基于卷积神经网络的热轧带钢力学性能预报

Mechanical property prediction of hot-rolled strip based on convolutional neural network

为解决热轧带钢力学性能预报精度的问题,本文提出了一种将一维数值型数据转换为二维图像型数据的建模方法,基于LeNet-5和GoogLeNet卷积神经网络,构建了一种新型的热轧带钢力学性能预报模型,并利用实际生产数据对模型的适用性进行了测试。结果表明,所建模型的抗拉强度预报误差为2.49%,均方根误差为19.15 MPa,预测精度高于BP神经网络和单独的LeNet-5和GoogleNet卷积神经网络模型,所建模型的有效性和准确性均得到了验证。

In order to improve the prediction accuracy of mechanical properties of hot-rolled strip,a modeling method for transforming one-dimensional numerical data into two-dimensional image data was proposed.Based on convolutional neural network LeNet-5 and GoogLeNet,a new type of model for mechanical property predition of hot-rolled strip was established,and its applicability was tested by employing the actual production data.The results reveal that the T s prediction error of the new model is 2.49%and the root mean square error is 19.15 MPa,which is more accurate than that of BP,LeNet-5 and GoogLeNet nerual network models.Therefore,the validity and accuracy of the established model have been verified.