论文部分内容阅读
用7个压电晶体组成传感器阵列，每个晶体上分别涂有不同种类的气相色谱固定液，通过测定各种可燃物质燃烧时放出的混合气体来识别所燃物质，在识别中分别应用了人工神经网络法（ANN）和逐步判别分析法（SDA）．讨论了解决神经网络开始训练时不收敛或产生麻痹现象的方法，提出了训练数据选取的新方法─—训练集逐步扩展法．实验证明：人工神经网络对被测物质的识别准确率达100％，高于逐步判别分析法（83％）．
The sensor array is composed of seven piezoelectric crystals. Each crystal is coated with different kinds of GC fixatives. The combustible substances are identified by measuring the mixed gas released during the combustion of various flammable substances. In the identification, artificial sensors Neural Network (ANN) and Step by Step Discriminant Analysis (SDA). The method to solve the problem that the neural network does not converge or produce paralysis when it starts training is proposed. A new method of training data selection - training set stepwise expansion method is proposed. Experiments show that the accuracy of artificial neural network in the identification of tested substances is 100%, higher than that of stepwise discriminant analysis (83%).