基于人工神经网络的多气体分析系统研究与设计
摘要
传感器在信息系统中的重要性不言而喻，它特性的好坏、输出信息的可靠性对整个系统的质量至关重要。各行各业的自动化程度的迅速提高，特别是工业生产的自动化程度的提高，对传感器的性能提出了更高的要求。由于气体传感器自身的交叉敏感等物理缺陷，单一气体传感器无法对多种气体进行准确的定性识别和定量检测。随着智能理论和技术的发展，研究出用人工神经网络和气体传感器阵列组成的智能传感器，实现了多气体的智能测量。
随着石油化学工业的发展和人民生活水平的提高，易燃、易爆、有毒气体的种类在不断增加，其应用范围也在不断扩大。这些气体在生产、运输、使用过程中一旦发生泄漏，将会引发中毒、火灾甚至爆炸事故，严重危害人民的生命和财产安全。因此，开发可以进行多种气体分析的仪器设备具有很大的实用价值。本文的主要目的是以电子嗅觉系统原理为基础，研究出基于气体传感器阵列和人工神经网络的多气体分析系统，利用传感器阵列对各种气体的高维响应模式来实现对不同气体的识别，具有很强的选择性，解决了目前单个气体传感器选择性差的问题。
基于人工神经网络的多气体分析系统主要包括气体传感器阵列、数据采集模块和数据处理模块三大部分。本文的研究内容主要涉及到传感器、单片机、串口通信、数据预处理、BP算法及模式识别等相关技术。
气体传感器是气体传感器阵列的核心部件，是气体分析系统进行气体分析的关键组成单元，其选择性、重复性、交叉敏感性的指标直接影响到系统进行气体分析的质量。通过对气体传感器原理和MQ系列气体传感器阵列的介绍，为测量线路设计和数据采集分析提供了基本测量条件。
基于RS-232C标准的数据采集模块实现了从气体传感器模拟信号到计算机数字信号的转换以及数据的有序存储，为模式识别和仿真算法提供了标准化的数据集。
数据处理模块是系统完成气体分析的关键部分，其特性的好坏直接决定了气体检测精度。文中首先论述了多气体分析系统的理论模型，BP训练算法，以及使用MATLAB神经网络工具箱进行网络设计的方法和将其应用于多气体定性定量分析的实验步骤。然后将气体传感器阵列和模式识别技术相结合，进行3种气体的定性定量分析实验。其中涉及信号预处理算法、人工神经网络设计、进行多气体定性和定量分析的BP网络的输入输出和隐层单元确定方法、网络的训练及实验结果分析等。
通过实验验证，本文所设计的系统实现了对三种气体100%的定性识别和较低误差的定量检测。论证了采用气体传感器阵列与模式识别技术相结合的方法进行多气体分析的可行性。在分析系统误差产生原因的基础上，提出了改进系统识别精度的方法和建议，具有一定的工程应用价值。
It is went without saying that the sensor is very important to the information systems. The sensor’s characteristics and the reliability of the output are mostly significant to the whole system’s quality. The improvement of the automatization level in every field of life, especially the improvement of the automatization level in industry production requires higher performance of the sensor. Because of the gas sensor’s physical shortcoming such as its cross sensitivity, a single gas sensor can’t accomplish the multiple gases’ qualitative identification and quantitative detection. With the development of intelligent theory and technology ,intelligent gas consisting of gas sensor array and artificial neural network is excogitated, which achieve the intelligent detection of multiple gases.
Along with the development of petrochemistry industry and improvement of people’s standard of living, combustible, explosive and poisonous gases’ category and use sphere have an increase at the same time .Once these gases are leaked out when they were in the course of production, transportation and use ,will induce poisoning ,fire and explosion accident and jeopardize the security of people’s life and belongings. So, it is valuable to exploit such an instrument that can analyze multiple gases .In this paper ,based on the theory of electronic nose, we study a kind of multiple gases analysis system, consisting of gas sensor array and artificial neural network, which using gas sensor array’s multidimensional response mode for multiple gases to accomplish the identification of different gases .This multiple gases analysis system has more selectivity, and resolves the selective problem that the single gas sensor can’t resolve.
The multiple gases analysis system based on artificial neural network mainly consists of gas sensor array ,signal detection module and signal processing module three parts. This paper involves the technology of sensor, Single Chip Micyoco ,serial interface communication ,signal preprocessing ,algorithms of Back Propagation network and pattern recognition.
Gas sensor is the kernel of gas sensor array and the most significant component of multiple gases analysis system .Its level of selectivity, reiteration and cross sensitivity will affect the whole system’s quality directly .The introduce of gas sensor theory and MQ series gas sensor array provide the basic measure condition for design of detection circuit and signal processing.
