Aiming at the problems that the adjustment parameters of rolling mill has a great influence on the quality of steel pipe in the two-roll skew rolling and punching production process of seamless steel pipes, and the accuracy of the set value calculated by the traditional mechanism formula is not high, a prediction model for adjustment parameters of skew rolling puncher based on deep neural network was proposed. Firstly, the traditional mathematical model of adjustment parameters was analyzed comprehensively, and the main influencing factors were determined on this basis. Then, based on the data set collected in the field, the deep neural network prediction model of rolling mill parameters during the two-roll skew rolling and punching was trained, and in the deep neural network training, the gradient estimation correction was conducted by using the combination of mini-batch gradient descent method and Adam algorithm to optimize the training speed. The simulation results show that the adjustment parameters of rolling mill predicted by the deep neural network model are compared with the measured data, the R-squared value of the prediction model is controlled at about 0.98, and the relative error of the adjustment parameters can be controlled within 5%. Compared with the traditional mathematical model, this prediction model has higher prediction accuracy, and can realize the high-precision prediction of rolling mill adjustment parameters and be used to guide production.