机械拉杆锻压工艺的神经网络优化研究

Study on Mechanical Tie Rod Forging Process Optimized by Neural Network

采用不同的隐含层传递函数构建了4×28×1三层拓扑结构的机械拉杆锻压工艺神经网络优化模型,并进行了学习训练和预测验证。结果表明：与logsig函数相比,以tansig函数作为隐含层传递函数的模型平均相对训练误差从5.4%减小至2.5%。以tansig函数作为隐含层传递函数的模型平均相对预测误差为2.4%,具有较强的预测能力和较高的预测精度。

The neural network optimization model（three layers topology structure of 4×28×1） of the mechanical tie rod forging process was built by using different the hidden layer transfer functions, and learning training and prediction verification were carried out. The results show that compared with the logsig function, the average training relative error of the model with the tansig function as the hidden layer transfer function is reduced from 5.4% to 2.5%. The average relative prediction error of the model with tansig function as the hidden layer transfer function is 2.4%, which has strong prediction ability and high prediction accuracy.