Abstract:The drift is the inherent behavior of gas sensor, so it is more generality to reveal drift phenomena with no-load data. In order to remove the drift effectively, under the no-load condition, a drift removal method based on wavelet packet decomposition was proposed. Firstly, wavelet packet decomposition was employed to decompose the no-load data of the E-nose, and the approximation coefficient set of wavelet packet decomposition could be obtained. After the discrete analysis of the approximation coefficient set was carried out, a threshold function based on no-load data of the E-nose was constructed. And then the drift threshold function based on the sample data (loaded data) was obtained by extending the threshold function based on no-load data;furthermore, a drift elimination method for sample data was given. To test the effectiveness and practicability of the above method, it was applied to identify four kinds of white spirit samples by using the E-nose. The E-nose data of the four kinds of samples were divided into training set and test set according to the test time sequence, the identification results of linear Fisher discriminant analysis (FDA) indicated that the identification correction rates of training set and test set were all improved after their data were processed by the above drift removal method, and the minimum improvement was 23.65%, which showed that the method can effectively enhance the detection ability of the E-nose. At the same time, in order to further test the performance of the drift removal method, the nonlinear BP neural network was used to identify the four kinds of samples, and its identification results displayed that after treatment with the method, the identification correction rate of the training set was from 65.5% up to 100%, and the identification correction rate of the test set was also up to 97.5%. This not only showed that the identification of the four kinds of white spirit samples was a complicated nonlinear classification problem, but also showed that the proposed drift removal method was very effective. In addition, the drift removal method was proposed according to the no load data of the E-nose, thus it was considered to be general.
