Fourth Party Logistics Routing Optimization Problem Based on Conditional Value-at-Risk Under Uncertain Environment

Guihua Bo, Qiang Liu, Huiyuan Shi, Xin Liu, Chen Yang, Li-Yan WangInternational Journal of Advanced Computer Science and Applications 2025 · DOI

Abstract

In order to improve the level of logistics service and considering the impact of uncertainties such as bad weather and highway collapse on fourth party logistics routing optimization problem, this paper adopts Conditional Value-at-Risk (CVaR) to measure the tardiness risk, which is caused by the uncertainties, and proposes a nonlinear programming mathematical model with minimized CVaR. Furthermore, the proposed model is compared with the VaR model, and an improved Q-learning algorithm is designed to solve two models with different node sizes. The experimental results indicate that the proposed model can reflect the mean value of tardiness risk caused by time uncertainty in transportation tasks and better compensate for the shortcomings of the VaR model in measuring tardiness risk. Comparative analysis also shows the effectiveness of the proposed improved Q-learning algorithm.

Introduction

In the modern logistics landscape, Fourth Party Logistics (4PL) providers act as "super-integrators," managing vast networks of Third-Party Logistics (3PL) providers to create seamless supply chains. However, the primary challenge for 4PLs is uncertainty. Factors such as traffic congestion, severe weather, or infrastructure failures render delivery times stochastic rather than deterministic.

Traditional optimization models often rely on Value-at-Risk (VaR), which identifies the maximum delay expected within a specific confidence level (e.g., 95%). However, VaR is inherently limited because it ignores the "tail risk"—the severity of delays that exceed the threshold. This paper addresses this gap by utilizing Conditional Value-at-Risk (CVaR) to focus on the average severity of extreme delays, providing a more robust framework for risk-averse logistics management.

Methodology

The research approaches the routing problem through two primary innovations: a risk-aware mathematical model and an improved reinforcement learning algorithm.

The CVaR Model

The authors propose a nonlinear programming model that minimizes CVaR instead of simple cost or average time. CVaR calculates the expected value of delays given that they have already exceeded a defined threshold. This approach incentivizes the selection of routes that remain resilient even under worst-case scenarios.

Improved Q-Learning on Multi-Graphs

Routing in a 4PL environment is complex because it involves a "multi-graph" structure, where multiple 3PL providers may offer different services (with varying costs and times) between the same two nodes.

Illustration of a 4PL logistics network as a multi-graph. The red path highlights a specific route selection.

To solve this, the authors developed an improved Q-learning algorithm. Unlike standard algorithms that require simplifying the graph, this Q-learning agent learns the value of selecting specific 3PLs at each node. The reward function is customized to balance delivery cost, average transit time, and time variance (reliability).

Experimental Setup

To test the model's performance and scalability, the authors generated synthetic logistics network instances ranging from small-scale (7 nodes) to large-scale (50 nodes). Each edge in these networks was assigned random variables for cost and time, simulating the diverse offerings of real-world 3PL providers. The algorithm's performance was benchmarked against traditional methods like Genetic Algorithms (GA) and standard VaR models.

Results and Discussion

The experimental results highlight three key advantages of the proposed approach: speed, stability, and risk accuracy.

Performance of the improved Q-learning algorithm across different network sizes.
  1. Computational Efficiency: The improved Q-learning algorithm demonstrated remarkable speed. For a 30-node problem, it identified the optimal path in just 1.5 seconds, significantly outperforming standard Genetic Algorithms which required 28.5 seconds for the same task.
  2. Scalability and Stability: As shown in the performance data, the algorithm remains effective even as the network size increases to 50 nodes. It maintained a high "Best Rate," finding optimal solutions consistently without becoming trapped in local optima.
  3. Risk Sensitivity: The CVaR model consistently produced higher risk values than the VaR model. This confirms that VaR tends to underestimate the potential impact of extreme delays, whereas CVaR provides a more realistic view for high-stakes logistics planning.

Conclusion

This study demonstrates that CVaR is a superior metric for managing tardiness risk in 4PL routing under uncertainty. By integrating CVaR with an improved Q-learning algorithm, 4PL providers can offer more reliable service tiers and better manage extreme "tail risks." Future work will explore the model's performance under non-normal probability distributions (such as fat-tailed distributions) and consider multi-objective optimization involving cargo damage risks alongside delivery delays.

Summary updated

Abstract

In order to improve the level of logistics service and considering the impact of uncertainties such as bad weather and highway collapse on fourth party logistics routing optimization problem, this paper adopts Conditional Value-at-Risk (CVaR) to measure the tardiness risk, which is caused by the uncertainties, and proposes a nonlinear programming mathematical model with minimized CVaR. Furthermore, the proposed model is compared with the VaR model, and an improved Q-learning algorithm is designed to solve two models with different node sizes. The experimental results indicate that the proposed model can reflect the mean value of tardiness risk caused by time uncertainty in transportation tasks and better compensate for the shortcomings of the VaR model in measuring tardiness risk. Comparative analysis also shows that the effectiveness of the proposed improved Q-learning algorithm.
BibTeX
@article{bo2025fourth,
  author = {Guihua Bo and Qiang Liu and Huiyuan Shi and Xin Liu and Chen Yang and Li-Yan Wang},
  title = {{Fourth Party Logistics Routing Optimization Problem Based on Conditional Value-at-Risk Under Uncertain Environment}},
  journal = {International Journal of Advanced Computer Science and Applications},
  year = {2025},
  doi = {10.14569/ijacsa.2025.01602114},
}
Figure