ABSTRACT Dynamic mobile wireless sensor networks (WSNs) have been a challenge in energy efficiency, network life, and quality of service (QoS) because of the mobility of the nodes, limited battery capacity, and network dynamicity. The study will suggest a scalable federated multi‐agent deep reinforcement learning (FMADRL) model in order to resolve these problems. Every sensor node acts as an autonomous agent that monitors the local conditions, such as residual energy, queue length, distance to neighbors, and velocity, and decides intelligently on routing and transmission. Federated learning makes it possible to perform unified model updates without the exchange of raw data, which is less stressful in terms of communication overhead and also maintains privacy. The framework uses energy conscious and QoS focused reward functions to satisfy energy consumption, latency, packet delivery ratio (PDR), and throughput while dynamically adapting to changes in the topology. The results of simulation prove that FMADRL greatly exceeds the baseline techniques. The average node energy consumption is brought down to 301 mAh (as compared with 455 mAh in regular DRL and 482 mAh in Q‐learning). Network lifetime is improved to 146 h, and PDR has been increased to 97% with throughput of 580 kbps and latency of 72 ms. Federated aggregation will save 70% in communication overhead. These findings confirm that FMADRL is a scalable, energy‐efficient, and QoS‐aware system, which can sustain a high level of performance in mobile WSN systems under large and highly dynamic settings.
Senthamilselvi et al. (Wed,) studied this question.