神经网络法在使用裂解气相色谱鉴别中草药中的应用

Identification of Chinese Herbal Medicine Using Artificial Neural Network in Pyrolysis-Gas Chromatography

将以误差反向传播为训练算法的前馈式人工神经网络（BP-ANN）首次用于中草药的裂解气相色谱谱图解析。重点考察了如何表征和提取复杂的裂解色谱图中有价值信息，用主成分分析方法处理后输入到参数经优化的神经网络中。实验证明，该方法不仅可以正确识别样品所属种类，而且对于不同实验时间、数据残缺等原因造成的噪音具有优异的抗干扰能力。

The potential utility of feed forward artificial neural network using the back propagation algorithm (BP-ANN), in interpreting pyrogram data from traditional Chinese medicine was discussed. We laid stress on how to extract and encode the most meaningful information from pyrogram to use as the input matrix in neural network, such as data representation and preprocessing. After network topology analysis,several parameters of neural network were optimized. The study revealed, that after training, utilizing principal component analysis (PCA) in conjunction with BP-ANN was robust in respect to small variances presented in data, such as noise and distortion.