Model Predictive Control is an effective control method for nonlinear plants with constraints. However, application to large-scale systems and/or processes that require high sampling rate is quite challenging because of the heavy computation. Here we consider MPC approach for the cooling process of hot-rolled strip having 3D thermal distribution. To solve this large-scale problem, we propose a method to change the resolution of the model dynamically according to the uniformity of the thermal distributions in the directions of interest. Combined with a hierarchical control strategy to flatten the thermal distribution in a certain direction, one can balance the computational load across the cooling process.