Energy-Efficient Adaptive Hierarchical Federated Learning Architecture for Mobile Target Detection in Wireless Sensor Networks

Wireless sensor networks (WSNs) support mobile target detection in various cyber–physical applications, yet their long-term operation is constrained by limited battery capacity and costly wireless communication. Federated learning (FL) enables collaborative model training without centralizing raw sensor data, improving privacy preservation in distributed settings. However, conventional FL can impose substantial communication overhead and consequently accelerate energy depletion in WSN deployments. To address these limitations, we propose an adaptive hierarchical federated learning architecture tailored to mobile target detection in clustered WSNs. The proposed approach organized nodes into clusters and performed multi-level aggregation to reduce the frequency and volume of uplink transmissions. In addition, reinforcement learning is employed to adaptively select cluster heads based on network conditions, with the objective of balancing energy expenditure and prolonging network lifetime while maintaining effective learning. By coupling hierarchical aggregation with adaptive cluster-head selection, AH-FL aims to decrease redundant communication, mitigate bottlenecks at energy-critical nodes, and improve overall energy efficiency during distributed training and inference. Simulation-based evaluation indicates that AH-FL outperforms representative baseline approaches in terms of energy consumption and total transmitted data, leading to an extended operational lifetime of the network under mobile tracking workloads. These results suggest that adaptive hierarchical FL is a practical direction for privacy-aware, energy-efficient learning in resource-constrained WSNs. Keywords: Wireless sensor networks; federated learning; hierarchical aggregation; reinforcement learning; clustering; energy efficiency; mobile target detection. DOI： https://doi.org/10.35741/issn.0258-2724.60.5.2