The rolling force is the most active factor in the cold continuous rolling process, which seriously affects the accuracy of its shape and thickness control. In order to predict the rolling force with high accuracy, this paper establishes the deformation resistance model of DP590 dual-phase steel through rolling and tensile experiments, and obtains the deformation resistance of DP590 dual-phase steel from the cumulative reduction rate of each pass in the production history data. It is used as a rolling force feature input. After screening redundant features and irrelevant features, the three rolling force prediction methods based on random forest, XGboost and least squares linear regression are compared; the results show that the XGboost algorithm has R2 and mean square error Better, the rolling force prediction accuracy after optimizing through the learning curve is higher, R2 can reach 0.99, and the mean square error MSE reaches 121 kN.