The simulation and optimization of industrial-scale simulated moving bed for para-xylene purification from a mixture of C-8 aromatics are presented. The separation process was modeled using true moving bed modeling strategy. The multi-objective teaching-learning-based optimization algorithm (MOTLBO) is improved by introducing alpha constrained technique, which is employed to optimize the yield of PX and consumption of desorbent. The improved MOTLBO has advantages in both convergence and distribution as compared to NSGA-II and MOTLBO. The optimized results suggest that the extract flowrate and consumption of desorbent should be increased, and simultaneously the step time and raffinate flowrate keep constant, so as to achieve a higher yield of PX.