Electronic nose drift general calibration method based on convex set projection and extreme learning machine

Abstract

The invention relates to the technical field of electronic nose calibration, and particularly discloses an electronic nose drift general calibration method based on convex set projection and an extreme learning machine. From the perspective of machine learning, a constraint network net1 and a calibration network net2 are established based on the extreme learning machine, the constraint network net1 is trained by using an electronic nose in a feature data set X without drift, network parameters are stored, then a drifting feature data set Xd is taken as an input of a constraint network net1, iterative adjustment is performed on the input Xd of the net1 network based on convex set projection to obtain a calibrated sensor feature data set Xc, and then the feature data set Xc is taken as a label of a calibration network net2; and the feature data set Xd is input for common training so as to calibrate an unknown gas response signal so that the tolerance performance of gas identification after drifting of the electronic nose can be improved, and a network obtained by training can achieve a drifting compensation effect on an unknown gas sample. Therefore, the gas recognition precision of the electronic nose after other gas sensors drift is improved.

Description

technical field [0001] The invention relates to the technical field of electronic nose calibration, in particular to a general calibration method for electronic nose drift based on convex set projection and extreme learning machine. Background technique [0002] An electronic nose is an intelligent electronic device or artificial olfactory system that uses the response map of a gas sensor array to identify gases. Due to the crossover characteristics and broad spectrum of gas sensor arrays in electronic noses, the gas recognition capabilities of electronic noses are widely used in medical diagnosis, tea quality assessment, environmental detection, and gas concentration prediction. [0003] However, the gas sensor of the electronic nose is aging continuously with the increase of usage time, which greatly shortens the service life of the gas sensor array of the electronic nose. Poisoning, aging or environmental variables can cause the gas sensor drift of the electronic nose, a...

AI Technical Summary

Problems solved by technology

[0008] The typical multivariate component calibration principal component analysis method uses principal component analysis to find the drift direction, thereby removing the drift component; however, the compensation idea of ​​the component calibration principal component analysis method needs to be based on the same drift direction of all categories of data However, the actual drift of the electronic nose is not the case, so it is difficult to effectively apply this method to the drift compensation of the electronic nose; and if a multiplier calibration variable is added on the basis of the component calibration principal component analysis method to improve the The limitation of data drift direction consistency will bring about the generalization of its drift compensation, which will be restricted by the nonlinear dynamic characteristics of gas sensors in online applications, making its drift compensation effect difficult for different gas identification applications.

[0009] The adjustment and compensation method is to adjust the difference in the characteristic distribution of the gas sensor array of the electronic nose during gas recognition and detection at different stages, and then realize drift calibration; When the gas sensor array has a transient response, it is misjudged that the gas sensor array is undergoing drastic changes in drift, and then frequently adjusted and calibrated, it is easy to interrupt the original eigenvalue distribution of the electronic nose gas sensor array, resulting in a more accurate After drift calibration, the identification neural network cannot correctly identify the matching gas, which affects the gas identification accuracy of the electronic nose.

[0010] Previously, researchers have also carried out some research on drift calibration of electronic noses through machine learning methods, but the current machine learning methods are mainly based on support vector machines, which often require a large number of training samples for learning. In the case of limited training samples, the calibration effect is not good, and it is still not good to improve the gas recognition accuracy of the electronic nose through drift calibration. In addition, this type of machine learning method usually needs to train many base classifiers. Therefore, its domain transfer ability and general flexibility is restricted

[0011] To sum up, the existing calibration methods for electronic nose gas sensor drift generally have the problems of low gas recognition accuracy and poor generalization of the electronic nose after calibration.

Images