轧制力预测中RBF神经网络的组合应用

 传统的数学模型无法达到冷连轧控制的尺寸精度要求。针对传统轧制力模型的固有缺陷，为提高冷连轧机组轧制力计算精度，合理选择、更新和预处理训练样本，采用RBF神经网络预测冷轧带钢屈服应力并把它用于传统轧制力计算模型，获得较高的轧制力预测精度。而后使用RBF长期数据修正网络和RBF短期数据修正网络得到长期数据修正网络和短期数据修正网络的修正系数，对轧制力计算值进一步修正，从而进一步提高轧制力预报精度。上述方法直接用于某冷连轧机组，轧制力预测误差在±6%之内。这充分证明RBF网络可以成功用于轧制过程控制并满足实际生产的需要。

Application of Combination of RBF Neural Network to Prediction of Roiling Force

Traditional mathematic model can not meet the requirement of dimension precision that is needed for tandem cold mill control. In view of intrinsic imperfection of traditional models of rolling force, in order to improve the rolling force calculation precision of tandem cold mill, choose, update and pretreat training patterns reasonably, RBF neural network is used here to predict yield stress of cold strip and this yield stress is used in traditional rolling force calculation model, as a result, the good precision of rolling force forecasting is received; and then the RBF long time data correcting network and the RBF short time data correcting network are used to get the correction factors of rolling force which belongs to these two networks for the purpose of modifying calculated value of rolling force, as a result, the rolling force forecasting precision is improved further. The method above is used in certain tandem cold mill directly and the final error of rolling force forecasting is within 〖JP2〗±6%. This result proved that RBF neural network can be used in rolling process control successfully and has the ability to satisfy production.