【摘要】 以白口铸铁力学性能与成分的关系为研究对象,以L9(34)型正交实验数据作为训练学习样本,以与正交实验成分有关的任意6个样本作为预测与评估样本,用学习率为0 5的一层10节点隐含层的BP神经网络进行了预测和评估,结果表明不加评估方案时,训练12913次,其最大误差不超过8%;在加评估方案时,训练了34919次,其最大误差为19 74%。因此,用正交实验测得的数据作为样本进行训练学习,可以对与正交实验成分有关的其余样本进行预测与评估,结果是令人满意的、现实的、可行的。人工神经网络与正交实验相结合,能大大节省时间和费用。更多还原

【Abstract】 The studied object was considered as the relation between the mechanical properties and the ingredients of the white cast and the trained samples considered as the data of the L9(34) orthogonal test. The random six samples related to the orthogonal test were forecast and assessed in the artificial neutral network of BP, in which the learning rate was 05 and an implicit layer contained ten nodes. The results were showed that, when the assessment was not discussed, the trained times were 12913 and the maximum forecasting error was not more than 8%; When the assessment was discussed, the trained times were 34919 and the maximum error was 1974%.Therefore if the samples from the data of the orthogonal test were trained by the artificial neutral network of BP, other samples of the ingredients related to the orthogonal test were better forecast and assessed, the result was satisfactory, practical and feasible. Combining the orthogonal test and the artificial neutral network, the time and cost was greatly saved.更多还原