The signal detection system based on RS-232C standard fulfills the diversion of gas sensors’ analog signals to computer’s digital signals and digital signals’ storage ,and provides the standard datasheet for pattern recognition and emulational algorithms.
The signal processing module is the most significant part of the system to complete the multiple gases’ analysis .Its characteristics mainly affect the precision of gases’ detection.
The theory model of the multiple gases analysis system, algorithms of Back Propagation network, design of ANN based on MATLAB neural network toolbox are discussed firstly .Then described the detection ways in multiple gases identification in order to validate the system’s feasibility .Qualitative and quantitative analysis experiments of three kinds of gases are completed with gas sensor array combined with pattern recognition. The experiment includes: signal preprocessing algorithms; ANN design, number determination of the input layer, output layer and the hidden layer of the BP network; train of the neural network and analysis of the results.
Through the experiment, the system designed achieved 100% in qualitative analysis of three gases and low error in quantitative analysis ,which validated the system’s feasibility. Deviation analysis and suggestions for improving the system’s performance have some application value for engineering.
随着石油化学工业的发展和人民生活水平的提高，易燃、易爆、有毒气体的种类在不断增加，其应用范围也在不断扩大。这些气体在生产、运输、使用过程中一旦发生泄漏，将会引发中毒、火灾甚至爆炸事故，严重危害人民的生命和财产安全。因此，开发可以进行多种气体分析的仪器设备具有很大的实用价值。本文的主要目的是以电子嗅觉系统原理为基础，研究出基于气体传感器阵列和人工神经网络的多气体分析系统，利用传感器阵列对各种气体的高维响应模式来实现对不同气体的识别，具有很强的选择性，解决了目前单个气体传感器选择性差的问题。
基于人工神经网络的多气体分析系统主要包括气体传感器阵列、数据采集模块和数据处理模块三大部分。本文的研究内容主要涉及到传感器、单片机、串口通信、数据预处理、BP算法及模式识别等相关技术。
气体传感器是气体传感器阵列的核心部件，是气体分析系统进行气体分析的关键组成单元，其选择性、重复性、交叉敏感性的指标直接影响到系统进行气体分析的质量。通过对气体传感器原理和MQ系列气体传感器阵列的介绍，为测量线路设计和数据采集分析提供了基本测量条件。
基于RS-232C标准的数据采集模块实现了从气体传感器模拟信号到计算机数字信号的转换以及数据的有序存储，为模式识别和仿真算法提供了标准化的数据集。
数据处理模块是系统完成气体分析的关键部分，其特性的好坏直接决定了气体检测精度。文中首先论述了多气体分析系统的理论模型，BP训练算法，以及使用MATLAB神经网络工具箱进行网络设计的方法和将其应用于多气体定性定量分析的实验步骤。然后将气体传感器阵列和模式识别技术相结合，进行3种气体的定性定量分析实验。其中涉及信号预处理算法、人工神经网络设计、进行多气体定性和定量分析的BP网络的输入输出和隐层单元确定方法、网络的训练及实验结果分析等。
通过实验验证，本文所设计的系统实现了对三种气体100%的定性识别和较低误差的定量检测。论证了采用气体传感器阵列与模式识别技术相结合的方法进行多气体分析的可行性。在分析系统误差产生原因的基础上，提出了改进系统识别精度的方法和建议，具有一定的工程应用价值。
It is went without saying that the sensor is very important to the information systems. The sensor’s characteristics and the reliability of the output are mostly significant to the whole system’s quality. The improvement of the automatization level in every field of life, especially the improvement of the automatization level in industry production requires higher performance of the sensor. Because of the gas sensor’s physical shortcoming such as its cross sensitivity, a single gas sensor can’t accomplish the multiple gases’ qualitative identification and quantitative detection. With the development of intelligent theory and technology ,intelligent gas consisting of gas sensor array and artificial neural network is excogitated, which achieve the intelligent detection of multiple gases.
Along with the development of petrochemistry industry and improvement of people’s standard of living, combustible, explosive and poisonous gases’ category and use sphere have an increase at the same time .Once these gases are leaked out when they were in the course of production, transportation and use ,will induce poisoning ,fire and explosion accident and jeopardize the security of people’s life and belongings. So, it is valuable to exploit such an instrument that can analyze multiple gases .In this paper ,based on the theory of electronic nose, we study a kind of multiple gases analysis system, consisting of gas sensor array and artificial neural network, which using gas sensor array’s multidimensional response mode for multiple gases to accomplish the identification of different gases .This multiple gases analysis system has more selectivity, and resolves the selective problem that the single gas sensor can’t resolve.
The multiple gases analysis system based on artificial neural network mainly consists of gas sensor array ,signal detection module and signal processing module three parts. This paper involves the technology of sensor, Single Chip Micyoco ,serial interface communication ,signal preprocessing ,algorithms of Back Propagation network and pattern recognition.
