个SnO2气体传感器而非阵列在方波温度调制的状态下,可实现多种有机气体的定性分析。在0.02Hz的调制频率、250～300℃的温度调制范围,测得传感器对不同体积分数异丙醇、乙酰丙酮和乙醇气体的动态响应。通过小波变换对单个周期测试信号进行特征提取,并将提取的特征值输入神经网络进行定性识别和体积分数估计,仿真识别的成功率高达100%,同时,较准确地实现待测有机气体的体积分数估计。A new method combining the gas sensor’s dynamic detection and neural network is described for the rapid identification and determination of volatile organic compounds in ambient air.For the qualitative recognition of the volatile organic compounds,only a single SnO_2-based gas sensor operating in a rectangular temperature modulation mode is required.At 0.02Hz,in the range of 250~300℃,dynamic response of the sensor to different volume fraction of propane-2-ol,acetyl acetone and ethanol is measured.The discrete wavelet transform (DWT) is used to extract important features from the sensor response.These features are then inputted to the neural network and qualitatively recognized,and the volume fraction is predicated.The species considered can be discriminated with a 100% success rate using back propagation network and the volume fraction of the gases studied can also be accurately predicted.

关 键 词：动态检测 小波变换 有机气体 神经网络 定性识别 定量估计 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]