电子鼻在检测食品等复杂样品时具有很大的潜力，但至今仍不具有长期稳健检测的能力。其主要原因是，漂移问题还没有得到彻底解决。另外，电子鼻测试信号的全面表征与可靠的非线性建模方法，对构建长期稳健的检测模型也至关重要。为此，本项目运用独立分量分析、多尺度小波包分解、能量分析等手段研究电子鼻漂移实时分离方法；运用变量可分性评价方法对电子鼻中每个传感器提取的不同特征进行鉴别能力评价，在此基础上，耦合多变量判别能力分析与多态检验分析，研究能够全面反映电子鼻测试信号的组合特征表征方法；以通过变换函数把非线性问题线性化为出发点，运用距离测度、相关系数等度量指标，构建能够评判空间变换优劣的准则函数，并借助于优化分析，研究变换函数中未知参数的确定方法，以获取最佳的线性分析空间，进而提出基于空间变换的非线性稳健建模方法。这些研究必将为电子鼻能够长期稳健检测食品等复杂样品奠定基础，并具有不失一般性的指导价值。

Electronic nose (E-nose) is a very promising tool to test volatile complex samples such as food. However, it has not the long term robust detection ability till date. The main reason is that the problem of E-nose drift has not been solved comprehensively and thoroughly. In addition, the comprehensive characterization method of E-nose response signals and the robust nonlinear modeling method are also very important for long term robust detection model of E-nose. Thus, in the project, the independent component analysis, multi-scale wavelet packet decomposition, energy analysis and others will be employed to investigate the real-time separation method of E-nose drift. Then, the separability analysis will be used to estimate the discriminant ability of every kind of feature variate corresponding to each gas sensor in E-nose, and on the basis of the estimation results, the characterization method of different combined feature variates is going to be explored by coupling multivariate discriminant ability analysis with multivariate hypothesis test. At the same time, to convert nonlinear problem into linear problem by transformation function with unknown parameter as a starting point, when the criterion function for evaluating the converting quality is established by some measuring indexs such as distance and correlation coefficient, the definition method of the unknown parameter will be researched with the help of optimization analysis so as to obtain the best linear analysis space. And then the nonlinear robust modeling method based on analysis space transformation will be proposed. These research findings will lay the solid foundation for the long term robust detection of food/complex samples using E-nose, and they may possess wide and general guidance value.