Research On Binary Gas Mixture Detection Method Based On Virtual Array Of Gas Sensors

With technological progress,many industries in our country,such as manufacturing,industry,etc.,are in the forefront of the world.The development of science and technology is inseparable from the use of energy.my country's coal reserves and mining volume rank first in the world,but high production is accompanied by high risks.In the process of coal mining,there are many flammable,explosive and poisonous underground mines.gas.Therefore,the detection of dangerous gases in the mine is extremely important.Due to the cross-sensitivity of a single sensor,the detection result is inaccurate.Therefore,this paper uses the sensor virtual array combined with the recognition algorithm to perform pattern recognition on the mixed gas of methane and hydrogen.In order to solve the problem of inaccurate detection of a single semiconductor gas sensor,this paper uses two different sensitivity sensors to form a sensor virtual array,builds an experimental system platform,analyzes the relevant performance of the sensor,and collects experimental data.For traditional sensor arrays,there are many types and numbers of sensors,causing problems of high cost and large size.In this paper,the operating temperature of the sensor is changed by changing the voltage,so that the sensor has different characteristics.Therefore,the same or even more data samples can be obtained by using a smaller number of sensor virtual arrays.In view of the low recognition accuracy of the sensor array's nonlinear response to the mixed gas,this paper uses the KPCA improved K neighbor algorithm to analyze the mixed gas composition.Using KPCA to extract the features of the mixed gas,and then use the K neighbor algorithm to have a better classification effect and anti-noise ability for the target to be classified in many cases,and to identify the components of the mixed gas.Through experiments,it is proved that KPCA's improved K neighbor algorithm can effectively classify the mixed gas,and the classification accuracy reaches 98.3%.Aiming at the problem of inaccurate recognition of mixed gas concentration,this paper proposes to use trainlm learning algorithm function to improve BP neural network to realize the concentration recognition of mixed gas.The traditional BP neural network can not only deal with nonlinear problems,but also includes strong self-learning ability and adaptability,generalization ability and fault tolerance.But there are also problems such as slow convergence,unstable results,and overfitting.In order to solve the problems of traditional BP neural network,this paper proves through simulation experiments that the improved BP neural network has a good effect on the recognition of mixed gas concentration.After the sample data is fused by the BP neural network,the maximum relative error is 10.3% and the minimum relative error 0.2%,compared with traditional sensor calibration methods,the improved BP neural network can effectively solve the problem of sensor cross-sensitivity,and is suitable for the detection and monitoring of target gases.