Application and optimization of reinforcement learning algorithm for coal yard transportation scheduling

Aiming at the transportation scheduling problem of coal yard, this paper proposes an optimization algorithm framework based on reinforcement learning (RL). Firstly, the coal yard transportation scheduling problem is modeled as markov decision processes (MDP), and a state space including dynamic environment and equipment state is designed, as well as an action space including vehicle path planning, equipment start-stop strategy and task priority adjustment. Furthermore, a multi-objective dynamic reward function is constructed, and an adaptive weight mechanism is introduced to balance short-term tasks and long-term equipment maintenance costs. To enhance algorithm performance, this paper adopts a hybrid framework that combines a rule engine with the Proximal Policy Optimization (PPO) algorithm to improve system robustness. Multi-Agent Reinforcement Learning (MARL) with the QMIX algorithm is utilized to achieve coordinated scheduling between vehicles and equipment, reducing path conflicts. Through simulation experiments and validation with real-world data, the results demonstrate that the proposed algorithm outperforms traditional methods in transportation costs, penalty costs, and maintenance costs. It significantly improves task completion rates, reduces the number of path conflicts, and optimizes equipment utilization. Sensitivity analysis of dynamic weights shows that the adaptive weight mechanism effectively responds to environmental changes, achieving more efficient, environmentally friendly, and safer operational management.