带钢平整张力神经网络预设定模型研究

Neural networks based preset model of strip tension for CAPL temper mill

针对某冷轧厂1 730mm连退机组六辊CVC平整机组出现张力设定表数值不准确或缺失的问题,建立了基于BP神经网络的平整张力预设定模型,对平整机张力设定表进行了完善并实现了张力设定值的在线预报。在线运行结果表明:张力预报值与实测值的相对误差在±8%以内,实现了平整机组张力的高精度预报。进一步离线测试表明:采用本文所建立的平整张力神经网络预设定模型后,平整机出口成品板形质量得到较大提升,平直度横向分布均值降低约78%,满足生产要求,并提高了轧制的稳定性。

For a six roller CAPL temper mill of a certain steel company, some tensions tabulated in the preset tables were inaccura- cy or incomplete,which brings much trouble into the production line of strips. So a neural networks-based preset model for the tensions of the temper mill was proposed. This model could be used to fill vacancies of the preset tension tables and further pre dict appropriate values for the online tensions control. Experiments showed that the relative errors between this predicted tensions and the measured values were less than ~8 %. Furthermore,offline tests on the productions showed that the transverse distribution averages of their flatness reduced almost 78%. This work could not only improve the quality of the strip steels under CAPL temper mill but also stabilize the rolling process of the production line.