Multiple Harmful Quantitative Detection Method Research Based On The Integrated Neural-network

Since the 19 th century, with the rapid development of the world industrial output, large amounts of energy are consumed no matter in the heavy industry or in light industry, such as the coal consumption at the beginning and oil consumption today. As a result, in this short 150 years, a large amount of toxic and harmful gases was produced, such as CO2 which causes the greenhouse effect, NOx which causes haze and photochemical smog, toxic gas CO and so on. The emissions of these poisonous and harmful gases not only caused great damage, but also produced a series of vicious incidents, such as fire, explosion, poisoning, etc. In the human survival environment, O2 as the basic nutrients for human life, is also one of the most important indicator of the environment. Therefore, monitoring of toxic and harmful gas in the environment becomes particularly important. Using the gas sensor to detect single gas is feasible, but the gases are usually mixed together in the daily life, so using gas sensor to detect them will result in the phenomenon of cross sensitivity, causing failing detection of multiple harmful gases adopting the traditional gas technology.Aiming at the defects of the current method to the detect poisonous and harmful gas, this paper constructed a set of multiple harmful gas detection system, which includes two parts of sensor array and pattern recognition. The organic combination of these two parts can effectively eliminate sensor’s cross-sensitivation. Sensor array of the test system can intake and record enough harmful gases concentration information, and then implement quantitative regression analysis through the pattern recognition part. Main contents include:1) construvt experiment device test system: preparing gas sensor array and designing the corresponding sensor preprocessing circuit; 2) simulation test: gas distribution scheme, Lab VIEW flow control software design, multi-channel signal acquisition system design based on Lab VIEW; 3) array signal preprocessing: complete array signal eigenvalue extraction using eigenvalue extraction method, and normalize the signal characteristic value; 4) pattern recognition algorithm design: establish an effective model of pattern by conbining the particle swarm optimization(pso) algorithm and neural network algorithm, then identify and quantitativly analysis the array signal; 5) algorithm improvement and comparison: the algorithm is improved and applied to the analysis of signal detection system for multiple harmful gases, and finally the classic pattern recognition algorithm performance comparison.The experimental results of this paper show that the sensor array technology can obtain the cross-sensitivity multivariate response signal, and pattern recognition technology can fully extract cross sensitive signal information, then complete the regression analysis of the atmosphere. The testing system accurately classifies the mixed gas of CO, CO2, NOx and O2 detected in this paper, and keep the average relative prediction error of quantitative analysis less than...