基于BP网络混合气体浓度识别

【Abstract】wWw.shuoshilunwen.com At present, China’s economic is rapidly developing, the people are building a harmonious socialist society, however, frequent mining accidents he brought untold suffering to people, Coal mine safety, especially the gas explosion problem has become the social focus. This thesis designs the combustible gas concentration recognition system, the system can identify concentration of the three coal mine flammable gases:CH4, H2S, CO, It he some practical significance for the effective prevention of gas explosions.In this thesis, the system is designed from four aspects:the semiconductor sensor selection, gas distribution systems, data acquisition, BP network design and algorithm improvements.First,this paper introduces the background and main research contents. chapterⅡintroduces gas sensors and reason of sensor choice, combined the actual situation in the mine, choose temperature sensor modules,humidity sensor modules and gas sensors which is sensitive for CH4, H2S, CO.these sensors form sensor array. ChapterⅢdescribes the static gas distribution and dynamic gas distribution, and design the gas distribution system from gas gathering, experimental apparatus, experimental theory and experimental methods combined with the actual situation. ChapterⅣintroduce the data acquisition system, this chapter first outlines the data collection principles and features of current data collection system, selected the Advantech PCI-1710L data acquisition card. We design a signal conditioning circuit, complete the data acquisition hardware design. Then we introduces the VC++6.0 development environment, and gives the procedures for the preparation of data collection. ChapterⅤdescribes the development of artificial neural networks and the main stream of network structure, neuronal function, neural network learning, BP network structure and basic algorithm, point out its deficiencies department. With the need for mixed gas concentration recognition, we design structure of BP network and performance function, the basic algorithm has been improved and optimized:an increase of weight change in momentum items and variable learning speed which speed up the nerve network convergence. This chapter also gives the neural network learning process flow chart for the mixed-gas concentrations recognition and the neural network training program code and system identification results:when the concentration of flammable gas (CH4, H2S, CO) is 10PPM-500PPM, the trained BP network can identify concentration with acceptable error range, reach the purpose that gas concentration is identified, we analyze and evaluate the identification results and error. Finally, this thesis summarizes the work which was done, and put forward future further research work.