基于模拟退火遗传神经网络的电子鼻已对二组分、三组分的混合气体模式进行了识别 ,识别精度及学习速度都较 BP神经网络大为提高 [1 ] ,但仍无法满足实用的要求。为了提高遗传神经网络对混合气体越限值的识别精度 ,本文在电子鼻已有的基础上提出分步分档识别法 ,在大范围内保证了识别准确性 ,提高了电子鼻的实用性。本文成功地将其用于四组分混合气体的精确识别。应用本方法的电子鼻既可用于正常环境气氛也可适用于危险气氛环境的气体模式识别。不同的档 ,其学习样本不同 ,识别精度不同。分档识别精度取决于学习样本的最小步长 ,最小步长越小 ,识别精度越高 Electronic nose based on simulated annealing genetic neural network has identified two-component and three-component mixed gas model, which has higher recognition accuracy and learning speed than BP neural network [1], but still can not meet the practical requirements . In order to improve the identification accuracy of genetic neural network to the limit value of mixed gas, this paper proposes step-by-step document recognition based on the existing electronic nose, which ensures the accuracy of recognition and improves the practicability of electronic nose. This article has successfully used it for accurate identification of four-component gas mixtures. The electronic nose to which this method is applied can be used both for the normal ambient atmosphere and for the gas pattern recognition in a dangerous atmosphere. Different files, their learning samples are different, recognition accuracy is different. The accuracy of sub-file recognition depends on the minimum sample size, the smaller the minimum size, the higher the recognition accuracy