Reinforcement Learning-Based Intelligent Decision Modeling for Large-Scale Distributed Systems

Abstract

To address challenges in large-scale distributed systems, including high-dimensional state spaces, dynamic operating environments, and strongly coupled multi-node decisions, this study proposes a reinforcement learning based approach for intelligent decision modeling that provides a unified sequential formulation of the scheduling process. System states are abstracted into structured representations, while policy networks and value evaluation mechanisms jointly model the long-term impact of scheduling actions on system objectives. This enables integrated optimization of task completion efficiency, resource utilization, and load distribution. The method explicitly accounts for parallel operation across multiple nodes and the requirement of global objective consistency, allowing stable and effective scheduling behavior under complex dependencies. Comparative evaluations conducted in a public distributed system data environment demonstrate clear advantages across multiple system-level metrics, confirming the effectiveness and practical value of reinforcement learning driven intelligent decision modeling for improving the overall operational quality of large-scale distributed systems.