针对1450HC轧机，利用大型非线性有限元软件MSC.Marc建立仿真模型，对多种轧制工况进行了模拟，得到了板凸度值。研究了不同板带参数、工艺参数、板形调控参数对轧后板凸度的影响规律。以有限元计算值为训练样本,利用BP神经网络强大的非线性映射功能，建立了板凸度预报模型，在训练过程中采用了改进的快速BP训练算法，从而提高了训练速度，加快了网络收敛速度，增加了算法的可行性。该网络模型解决了有限元计算时间长，难以在线应用的问题。

Using nonlinear elastic-plastic finite element method, a 3D FE simulation model of 1450 HC rolling process was developed with the nonlinear FE software MSC.Marc. Based on the model, a large amount of models which contain different rolling parameters was simulated and the strip crown was obtained. The effects of different rolling parameters, strip and rollers parameters on the strip crown were investigated and the main rules were mastered. The simulation results of the strip crown are served as the sample database of BP neural network by which the prediction model of cold rolled strip crown was built. During the training, an improved arithmetic was used in the BP neural network. Practice had proved: an improved arithmetic of BP neural networks improves the speed of learning and builds up the feasibility of arithmetic. The problem that the finite element method is time consuming and difficult to be used to the online flatness control is solved, and the precision of flatness online control is enhanced by this method.