Gas sensor is the kernel of gas sensor array and the most significant component of multiple gases analysis system .Its level of selectivity, reiteration and cross sensitivity will affect the whole system’s quality directly .The introduce of gas sensor theory and MQ series gas sensor array provide the basic measure condition for design of detection circuit and signal processing.
The signal detection system based on RS-232C standard fulfills the diversion of gas sensors’ analog signals to computer’s digital signals and digital signals’ storage ,and provides the standard datasheet for pattern recognition and emulational algorithms.
The signal processing module is the most significant part of the system to complete the multiple gases’ analysis .Its characteristics mainly affect the precision of gases’ detection.
The theory model of the multiple gases analysis system, algorithms of Back Propagation network, design of ANN based on MATLAB neural network toolbox are discussed firstly .Then described the detection ways in multiple gases identification in order to validate the system’s feasibility .Qualitative and quantitative analysis experiments of three kinds of gases are completed with gas sensor array combined with pattern recognition. The experiment includes: signal preprocessing algorithms; ANN design, number determination of the input layer, output layer and the hidden layer of the BP network; train of the neural network and analysis of the results.
Through the experiment, the system designed achieved 100% in qualitative analysis of three gases and low error in quantitative analysis ,which validated the system’s feasibility. Deviation analysis and suggestions for improving the system’s performance have some application value for engineering.
引文
[1] 刘君华.智能传感器系统[M].西安:西安电子科技大学出版社，1999.371-372.
[2] 王 祁,聂 伟,王少军.人工神经网络智能气体传感器[J].传感器技术,1998，（5）:55-57.
[3] 彭备战，林德杰，欧金成.人工智能在传感器领域的应用.传感器技术，2002，（3）:5-7.
[4] 寇雪芹,师帅兵.人工神经网络在传感器数据融合中的应用.传感器技术.2002，（10）:49-51.
[5] 林康红,施惠昌,卢 强,奉 玲.基于神经网络的传感器非线性误差校正.传感器技术.2002（1）:42-43,47.
[6] 梁 坚,辛德瑗,陆毅新,沈 悦,胡 真.电子鼻—智能气体传感器.传感技术学报,1995,（2）:11-14.
[7] 王化祥,张淑英.传感器原理及应用.天津:天津大学出版社,1988.4-18.
[8] 焦李成.神经网络的应用与实现[M].西安:西安电子科技大学出版社,1993.26-45.
[9] 王 伟.人工神经网络原理[M].北京:北京航空航天大学出版社,1995.1-2.
[10] 刘同明,夏祖勋,解洪成.数据融合技术及其应用.北京:国防工业出版社,1998.
[11] 王 祁,聂 伟,张兆礼.数据融合与智能传感系统[J].传感器技术,1998,17（6）:51-53.
[12] 闻 新,周 露,王丹力,熊晓英.MATLAB神经网络应用设计.北京:科学出版社,2000.
[13] 刘迎春，叶湘滨.现代新型传感器原理与应用. 北京:国防工业出版社,1998.150-179.
[14] G.Faglia, F.Bicelli, G.Sberveglieri, P.Maffezzoni, and P.Gubian, “Identification and quantification of methane and ethyl alcohol in an environment at variable humidity by an hybrid array”, Sensors and Actuators B, vol. 44, 1997, pp. 517-520.
[15] 曲建岭,王磊,高峰.运用人工神经网络进行混合气体定量分析的研究.西北工业大学学报.2001,19（3）:399－402.
[16] 马常霞.基于传感器阵列与前馈神经网络的气体辨识系统.测控技术，2001,20(3)：18-19.
[17] 何芸.运用神经网络进行气体分析的研究.传感器技术.1995，
（3）:12-13.
[18] 马学童,马奎,王磊.多组分气体定量检测研究.仪器仪表学报,2001,22（4）:58-59.
[19] 毛六平，王小华，卢小勇. 微型计算机原理与接口技术.北京：清华大学出版社，北方交通大学出版社，2002.230-253.
[20] 曲建岭，人工嗅觉系统中的信息处理技术.西北工业大学博士学位论文，2000.
[21] 刘岚，秦洪.用于神经网络模式识别的一种改进的BP算法.信息技术，2002，（6）：6,7,10.
[22] 潘盛辉，韩峻峰.基于神经网络的模式识别及其应用.广西工学院学报，2000，11（4）：33-36.
[23] 罗中良，麦宜佳，余剑峰.基于多传感器的神经网络模式识别方法.中山大学学报，2001，40(5)：21-24.
[24] 飞思科技产品研发中心.MATLAB6.5辅助神经网络分析与设计.北京：电子工业出版社，2003.
