特征选择在电子鼻系统阵列优化中的应用

阵列优化是优化电子鼻系统性能的重要方法之一,其通过特征选择来确定阵列中合适的传感器数目和种类,进而优化传感器阵列。特征选择一般有搜索性和非搜索性两类方法。实验采用广义顺序前进算法、标准遗传算法、模拟退火算法和随机搜索算法4种搜索性特征选择算法,实现了可燃性液体检测实验中的阵列优化。比较了这4种算法的搜索策略,同时,还对搜索性特征选择算法中常用的5种类别可分性准则进行了比较。结果表明:广义顺序前进算法在本实验条件下具有较优的搜索效率;而基于类内类间距离的准则J3更能准确反映特征集的类别可分离性。

Application of feature selection in sensor array optimization of E-nose

Sensor array optimization is one of the methods to enhance the performance of electronic noses,which uses the feature selection to decide the right kinds and number of gas sensors.Generally there are two different methods for feature selection,searching method and non-searching method.Four searching strategies are adopted in a flammable liquids detection experiment,which are genetic algorithm(GA),simulated-annealing algorithm,generalized sequential forward selection and random searching algorithm,and their performances are compared.Five kinds of class separative criterion are compared here as well.Results show that generalized sequential forward selection possesses excellent optimization property and spended the least time in this experiment,thus having higher efficiency than other three methods.And the criterion J3 based on class distance can estimate the class separability relatively appropriately.