An Intelligent Deep Reinforcement Learning Framework for Cost-Efficient and Energy-Aware Cloud Resource Management

Abstract

The rapid growth of cloud computing and Internet of Things (IoT) applications has intensified the need for efficient resource allocation, energy management, and cost optimization in large-scale data centers. Traditional optimization and machine learning approaches, while effective in static environments, often fail to adapt to dynamic workloads and heterogeneous infrastructures, resulting in performance degradation, increased operational costs, and reduced quality of service (QoS). Addressing these challenges is essential for sustainable and scalable cloud service management. Despite recent advances, existing frameworks still suffer from slow convergence, high sensitivity to learning parameters, and limited scalability in real-world deployments. Furthermore, the imbalance between multiple objectives such as cost, energy, and QoS remains unresolved. In this paper, a DQN-driven resource optimization framework is proposed. The method employs advanced variants of Deep Q-Networks and introduces a multidimensional reward function that integrates virtual machine rental cost, energy consumption, and SLA compliance. By dynamically learning adaptive policies, the framework enables real-time resource allocation and scaling decisions in complex cloud environments. The simulation results demonstrate that, compared to the average performance of the related work, the proposed method achieves a 16.3% reduction in average response time, 7.7% lower energy consumption, and 11.3% reduction in operational cost, while simultaneously improving the task acceptance rate by 6.2%.