An Entropy-Guided Monte Carlo Tree Search Approach for Generating Optimal Container Loading Layouts

Area of Science:

• Operations Research
• Logistics and Supply Chain Management
• Artificial Intelligence

Background:

• The container loading problem is a complex optimization challenge.
• Existing deterministic algorithms often fail to provide efficient solutions.
• There is a need for algorithms that balance space utilization with packing neatness.

Purpose of the Study:

• To propose a novel approach for the container loading problem using spatial entropy.
• To develop an algorithm that generates neat and easily applicable container layouts.
• To compare the performance of entropy-driven algorithms against traditional methods.

Main Methods:

• A Monte Carlo Tree Search (MCTS) algorithm was developed.
• A spatial entropy measure was introduced to bias the MCTS.
• Three algorithms were analyzed: basic MCTS, entropy-driven MCTS, and a combined approach.
• Performance was compared against a classical deterministic algorithm.

Main Results:

• The entropy-driven MCTS algorithms generated more consistent and practical layouts.
• The combined entropy-driven algorithm showed superior performance.
• Entropy-driven methods provided good results even when classical algorithms failed.
• These algorithms achieved good results in short computational times.

Conclusions:

• Spatial entropy is an effective measure for improving container loading algorithms.
• The proposed entropy-driven MCTS approach offers a robust and efficient solution.
• This method enhances both space utilization and packing neatness for practical applications.