The Application of Artificial Intelligence in Logistics Management Optimization Research

Chunxiang Wu, Rong WangScalable 2025 · DOI

TL;DR

This paper presents an AI-enhanced warehouse management system (WMS) that replaces error-prone manual data entry with an automated RFID-based architecture. By integrating a Naive Bayes classifier for data mining and simulating the workflow in Flexsim, the authors demonstrate a highly efficient inbound process achieving a 72% total utilization rate for Automated Guided Vehicles (AGVs).

What Problem Does This Paper Address?

Traditional logistics and warehouse management often rely on manual recording and barcode scanning, which are notoriously slow and prone to human error. These inefficiencies lead to "location conflicts" (where the system thinks a spot is empty, but it isn't), inaccurate inventory counts, and poor space utilization.

As e-commerce continues to scale, the "Internet Plus" era demands a level of speed and accuracy that manual systems simply cannot provide. The authors aim to bridge this gap by creating a "smart warehousing" framework that uses digital carriers (RFID) and artificial intelligence to perceive management objects dynamically and intelligently.

Methods

The researchers proposed a multi-layered approach to modernize the warehouse:

  1. Three-Layer Architecture:
    • Management Layer: The "brain" that handles querying, statistics, and high-level scheduling.
  2. Middleware Layer: Acts as the bridge, integrating RFID data collection with the database and translating management instructions into physical tasks.
  3. Physical Layer: The "boots on the ground," consisting of RFID readers, tags, AGVs, and destacking robots.
  4. Naive Bayes Algorithm: For the data mining component of the logistics monitoring system, the authors selected the Naive Bayes classifier. They chose this specifically for its simplicity and effectiveness in categorizing logistics information based on feature attributes, which helps in identifying and routing goods correctly.
  5. Flexsim Simulation: To validate their design without the cost of a full-scale physical pilot, the authors built a digital twin of a manufacturing workshop using Flexsim software. They simulated 24 hours of warehouse operations (86,400 simulation units) to analyze equipment efficiency and bottlenecks.

Datasets

The paper does not utilize a standard public benchmark dataset. Instead, the evaluation is based on simulated operational data generated within the Flexsim environment. The model was configured with specific parameters:

  • A storage area of 1840m².
  • A fleet of 12 AGVs.
  • Two storage ports and two destacking robots.
  • Input data based on a hypothetical production and manufacturing workshop's inbound logistics flow.

Key Insights and Ideas

  • The Power of Digital Carriers: By embedding RFID tags into product packaging, the system achieves "spatial coupling," allowing for non-contact, high-speed identification that far outpaces traditional manual logging.
  • Automated Exception Handling: The system includes a dedicated "RFID exception processing area." If a tag is misread or data is inconsistent with the cloud database, the system automatically reroutes the goods for human intervention, ensuring the main workflow remains unblocked.
  • The "Too Many Cooks" Problem: A fascinating insight from the simulation was that while the overall system was efficient, specific AGVs (AGV11 and AGV12) showed very low utilization rates (18.64% and 36.70%). This highlights that simply adding more hardware doesn't always lead to linear gains; it can actually lead to resource waste if the fleet size isn't optimized for the specific order volume.

Key Results

The simulation yielded concrete quantitative metrics for the proposed system:

  • Total AGV Utilization: The system reached an impressive 72% total utilization rate across the fleet.
  • Robot Efficiency: The destacking robots at the storage ports operated at 58.84% and 60.11% efficiency, respectively.
  • Throughput: In a single simulation cycle, the system successfully managed the storage of 5,756 boxes across two storage areas.
  • Error Reduction: Qualitatively, the system effectively resolved the slow data collection and low accuracy issues inherent in manual traditional logistics.

Limitations and Future Work

The primary limitation acknowledged is the idle equipment issue. The simulation revealed that having a fixed number of AGVs leads to low utilization for some units, which translates to unnecessary energy consumption and capital expenditure.

Future Work suggested by the authors includes:

  • Dynamic Fleet Scaling: Developing algorithms to set the number of active AGVs based on real-time order quantities to reduce idle time.
  • Energy Optimization: Further research into reducing energy consumption during "idle running" periods.
  • Real-world Implementation: Moving beyond the Flexsim environment to test the robustness of the Naive Bayes classifier against the physical interference often found in industrial RFID environments.

Summary updated

Abstract

In order to solve the problem of low efficiency and low accuracy in manual data collection in traditional logistics management systems and traditional warehouse management, the author proposes the application of artificial intelligence in logistics management optimization research. The author designed an RFID based warehouse management system that includes management layer, intermediate layer, and physical layer, focusing on the warehousing and warehousing operations in the production and manufacturing workshop. This system can optimize the traditional warehousing process, dynamically and intelligently perceive management objects, and achieve effective management and monitoring of warehousing management. Establish a simulation model using Flexsim software to simulate the inbound process and simulate the model entity in a relatively short period of time. The experimental results show that the total utilization rate of AGV in the simulation model reaches 72%. This method effectively solves the problems of slow data collection and low accuracy in traditional logistics warehousing processes.
BibTeX
@inproceedings{wu2025application,
  author = {Chunxiang Wu and Rong Wang},
  title = {{The Application of Artificial Intelligence in Logistics Management Optimization Research}},
  booktitle = {Scalable},
  year = {2025},
  doi = {10.12694/scpe.v26i3.4200},
}
Figure