基于时滞BP神经网络的铸坯表面温度预报研究

为了快速准确地计算连铸方坯表面温度,提出了基于时滞BP神经网络的温度预报模型.由于传统的基于机理的模型需要直接求解偏微分方程,导致模型的求解消耗时间过长.为了满足连铸动态控制与优化的需求,研究了连铸方坯的传热机理,根据连铸过程热传导的特点,建立了基于时滞BP神经网络的温度预报模型,以历史时刻的表面温度和当前时刻水量作为神经网络的输入,以下一个时刻表面温度作为神经网络的输出.该模型可以快速预报连铸方坯表面温度的动态变化情况.并且依据某工厂的真实数据进行实验,表面温度的预报最大相对误差小于0.1%,取得了较好的实验结果.该方法可以准确地预报连铸坯的表面温度.

On Prediction of Slab Surface Temperature Based on Time Delay BP Neural Network

Accuracy and rapid calculation of continuous casting square billet surface temperature of continuous casting production process for optimization and control is of great significance.The mechanism of heat transfer is essentially a nonlinear parabolic partial differential equation.Based on solidification heat transfer model of continuous casting square billet surface temperature calculation for a long time,it is difficult to meet the need of optimization and control.Based on Recurrent BP neural network to approximate two-dimensional heat transfer model for continuous casting,according to the characteristics of the continuous casting process of heat conduction,a Recurrent BP neural network has been set up,which can predict the dynamic change of continuous casting billet surface temperature rapidly.The actual data of some factory as the foundation,the surface temperature forecast the maximum relative error less than 0.1%,good results have been achieved.The method can fast accurate prediction of continuous casting billet surface temperature.