Hits:

Date of Publication:2013-01-01

Journal:传感技术学报

Affiliation of Author(s):电子信息与电气工程学部

Issue:12

Page Number:1649-1654

ISSN No.:1004-1699

Abstract:Sensor response drift remains to be the most challenging problem in gas sensing. We proposed a novel ensemble method with dynamic weights to solve the gas discrimination problem regardless of their concentration with high accuracy over extended periods of time. The method uses a dynamic weighted combination of classifiers trained at different points of time. Their weights in testing future datasets are predicted by fitting functions which are obtained by proper fitting of optimal weights in training. We compared the performances of the proposed method and competing methods in experiment based on the public dataset over a period of three years. As results illustrate,the proposed method performs better than others. Furthermore,the method can be further optimized by applying a fitting function that is better match variation of the optimal weight over time.

Note:新增回溯数据