Digital twin-driven joint predictive and optimization scheduling method for parcel sorting systems

This study propose a joint predictive and optimization scheduling method based on a digital twin system to address challenges in dynamic scheduling for chute resources in sorting systems during an intensive operational transformation at a distribution center. First, a four-layer virtual-physical collaborative framework is constructed, establishing a closed-loop mechanism based on iterative cycles of perception, analysis, decision-making, and verification to overcome the limitations of traditional static scheduling in terms of adaptability. Subsequently, a Lagrangian relaxation-corrected long short-term memory prediction model is combined with an improved genetic optimization algorithm to achieve minute-level parcel flow forecasting and rapid rescheduling in response to disturbances. Empirical validation through simulations at an East China parcel processing center demonstrates significant performance advantages of the proposed method. At the algorithmic level, the Improved Genetic Algorithm outperforms other metaheuristics across various scales, achieving lower mean objective values and superior stability. At the system level, the JSMDTSS framework significantly improves operational efficiency compared to rule-based scheduling methods, reducing bagging completion time by 18.2%-20.2% and decreasing secondary sorting volume by 17.0%-44.4%. This approach provides an efficient technical pathway for intensive distribution operations. The findings serve as theoretical and engineering guidelines to enhance sorting efficiency and resource utilization in processing centers.