Research On Drift Suppression Of Electronic Nose Sensors Based On Domain Correction

The electronic nose(E-nose)is an artificial olfactory system that imitates the human sense of smell.When the gas to be detected enters the E-nose,sensor arrays generate sensor responses,and then analyze and process the sensor response signal with the intelligent signal processing method,and makes an odor prediction to complete the identification of the odor.However,in the actual application scenario,the E-nose is affected by sensor drift whose direction is uncertain and unpredictable.This will makes the data distribution of the previously collected data and the data collected later inconsistent,which casued the performance of the model trained with previous data egradation or even failure,and then the performance and service life of the E-nose system will be reduced.Therefore,this thesis mainly studies how to suppress sensor drift and improve the accuracy and robustness of the prediction model.In this thesis,the data without sensor drift is regarded as the source domain,and the data with sensor drift is treated as the target domain.In view of the drift problem faced by the electronic nose,this thesis proposes corresponding solutions.Details as follows:1.A cross-domain subspace alignment(CDSA)drift suppression model is proposed.The source domain and the target domain are simultaneously mapped to the same subspace through a linear mapping relationship P.In this subspace,the distance between the source domain and the target domain is minimized.The distinguishability between classes is guaranteed,and the local neighborhood structure is preserved to the greatest extent.Because the data distribution of the source domain and the target domain are inconsistent,the main idea of CDSA is to learn a new common subspace through a linear mapping,so that the data distribution consistency between the source domain and the target domain is improved,and the class separability is guaranteed.Strengthen to achieve drift suppression.2.Domain correction based on kernel transformation(DCKT)for drift suppression is proposed.The source domain and the target domain are mapped to a high-dimensional Hilbert space by a nonlinear transformation,in which the domain distance between the source domain and the target domain is minimized.At the same time,the data information of the original source domain and the target domain is preserved to the greatest extent.The mapping matrix P is then introduced to transform the empirical kernel mapping space into an m-dimensional subspace.In this subspace,the standard machine learning algorithm can be used to train the classification model with the source domain data,and then used for the prediction of the target domain.