Abstract: The method combining mathematical model and BP neural network was used to predict rolling force. Unlike most neural networks which only selected rolling variables as input variables, the BP neural network added lubricating oil injection quantity and calculated data of rolling force model into the input variables to reflect the impact of friction on rolling force and avoid large rolling force prediction deviation, so the 11×7×1 network structure was established, then the CSP line rolling force prediction model was formed in combination with rolling force model. The results show that the average relative error between the predicted data of the neural network model and the average measured data was only 1.08％, while the average relative error of rolling force model was 6.32%. So it can be concluded that the neural network model had a good ability in tracing average measured rolling force and high value in engineering application.