AI-Integrated Trust- and Energy-Aware Federated Learning for Intrusion Detection in Wireless Sensor Networks

Wireless Sensor Networks (WSNs) deployed in critical Internet of Things (IoT) environments face severe security risks while operating under tight computational, bandwidth, and battery constraints1. Centralized Intrusion Detection Systems (IDSs) introduce high communication overheads, latency spikes, single-point vulnerabilities, and privacy risks3. Although Federated Learning (FL) enables privacy-preserving collaborative model training by keeping raw sensing data localized, standard FL protocols remain highly vulnerable to malicious model poisoning attacks, stragglers, rapid battery depletion, and convergence degradation under non-IID data distributions5. To resolve these coupled security and resource challenges, this paper presents an AI-Integrated Trust- and Energy-Aware Federated Learning (TEA-FL) framework specifically designed for resource-constrained WSNs. The framework introduces a joint node selection and aggregation weight formulation combining dynamic trust scoring via Exponential Moving Average (EMA) weight-update similarity with residual energy state estimation. Furthermore, local sensor models utilize a lightweight 1D-Convolutional architecture enhanced with supervised contrastive representation learning to insulate training against distributional skews. Evaluated on the ToN-IoT, WSN-DS, and UNSW-NB15 benchmark datasets under non-IID Dirichlet partitions  the proposed system achieves a classification accuracy of 99.57%, cuts false positive rates to 0.87%, and achieves a 38.6% reduction in per-round energy consumption compared to standard FL baselines.