为提高轧制过程轧制力预报精度，建立了将轧制接触面积由几何关系确定，将影响因素复杂的轧制单位压力通过RBF神经网络预测模型。为适应工况的改变，提出了一种在线动态调整算法，利用新的测试数据对网络进行重新训练，使模型能够调整结构及网络参数，从而使最终设计的网络具有最佳结构。试验研究证明，所设计模型具有良好的适应能力，提高了轧制力的预报精度。

In order to improve rolling force prediction accuracy, a model was proposed with a geometrical relationship to determine roll contact area and RBF networks to predict unit rolling force which is affected by many factors. In order to suit the change of working conditions, an online dynamic adjustment algorithm is used to retrain new test data to adjust the configuration and parameters of RBF neural networks so that the optimal networks can be acquired finally. Experiment results indicate that the designed rolling force model has better adaptive capability and high prediction accuracy.