Research On VOCs Mixed Gas Detection Based On BP Neural Network

In order to estimate the concentrations of volatile organic compounds(VOCs)in mixtures,electronic nose which constitutes by sensor array and pattern recognition technology is applied to explore the problem.The sensor array consists of five metal oxide sensors made in the laboratory,which can form a complete response pattern for VOC mixed gases.Response sets about the sensor array in different VOC mixed gas are derived from actual experimental tests.In the experiment,the sensor test system is set up.The target analyte is composed of four kinds of typical VOC gas of ethanol,acetone,formaldehyde and toluene,and in order to develop the electronic nose for practical applications,the concentrations of VOCs and their combinations are randomly assigned in the mixture.In this paper,the back-propagation(BP)neural network is used to analyze and identify the sensor array signal,to complete the estimation of VOC mixture about components and concentration.The BP neural network is created in MATLAB.The first thing to do is to normalize the data preprocessing to prevent errors due to the number of levels.Then,the network prediction performance is explored which in the influence of structural parameters in the BP neural network such as the number of neurons,the activation function and the performance target and the optimal structural parameters suitable for the problem are debugged.The experimental results show that the BP neural network output node can give continuous prediction on the concentrations of each VOC in the mixture,and it can complete the quantitative analysis of the VOC mixed gas component within a certain error range.To improve the prediction accuracy of the system,the method of pattern recognition is improved based on BP neural network with two hidden layers and decision tree learning.First,the data set of the VOCs mixture is classified to sub-classes based on the concentration with the help of classification and regression tree.Second,every sub-class is classified and regressed by a corresponding BP neural network,which is in the appropriate structural parameters about sub-class samples.The experimental results show that the maximum error of the improved model is about 2 ppm in the concentration estimation for each VOC,and its accuracy is better than the result obtained from the single BP neural network.Moreover,the relative error is less than 5% when the predicted concentration is higher than 20 ppm.This study reports some aspects of the potential for using neural networks for quantitatively analyzing the concentrations of VOCs mixture.This study shows that the neural network has the potential to quantitatively analyze the concentration of VOCs mixtures and can be developed for the basis of electronic nose products for identifying the VOC mixture.