[25] 飞思科技产品研发中心.MATLAB6.5应用接口编程.北京：电子工业出版社，2003.
[26] 闻新，周露，李翔，张宝伟.MATLAB神经网络仿真与应用.北京：科学出版社，2003.
[27] 刘崇进、郑大昉.气体传感器的发展概况和发展方向.计算机测量与控制，1999，7(2)：54-56.
[28] A.Szczurek，P.M.Szecowka，B.W.Licznerski，Application of sensor array and neural networks for quantification of organic solvent vapours in air，Sensors and Actuators B，vol.58,1999，pp.427-432.
[29] 焦李成.神经网络系统理论.西安电子科技大学出版社，1990.
[30] 高大启，杨根兴. 电子鼻技术新进展及其应用前景.传感器技术，2001，20(9)：1-4.
[2] 王 祁,聂 伟,王少军.人工神经网络智能气体传感器[J].传感器技术,1998，（5）:55-57.
[3] 彭备战，林德杰，欧金成.人工智能在传感器领域的应用.传感器技术，2002，（3）:5-7.
[4] 寇雪芹,师帅兵.人工神经网络在传感器数据融合中的应用.传感器技术.2002，（10）:49-51.
[5] 林康红,施惠昌,卢 强,奉 玲.基于神经网络的传感器非线性误差校正.传感器技术.2002（1）:42-43,47.
[6] 梁 坚,辛德瑗,陆毅新,沈 悦,胡 真.电子鼻—智能气体传感器.传感技术学报,1995,（2）:11-14.
[7] 王化祥,张淑英.传感器原理及应用.天津:天津大学出版社,1988.4-18.
[8] 焦李成.神经网络的应用与实现[M].西安:西安电子科技大学出版社,1993.26-45.
[9] 王 伟.人工神经网络原理[M].北京:北京航空航天大学出版社,1995.1-2.
[10] 刘同明,夏祖勋,解洪成.数据融合技术及其应用.北京:国防工业出版社,1998.
[11] 王 祁,聂 伟,张兆礼.数据融合与智能传感系统[J].传感器技术,1998,17（6）:51-53.
[12] 闻 新,周 露,王丹力,熊晓英.MATLAB神经网络应用设计.北京:科学出版社,2000.
[13] 刘迎春，叶湘滨.现代新型传感器原理与应用. 北京:国防工业出版社,1998.150-179.
[14] G.Faglia, F.Bicelli, G.Sberveglieri, P.Maffezzoni, and P.Gubian, “Identification and quantification of methane and ethyl alcohol in an environment at variable humidity by an hybrid array”, Sensors and Actuators B, vol. 44, 1997, pp. 517-520.
[15] 曲建岭,王磊,高峰.运用人工神经网络进行混合气体定量分析的研究.西北工业大学学报.2001,19（3）:399－402.
[16] 马常霞.基于传感器阵列与前馈神经网络的气体辨识系统.测控技术，2001,20(3)：18-19.
[17] 何芸.运用神经网络进行气体分析的研究.传感器技术.1995，
（3）:12-13.
[18] 马学童,马奎,王磊.多组分气体定量检测研究.仪器仪表学报,2001,22（4）:58-59.
[19] 毛六平，王小华，卢小勇. 微型计算机原理与接口技术.北京：清华大学出版社，北方交通大学出版社，2002.230-253.
[20] 曲建岭，人工嗅觉系统中的信息处理技术.西北工业大学博士学位论文，2000.
[21] 刘岚，秦洪.用于神经网络模式识别的一种改进的BP算法.信息技术，2002，（6）：6,7,10.
[22] 潘盛辉，韩峻峰.基于神经网络的模式识别及其应用.广西工学院学报，2000，11（4）：33-36.
[23] 罗中良，麦宜佳，余剑峰.基于多传感器的神经网络模式识别方法.中山大学学报，2001，40(5)：21-24.
[24] 飞思科技产品研发中心.MATLAB6.5辅助神经网络分析与设计.北京：电子工业出版社，2003.
[25] 飞思科技产品研发中心.MATLAB6.5应用接口编程.北京：电子工业出版社，2003.
[26] 闻新，周露，李翔，张宝伟.MATLAB神经网络仿真与应用.北京：科学出版社，2003.
[27] 刘崇进、郑大昉.气体传感器的发展概况和发展方向.计算机测量与控制，1999，7(2)：54-56.
[28] A.Szczurek，P.M.Szecowka，B.W.Licznerski，Application of sensor array and neural networks for quantification of organic solvent vapours in air，Sensors and Actuators B，vol.58,1999，pp.427-432.
[29] 焦李成.神经网络系统理论.西安电子科技大学出版社，1990.
[30] 高大启，杨根兴. 电子鼻技术新进展及其应用前景.传感器技术，2001，20(9)：1-4.