Reinforcement Learning-Based Dynamic Resource Allocation and Optimization for Enhanced Performance and Energy Efficiency in IoT Systems

Abstract

The proliferation of the Internet of Things (IoT) has introduced significant challenges in resource management due to the diverse and dynamic nature of IoT devices and applications. This paper presents a novel approach that leverages reinforcement learning algorithms for dynamic resource allocation and optimization to enhance system performance and energy efficiency in IoT environments. By training models on historical and simulated data, the proposed solution learns optimal policies for resource distribution that maximize vital performance by achieving optimized throughput as 1378.7312500000116, latency as 0.09014474755345958, and energy consumption as -564.8125. Our experiments demonstrate that the reinforcement learning-based method effectively adapts to changing environmental conditions and varying workload demands, outperforming traditional static allocation strategies. We also enhance the security level by using machine learning methods like isolation forest; in this, we attain average model stability as 0.9792682926829268.