论文部分内容阅读
采用贝叶斯统计学原理改进传统神经网络算法,通过在神经网络的目标函数中引入了表示网络结构复杂性的约束项,避免了网络的过拟合以提高网络的泛化能力。将改进的神经网络应用于某钢铁公司1700 mm热连轧机带钢轧制力中,其预报精度、训练时间和网络稳定性均优于传统神经网络预测,其改进传统的轧制力预报方式,从而进一步提高轧制力预报精度和辊缝设定精度,以期进一步带钢厚度质量。
The Bayesian statistical theory is used to improve the traditional neural network algorithm. By introducing a constraint term that represents the complexity of the network structure in the objective function of the neural network, the network over-fitting is avoided to improve the generalization ability of the network. The improved neural network is applied to strip rolling force of a 1700 mm hot strip mill in a steel company. The forecasting accuracy, training time and network stability are all better than the traditional neural network prediction. It improves the traditional method of forecasting rolling force, So as to further improve the prediction accuracy of rolling force and the setting accuracy of roll gap, in order to further improve the quality of strip thickness.