Abstract

In Wireless Sensor Networks (WSNs), sensor nodes with limited resources must manage energy consumption and network efficiency. This work proposes an AM-DRL framework that adjusts transmission power, duty cycling, and clustering methods in response to real-time network conditions. Our methodology implements Multi-Agent Reinforcement Learning (MARL) to allow distributed decision-making by individual sensor nodes, which improves network lifetime while ensuring QoS parameters such as latency, packet delivery ratio, and throughput are met. The framework steers learning on the path of the energy-efficient decisions by introducing a novel hybrid reward function that accounts for energy expenditure, network topology, and data garnered. Transfer learning allows the model to be applied to different WSN configurations with minimal retraining efforts. AM-DRL outperformed other tested methods at saving energy and enhancing network lifetime and data transmission, during extensive simulations and real-world trials, proving more efficient than energy-saving approaches using reinforcement learning, and clustering. This paper provides scalable and intelligent WSN energy optimization solutions for industrial IoT, environmental monitoring, and smart city infrastructure.