Research On Optimization Algorithm Of Electronic Nose Sensor Array Based On Feature Selection

The electronic nose is a bionic system for gas identification,which can acquire the response of target gas through sensor arrays and then analyze the composition or type of gas based on pattern recognition algorithms.The array optimization algorithm in the electronic nose system can filter more effective sensors from the original array to form a new array and improve the recognition accuracy of the electronic nose system for the target gas.Feature selection is an important array optimization algorithm that can measure sensor characteristics based on evaluation criteria to filter out more relevant and less redundant sensors and improve the performance of classification models.However,existing feature selection algorithms fail to optimize for the characteristics of e-nose system: when the e-nose system recognizes mixed gases,it is difficult for sensor features to distinguish target gases with the same composition;when the sensors in the array have the same redundancy,the existing feature selection algorithms cannot further measure the sensor features performance.Therefore,there are limitations when applying the existing feature selection algorithms to array optimization.This thesis introduces the algorithms related to array optimization and feature selection in detail,and proposes two mutual information feature selection algorithms based on sensor characteristics from two perspectives: improving sensor feature correlation and optimizing sensor feature redundancy.1.A mutual information feature selection algorithm based on the performance weights of the electronic nose system is proposed.The algorithm designs a feature differentiation evaluation function based on the characteristics of the electronic nose system for screening features that can distinguish similar labels and enhance the ability to identify mixed gas labels;then a feature redundancy evaluation function is used to calculate the redundancy degree between the candidate features and the selected features to screen out mutually independent features;finally,a feature sensitivity evaluation function is designed for screening features with higher dispersion of feature variables and greater Finally,the feature sensitivity evaluation function is designed to screen sensor features with higher dispersion of feature variables and greater differences in response to different gases.Experiments show that the algorithm can filter better sensor features in the e-nose related data set.2.The proposed algorithm for mutual information feature selection based on the performance of the electronic nose sensor.The algorithm screens out features that are more relevant to the label by calculating the mutual information of feature variables and label variables;then constructs a weight function based on the standard deviation of intra-class redundancy to screen out features with higher dispersion of intra-class redundancy;and finally screens out features that can better discriminate between different target gases by calculating the variance of feature variables.The feature subset is validated using the classification algorithm in the electronic nose related dataset,and the results show that the algorithm effectively improves the performance of the classification algorithm.