引入模拟辅助样本提高BP网络的泛化能力

以2－吲哚醇在20种不同参数固定液上的保留值、19种不同物质在同一固定液上的保留值以及脂肪胺的电谱保留值分别作为网络的训练样本和检验样本，建立了多元线性回归（LR）模型和BP网络模型，并基于LR模型运用随机搜索最优化方法，产生模拟辅助样本并将其引入BP网络训练样本集。预测结果表明，该方法的使用提高了BP网络的泛化能力，对于残缺样本问题的预测研究，提供了一种有效的方法，与线性回归模型及原BP网络模型相比，预测精度有了明显的改善。

Improvement of the Generalization of BP Networks with the Aid of Simulating Assistant Samples

Retention indices in different stationary phases, in different solutes and in fat amine are used as the training and proving samples of the networks. A method producing simulating assistant samples is presented by use of direct search procedure utilizing pseudo random numbers based on linear regression models, and the samples were combined with original samples of BP networks. The predicting results via different GC retention indices are given, it indicates that the method can improve the generalization of BP networks, and also it provides an effective method for predicting research of insufficiency samples. Predicting precision of the networks was improved obviously compared with LR model and basic BP networks model.