Enhancing safety and efficiency in automated container terminals: Route planning for hazardous material AGV using LSTM neural network and Deep Q-Network

Abstract

As the proliferation and development of automated container terminal continue, the issues of efficiency and safety become increasingly significant. The container yard is one of the most crucial cargo distribution centers in a terminal. Automated Guided Vehicles (AGVs) that carry materials of varying hazard levels through these yards without compromising on the safe transportation of hazardous materials, while also maximizing efficiency, is a complex challenge. This research introduces an algorithm that integrates Long Short-Term Memory (LSTM) neural network with reinforcement learning techniques, specifically Deep Q-Network (DQN), for routing an AGV carrying hazardous materials within a container yard. The objective is to ensure that the AGV carrying hazardous materials efficiently reaches its destination while effectively avoiding AGVs carrying non-hazardous materials. Utilizing real data from the Meishan Port in Ningbo, Zhejiang, China, the actual yard is first abstracted into an undirected graph. Since LSTM neural network can efficiently conveys and represents information in long time sequences and do not causes useful information before long time to be ignored, a two-layer LSTM neural network with 64 neurons per layer was constructed for predicting the motion trajectory of AGVs carrying non-hazardous materials, which are incorporated into the map as background AGVs. Subsequently, DQN is employed to plan the route for an AGV transporting hazardous materials, aiming to reach its destination swiftly while avoiding encounters with other AGVs. Experimental tests have shown that the route planning algorithm proposed in this study improves the level of avoidance of hazardous material AGV in relation to non-hazardous material AGVs. Compared to the method where hazardous material AGV follow the shortest path to their destination, the avoidance efficiency was enhanced by 3.11%. This improvement demonstrates potential strategies for balancing efficiency and safety in automated terminals. Additionally, it provides insights for designing avoidance schemes for autonomous driving AGVs, offering solutions for complex operational environments where safety and efficient navigation are paramount.

Keywords

• container yard
• route planning
• hazardous material automated guided vehicle (agv)
• long short-term memory (lstm)
• deep q-network (dqn)