Optimizing cloud energy with deep reinforcement learning: an artificial bee colony-Q-learning hybrid approach

Abstract

Load balancing in the cloud is a crucial characteristic for distributing resources from data centers. However, load balancing in a cloud environment remains a challenging issue due to the influx of numerous incoming tasks, particularly in an infrastructure as a service (IaaS) cloud environment. To address the complexity and high dimensionality of the process, machine learning (ML) techniques combined with optimization algorithms are used, which fasten the process and minimize execution time. The deep reinforcement learning (DRL) – artificial bee colony with Q-learning (ABCQ) approach is proposed to solve load balancing in a cloud environment. The agent in DRL makes intelligent decisions to handle a large number of tasks efficiently at lesser time and increasing speed. The artificial bee colony (ABC) algorithm uses a global guiding factor to initialize position updates, enhancing the searchability and performance of the algorithm. The proposed DRL-ABCQ approach optimizes scheduling and resource utilization, increases the throughput of virtual machines (VMs), and balances the load between VMs. The DRL-ABCQ approach achieves a makespan of 125.52 s, energy utilization of 99.81 kJ, and a throughput of 2.39 jobs/s, compared to existing approaches.