Game-Theoretic Reinforcement Learning for Stable Competitive Resource Allocation in Dynamic Cloud Environments

Abstract

With the increasing prevalence of multi-tenant concurrent access, heterogeneous task hybrid operation, and dynamic fluctuations in resource demand in cloud computing platforms, the stability problem of competitive allocation under limited resource conditions is becoming increasingly prominent. Addressing the shortcomings of traditional methods in simultaneously considering resource utilization, fairness constraints, and system stability in complex competitive environments, this paper proposes a cloud resource competitive allocation method based on the fusion of game theory and reinforcement learning. This method first starts from the multi-agent resource competition relationship in the cloud environment, modeling resource requesters and the scheduling process as a dynamic interaction process. State vectors are used to uniformly characterize key information such as remaining resources, queue pressure, task priority, and competition intensity, thereby enhancing the model's ability to describe real-world cloud resource competition scenarios. Based on this, a game- theoretic interaction mechanism and a hybrid strategy selection mechanism are introduced to model the strategy coupling relationship between different agents under limited resource constraints, enabling resource allocation decisions to better reflect the mutual influence and equilibrium characteristicsin the multi-agent competition process. Furthermore, by combining reinforcement learning strategy optimization mechanisms, a reward-shaping guidance model simultaneously focuses on resource utilization efficiency, fair allocation effectiveness, and conflict suppression capabilities during long-term interactions. Stability regularization constraints are used to reduce fluctuations and imbalances in resource allocation, thereby improving the continuity, coordination, and robustness of the schedulingstrategy in dynamic environments. After constructing a resource competition allocation scenario using public cloud cluster data, this paper systematically validates the proposed method. Results show that this method effectively improves the overall performance of resource competition allocation, demonstrating good application potential in resource utilization, fairness assurance, system stability, and overall coordination. The research indicates that the synergistic integration of game theory mechanisms and reinforcement learning strategy Artificial Intelligenceand ComputingInnovations optimization provides a feasible and interpretable technical path for studying the stabilityof resourcecompetition allocationin complexcloud environments.