For the problem that the physics model is difficult to describe complex finishing mill process for hot rolling production, the rolling force data model of hot rolling is build by using the data subspace-based partial least squares method. Then it is used as part of the rolling force optimization model, and the modified particle swarm optimization intelligent optimization algorithm solves the problem. By the computational analysis, the rolling force data model established by the data-driven approach is able to reveal the physics laws of the rolling force, which can be used in the actual system instead of the physics model. By solving the overall optimization model, the quality of the hot rolling finishing products is improved and energy consumption is reduced. The data-driven based modeling and optimization method has great value in the actual production process.