Improving Energy Efficiency in Iot Edge Computing Through Dynamic Deep Q‐Network Based Resource Management

ABSTRACT Background With the rapid growth of IoT ecosystems, energy‐efficient resource allocation in edge computing has become critical. IoT devices produce dynamic, time‐sensitive workloads, demanding adaptive strategies that optimize energy usage without compromising performance. Existing methods often rely on static models or oversimplified assumptions, limiting their effectiveness in real‐world, variable environments. Objective This work proposes a novel, reinforcement learning‐based framework leveraging Deep Q‐Networks (DQN) for dynamic, energy‐aware resource allocation in edge computing environments, aiming to improve scalability and adaptiveness in the face of fluctuating demands. Methods The proposed approach utilizes Deep Q‐Networks to learn optimal resource allocation strategies based on real‐time system states. A replay buffer is incorporated to stabilize learning, and the impact of different discount factors is evaluated to balance short‐term energy use with long‐term efficiency. The framework is tested across varying device loads (10, 25, and 50 devices) to assess scalability and energy performance. Results Experimental results demonstrate significant reductions in energy consumption: 10 devices consume an average of 0.150 kWh, 25 devices consume 0.203 kWh, and 50 devices consume 0.290 kWh over a fixed simulation period. Q‐values ranging from 4.41 to 5.41 confirm effective learning and optimized decision‐making, showcasing the system's ability to adapt to real‐time environmental changes. Conclusion The proposed DQN‐based resource allocation framework effectively addresses the limitations of static models, offering a scalable, intelligent solution for energy‐efficient edge computing. This work contributes to sustainable IoT infrastructure by enabling responsive, adaptive, and optimized resource management in dynamic operational contexts. Future work includes real‐world deployment and further optimization under diverse workload conditions.