Dynamic Multi-Objective Service Resource Scheduling via LLM-Optimized Fuzzy State Fusion and Reinforcement Learning Closed Loop

Efficient resource scheduling is a cornerstone for optimizing service quality and operational efficiency in modern healthcare, logistics, and cloud computing systems. However, existing methods often struggle with dynamic environments, multi-objective trade-offs, and stringent Service Level Agreement (SLA) constraints. To address these challenges, this paper proposes a novel framework named LFRL-MOS (LLM-Fuzzy-Reinforcement Learning Multi-Objective Scheduling). The framework integrates three key innovations: (1) LLM (Large Language Model) -enhanced fuzzy state fusion, which unifies real-time and predictive states through dynamically adjusted membership functions; (2) SLA-constrained Multi-Objective Ant Lion Optimizer (MOALO), which incorporates dynamic Pareto frontier adjustments and cross-department penalty mechanisms to ensure low-violation solutions; and (3) a dual-loop Proximal Policy Optimization (PPO) mechanism, which adaptively tunes parameters across all components for robust performance. Extensive experiments on simulated medical resource scheduling scenarios demonstrate that LFRL-MOS significantly outperforms state-of-the-art baselines in terms of SLA violation rate, average response time, resource utilization efficiency, and cost effectiveness. Our findings highlight the potential of combining advanced machine learning techniques with evolutionary optimization for addressing complex, dynamic scheduling problems.