基于神经网络与线性回归的轧制力预报

Rolling Force Forecast Based on Neural Network and Linear Regression

通过将SIMS轧制力计算公式进行相应简约化处理,避免了模型软件在该公式计算时的重复迭代求解,缩短了计算时间,因此更适合在线软件计算。利用现场实际生产数据反向回归出变形抗力模型中的系数,提高了模型中系数的准确性。用神经网络对变形抗力与应力状态系数的乘积加以修正,进一步提高了轧制力预报的精度。预测结果与实测数据比较表明,轧制力预报误差基本在±5%以内,满足了轧制力预报的精度要求。

The formula deduction procedures of the SIMS rolling force formula are presented in order to avoid the repeated iteration in the calculation software, and the calculation time is reduced, which makes it more suitable for online software calculation. The actual production data reverse regression coefficients in the model of deformation resistance model were used to improve the model coefficient accuracy. The neural network used to amend the prod- uct of the coefficient of deformation resistance and stress state, thus the accuracy of the rolling force prediction was further improved. Predicted and measured data show that the rolling force prediction error was within ±5 % and could meet the forecast accuracy requirements of the rolling force.