Iterative learning control (ILC) system is essentially a special feedback control system with two-dimensional (2D) dynamics that can be designed and optimized under the framework of 2D system theories. Motivated by this viewpoint, it is proposed in this paper to describe a batch process with 2D dynamics directly using a 2D controlled autoregressive moving average model, and then, design a 2D feedback controller, referred to as two-dimensional generalized predictive control, in the framework of model predictive control. The proposed design method naturally results in an ILC algorithm when the process is assumed as a one dimensional process performing a given task repetitively and guarantees the better control performance along cycle by utilizing the cycle-wise dynamics of the process. The proposed control scheme is the further generalization and extension of the two-dimensional generalized predictive iterative learning control scheme which has been developed in the previous works. It solves the problem in some degree that conventional ILC cannot guarantee the convergence when there are non-repeatable dynamics in the processes and/or in desired trajectories. The effectiveness and the applicability are illustrated by the comparisons of the simulation results and the experimental results on packing pressure control of the injection molding process.