Research On The Detection Technology Of Toxic And Harmful Gases Based On Array Sensors

In the public transportation environment,people are exposed to potential dangers caused by toxic and harmful gases,so it is necessary for low-concentration gas detection.Generally,the gas color phase spectrometer is selected as gas detecting instrument,which is not suitable for gas detection of public transportation due to its large volume,high cost and only single gas could be detected.Semiconductor gas sensors are low in cost and small in size but greatly reduce detection accuracy due to cross-sensitivity when detecting multiple gases.Therefore,in the case of low concentration mixed gas and cross-sensitivity of semiconductor gas sensor,the hardware system of mixed gas is built,and a pattern recognition model is designed and constructed to achieve qualitative identification and quantitative identification of mixed gas with high precision.The error feedback neural network(BP)is used to complete the qualitative identification and quantitative identification of the mixed gas.The BP neural network model reduces the output error by changing the weight and threshold through the continuous back propagation of the error,but the continuous error back propagation is easy to fall into local convergence so that the output accuracy cannot improved and the calculation speed is slow.The artificial bee colony algorithm(ABC)is introduced to optimize the BP neural network.The ABC algorithm has strong global search ability and fast convergence speed,it can effectively improve the overall performance of the BP neural network,however,the good solutions of the ABC algorithm is easy to lose in the optimization process.The optimization mechanism of ABC algorithm is improved to obtain adaptive artificial bee colony algorithm(AABC).The improved AABC algorithm can further improve the overall performance of BP neural network model.Two common used BP neural network optimization algorithms,chaotic particle swarm optimization(CPSO)and fruit fly algorithm(FOA)are used to compare the AABC and ABC to verify whether ABC is optimized.By using the fuzzy C-means algorithm(FCM)to preserve the characteristics of the eigenvalues,the number of training samples is optimized and the number of samples is reduced under the condition of ensuring accuracy as much as possible to reduce the computation time of the model.The hardware system builds a mixed gas detection platform with MQ-3(alcohol sensor),MQ-4(methane sensor),MQ7(carbon monoxide sensor),MQ-8(hydrogen sensor)and uses STM32F103VET6 as the core of detection platform,and the system start-up initialization and data transfer of hardware are completed through STM32 library function and LabVIEW.The model establishment of BP neural network is completed by using python and its extension library and tensorflow.The comparison experiments on the optimization ability of AABC,ABC,CPSO and FOA algorithms are completed by five different test functions.The experimental results show that the global search ability and the optimal value of AABC algorithm are better than other optimization algorithms in the optimization process,and it also proves that the improvements to the ABC algorithm are effective.Through the training samples,the parameters optimization of BP neural network in qualitative identification and quantitative identification is completed,and the selection of the optimal number of training samples is completed.The gas test results show that the AABC-BP hybrid algorithm model has the fastest calculation speed and the highest accuracy in qualitative identification and quantitative identification.The test accuracy in qualitative identification is 96.42%,and the average test error in quantitative identification is 5.62%,the accuracy and error meets the design requirements and has a certain practical value.