摘要
基于模糊数学理论，提出了一个从神经网络训练样本中侦破矛盾样本的方法，建立了其数学模型；并用此方法对从工厂取得的测量数据样本进行了检查，找到了对训练及检验结果影响很大的矛盾样本，并对剔除矛盾样本前后建立的模型进行了对比分析，结果表明用此方法侦破出矛盾样本并加以剔除后可大大改善训练及检验的结果，由此建立的模型更为准确．
A fuzzy mathematical method is presented in this paper, by which the conflicting samples used for training artificial neural networks are detected according to the principle that the outputs of samples should be also similar if their inputs are similar. A mathematical model is constructed. Measurement samples data from the plant are tested by this method. The conflicting samples which greatly affected the training and test results are detected. The neural network models, which are constructed before and after the conflicting samples are deleted, are comparatively analyzed. The results show that the model constructed after deleting the conflicting samples is more correct and this method can greatly improve the training and test results.
出处
《计算机与应用化学》
CAS
CSCD
1997年第3期181-184,共4页
Computers and Applied Chemistry
关键词
神经网络
矛盾样本
模糊数学
Artificial neural network, Conflicting samples, Fuzzy mathematics
