DF-RL: A Dynamic Fuzzy-Neuro Reinforcement Learning Framework for Cloud Resource Management

Abstract

This paper presents a dynamic fuzzy-neuro reinforcement learning (DF-RL) framework designed for resilient cloud resource management. By integrating the complementary strengths of fuzzy logic, neural networks, and hierarchical reinforcement learning, the proposed framework effectively addresses the uncertainties and dynamic conditions prevalent in cloud computing environments. Specifically, DF-RL utilizes an adaptive neuro-fuzzy inference system (ANFIS) to dynamically fine-tune fuzzy rules, while decision-making is performed through a hierarchical deep Q-network (HDQN) structure. Experimental evaluations demonstrate that DF-RL substantially outperforms existing approaches in resource utilization efficiency, task completion speed, and overall quality of service (QoS). Furthermore, the framework exhibits robust adaptability to workload fluctuations, highlighting its significant potential to manage the complexities and dynamic challenges inherent in contemporary cloud computing scenarios.