基于数据库的神经网络轧制力建模

Rolling Force Modeling of Neural Networks Based on Database

冷连轧过程控制的轧制力模型对于提高轧制精度和降低生产成本具有重要的意义,而传统的轧制力模型结构简单,精度较低,即使在实际生产中采用自适应技术,也无法满足高精度轧制的需要。为此针对5机架冷连轧机,提出并联结构的BP神经网络模型;采用Levenberg-Marquardt算法进行训练,确定网络的结构和参数;在数据库中建立钢种与神经网络的结构和参数一一对应的关系表,保存网络训练结果。对神经网络模型的仿真测试表明该神经网络轧制力模型有较强的泛化能力,收敛速度快,不易陷入局部最优,精度明显高于传统的轧制力模型。

Rolling force model of cold continuous process control has great significance in improving rolling precision and reducing producing cost. With the characteristic of simpleness and low-precision, conventional rolling force model can't meet the request of high-precision rolling, although it adopts self-adapting technology in practical producing. So a parallel BP neural network model has been proposed against five mills cold continuous rolling. In this model, Levenberg-Marquardt algorithm has been used to train and find the network's structure and parameters. In the meantime, relation table has been established in database, which can connect special type steel with nerve network's structure and parameters accordingly and save the result of network training. Simulation result shows that the model has the following merits: definite extend ability, high convergence velocity, not easy getting in local optimization, and higher precision.