Gas Sensor Drift Compensation Based On Stacked Autoencoder

In the current society,gas sensor detection technology is widely used,but the drift problem of gas sensor will cause the detected data to be non-linear and chaotic,which will cause the classification performance of model to degrade severely after a period of time.It increases the system difficulty and maintenance cost in practical engineering.The drift of gas sensor Therefore,it's one of the urgent problems to find a method to compensate the drift of gas sensor in the current sensor research field.Automatic encoder is one of the several typical models in deep learning,which has a strong nonlinear fitting capability to extract the deep characteristics of the tested samples,which is more conducive to classification or regression prediction.So this paper proposed a gas sensor drift compensation method based on the auto-encoder.Specifcally,this paper mainly studied the following three aspects:?In this paper,it proposed a method to extract the characteristics of drifted feature values by stacking auto-encoders,and then use Support Vector Machine and K-nearest Neighbor Analysis algorithm to identify the gas.?Using kernel principal component analysis to extract the secondary features of the eigenvalues extracted by the stacked auto-encoder,then using K-nearest Neighbor Analysis to identify.?In this paper,a kind of classifier fusion method is proposed,which combines the cascading generalization method and the cascade method to fuse four base classifiers with different classification effects.Then gas identification is performed using fusion model.This paper used the auto-encoder network to study the deep connection between gas sensor drift data.And through the double feature extraction and classifier fusion methods,the drift compensation performance of the stacked auto-encoders network is further optimized respectively from the characteristics layer and the decision-making layer.Through experimental comparison,this paper verifies that the method proposed in this paper can effectively inhibit the expression of irrelevant characteristic values(i.e.,drift signals)and improve the accuracy of classification,so as to realize the compensation of gas sensor drift.