Adaptive Scheduling of Mixing Trucks in Construction Sites with an Improved Deep Q-Network

Abstract

The management of concrete mixing station distribution is evolving toward more intelligent and efficient methods. Additionally, in the context of the group operations of commodity concrete enterprises, the cooperation between each mixing station during the distribution process has become increasingly crucial. Therefore, establishing a scientific and effective cooperative scheduling system is of practical significance. This paper proposes a centralized hierarchical dispatch system for construction site mixer trucks, designed to meet the control requirements of cooperative transport. The upper computer control system handles single-vehicle scheduling tasks, while cooperative transport tasks are determined by the system based on task and environmental conditions. The system establishes the cooperative transport formation, with the primary vehicle responsible for both the assembly and scheduling paths. The "master vehicle" at the construction site identifies the "slave vehicles" and oversees control and coordination during collaborative operations. This study explores the scheduling method based on the Improved Deep Q-Network (IDQN) algorithm, leveraging a mathematical model of scheduling and a reinforcement learning environment. The paper details the basic principles of the DQN algorithm and outlines the learning process for the scheduling algorithm tailored to construction sites. Furthermore, it designs a local mechanism for the scheduling agent and an action selection method based on environmental state information specific to various construction sites. It also defines the interaction between scheduling agents, the scheduling algorithm, and the construction site mixer truck path planning algorithm. The simulation results show that the IDQN scheduling algorithm outperforms five other algorithms, demonstrating better performance, adaptability, and real-time responsiveness to environmental changes.