Optimization of shunting operation plan in large freight train depot based on DQN algorithm.

Area of Science:

• Railway Operations Research
• Artificial Intelligence in Transportation
• Logistics Optimization

Background:

• Shunting operations are critical for freight train depots, impacting overall railway efficiency.
• Optimizing shunting plans is essential for improving production and transportation throughput.
• Current methods may not fully address the complexity of shunting operations.

Purpose of the Study:

• To develop and validate a deep reinforcement learning (DRL) model for optimizing shunting operations.
• To minimize the number of shunting hooks required for train reorganization.
• To enhance the efficiency and intelligence of shunting operations in large freight train depots.

Main Methods:

• Constructed a DRL environment with defined actions, states, and rewards for shunting operations.
• Utilized the Deep Q network (DQN) algorithm, with the shunting locomotive as the agent.
• Designed a reward function based on the total shunting hooks generated post-reorganization.

Main Results:

• DQN significantly reduced the number of shunting hooks compared to Overall Planning and Coordinating (OPC) and Binary Search Tree (BST) algorithms.
• DQN-generated plans occupied fewer lanes and resulted in 10% fewer shunting hooks than OPC.
• DQN outperformed Branch and Bound (B&B) in solving time and reduced coupling/slipping operations, indicating superior plan quality.

Conclusions:

• The DQN algorithm provides an effective and intelligent solution for optimizing shunting operations in freight train depots.
• DRL offers a promising approach to enhance efficiency and reduce operational costs in railway logistics.
• This research contributes a novel method for the intelligent automation of complex shunting tasks.