Self-learning adaptive power management scheme for energy-efficient IoT-MEC systems using soft actor-critic algorithm

Abstract

The rapid increase of Internet of Things (IoT) devices in Mobile Edge Computing (MEC) environments requires effective energy management to ensure device operation and enhance network efficiency. IoT-MEC systems face challenges such as varying task loads, dynamic environmental conditions, and limited energy resources. These factors make it challenging to design adaptive and efficient energy strategies. Traditional methods, such as static scheduling and centralized control strategies, struggle to adapt to real-time fluctuations in task loads and network conditions, resulting in inefficient energy use, higher latency, and a lack of flexibility to respond to these demands in real-time. This paper proposes a self-learning power management model using the Soft Actor-Critic (SAC) algorithm. It is deployed on IoT devices to enable localized and context-aware power management. Our model includes modules for energy monitoring, adaptive task prioritization, and a self-adjusting reinforcement-learning mechanism, which dynamically fine-tunes energy policies based on real-time device conditions, allowing each device to optimize power use independently without heavy dependence on centralized control. MEC nodes gather data on battery health, load, and network conditions to support decentralized policy adjustments. Connected devices in simulated smart homes served as the primary context for evaluation. Experimental results show that our model achieves a 45 % reduction in energy consumption in smart home environments, a 49 % improvement in battery life (compared to baseline-like Local Computing), and high adaptability in diverse scenarios compared with other methods.