Decision Making Model for Frozen Distribution using Cross-dock

Minh Nhat Vu, Hoai-Phuc Tran, Huu‐Thanh Nguyen, Mai-Ha PhanOperations 2025 · DOI

TL;DR

This paper introduces a decision-making model for frozen product distribution using a cross-dock system. It optimizes vehicle allocation to minimize transportation costs, penalties for delivery time deviations, and penalties for quantity inaccuracies, crucial for time-sensitive frozen goods. The model is validated with CPLEX and enhanced with Simulated Annealing for efficiency on larger datasets.

What Problem Does This Paper Address?

The paper tackles the complex challenges of distributing frozen goods, particularly within a cross-docking framework. A key issue is the mismatch between received and demanded quantities at the cross-dock, leading to potential surpluses or shortages. This is exacerbated by the time-sensitive nature of frozen products and the unpredictable urban environments (like Ho Chi Minh City) that make strict adherence to delivery schedules difficult. The authors aim to create a model that balances three critical cost components: transportation expenses, penalties for untimely deliveries (early or late), and penalties for incorrect delivery quantities (over or under). This is a significant problem because traditional distribution models often struggle to account for these intertwined costs and the dynamic nature of real-world logistics.

Methods

The core of this research is a mathematical model designed to optimize vehicle allocation and delivery schedules. The model incorporates decision variables for truck usage, product quantities delivered, and delivery times. It features objective functions to minimize costs related to truck utilization, early/late arrivals, and quantity discrepancies. Crucially, it includes constraints to ensure deliveries fall within acceptable time windows and quantity ranges, and that truck capacities are not exceeded.

The model was initially validated using:

  • Trial-and-Error Method: A basic approach to test the model on small sample problems.
  • CPLEX: A commercial optimization solver used to find optimal solutions for the mathematical model.

Recognizing CPLEX's limitations with large datasets and time constraints, the researchers then employed:

  • Simulated Annealing (SA): A metaheuristic algorithm used to find near-optimal solutions more efficiently. SA was chosen for its ability to escape local optima and its effectiveness in complex optimization problems. The paper details SA's initial solution generation (based on vehicle capacity and customer delivery times) and its neighborhood search mechanisms (swap, reversion, insertion).

Datasets

The paper uses a synthetic dataset for model validation. This dataset includes:

  • Two trucks
  • Three customers
  • Two product types

This small dataset was used for the initial trial-and-error and CPLEX validation. For evaluating the Simulated Annealing algorithm, larger datasets were generated, ranging from 3 to 35 customers, with varying numbers of trucks and products. These datasets are not explicitly made public but are described as subsets (S3, S5, S7, etc.) used for testing.

Key Insights and Ideas

  • Holistic Cost Integration: The paper's main contribution is the integrated approach to minimizing three distinct but interconnected cost factors: transportation, time penalties, and quantity penalties. This is particularly relevant for frozen goods where all three can have significant financial and operational impacts.
  • Balancing Efficiency and Flexibility: The model acknowledges that rigid delivery schedules are often impractical in dynamic urban environments. It builds in flexibility by allowing for acceptable deviations within defined time windows and quantity ranges, but penalizes excessive deviations.
  • Simulated Annealing for Scalability: The research highlights a critical practical limitation of exact solvers like CPLEX: their inability to handle large-scale problems within reasonable timeframes. The successful application of Simulated Annealing demonstrates a viable path for real-world implementation where speed and near-optimality are often preferred over absolute optimality.
  • Frozen Goods Specifics: The emphasis on frozen products underscores the importance of time sensitivity and the need for precise quantity management to avoid spoilage and meet customer demand effectively.

Key Results

  • CPLEX Validation: The mathematical model, when solved by CPLEX, produced consistent results with the trial-and-error method for small datasets, confirming its correctness. For a sample problem, the total cost was $271.0.
  • Simulated Annealing Performance: For larger datasets (S20 to S35), where CPLEX failed to provide a solution within a 24-hour timeframe, Simulated Annealing delivered comparable results in approximately 1 to 5 minutes. This represents a significant speedup and demonstrates SA's practical utility.
  • Accuracy: The results from SA closely matched CPLEX for smaller datasets (S3, S5), with differences generally within a few percentage points for larger datasets where CPLEX struggled. For instance, for dataset S5, SA had a 3.60% difference in objective value compared to CPLEX.

Limitations and Future Work

  • Limitations:
    • CPLEX struggles with large datasets, requiring alternative algorithms like Simulated Annealing.
  • The paper focuses on a single-day distribution model.
  • Future Work:
    • Develop a multi-day distribution model for long-term cost forecasting and planning.
  • Incorporate a model where vehicles can return to pick up additional orders to optimize vehicle usage.
  • Explore combining Simulated Annealing with other algorithms to further improve solution quality.

Summary updated

Abstract

Within the cross-dock facility, suppliers transport goods on pallets; however, there are instances when the received amount does not align with the allocation demand, resulting in either surplus or shortage.This situation raises the challenge of determining the appropriate customer to handle the excess or shortfall while also addressing any potential penalties.In the case of frozen deliveries, time sensitivity is of utmost importance, necessitating swift and efficient procedures.However, Ho Chi Minh City's dynamic environment poses challenges when delivery schedules deviate, making it necessary to accept such variations.The problem also involves optimizing vehicle allocation and managing early or late deliveries, along with over-deliveries or under-deliveries.The objective of this study is to create an enhanced schedule and strategy for vehicle allocation that effectively reduces three cost components: transportation expenses, penalties associated with untimely deliveries, and fines for inaccuracies in product quantities.A robust model will be introduced to balance these costs effectively, providing valuable insights for frozen product distribution.The model was initially tested on a sample problem using the Trial-and-Error method with CPLEX.Given the large data set and time constraints in real-world scenarios, Simulated Annealing is used to optimize the solution faster, making it more practical for implementation.
BibTeX
@inproceedings{vu2025decision,
  author = {Minh Nhat Vu and Hoai-Phuc Tran and Huu‐Thanh Nguyen and Mai-Ha Phan},
  title = {{Decision Making Model for Frozen Distribution using Cross-dock}},
  booktitle = {Operations},
  year = {2025},
  doi = {10.31387/oscm0610467},
}
Figure