Iterative-learning-model predictive control (ILMPC) has been considered a potential control strategy for batch processes. ILMPC can converge to the desired reference trajectory with high precision along cycles and reject real-time disturbances within cycles. However, as a model-based control method, the control performance of the ILMPC algorithm deteriorates when it suffers the problem of model-plant mismatches. Therefore, simultaneously guaranteeing system convergence along the cycle-axis and robust stability on the time-axis is a challenging task in the framework designing of ILMPC control systems. To this end, we present a type of tube-based ILMPC strategy based on an empirical model for nonlinear batch processes. First, to describe the dynamic behavior of nonlinear batch processes as much as possible with the goal of low computational cost and high modeling accuracy, a pre-clustered just-in-time learning (JITL) model focused on operation data is developed. Then, an ancillary MPC controller is combined with the ILMPC algorithm to form a tube scheme to relieve the impact of model error. In addition, we construct an inverse model system (IMS) to systematically determine the first cycle control input trajectory of the proposed tube-based ILMPC algorithm. As a result, they simultaneously ensure the system convergence along the cycle-axis and robust stability on the time-axis. Still, they can improve convergence speed and tracking performance. Finally, comparative simulations demonstrate the superiority of the proposed control algorithm.