Federated Learning-Based Adaptive Protocol for Privacy-Aware and Energy-Efficient Wireless Sensor Networks in Edge-IoT Systems

Wireless Sensor Networks (WSNs) are increasingly being used in contemporary Edge-IoT systems, but face a number of persistent concerns such as energy constraints, private data, and dynamic network performance. In this study, we propose a Federated Learning-based Adaptive Protocol (FLAP) that allow for in-network intelligence, and it does so in decentralized manner, nurturing energy and protecting sensitive data in the heterogeneous sensor nodes. Experimental evaluations show that FLAP reduces the average energy consumption by 21.8%, prolongs the network lifetime by 32.5%, and reaches an accuracy of 94.2% under different network conditions. The communication overhead is also reduced by 39.6%, and latency is always maintained within the 95 ms in common update rounds. We use TensorFlow Federated for implementing the distributed training process, and develop dynamic scheduling strategies to manage the node participation by considering real-time energy status, latency conditions, and learning convergence rate. This includes a hybrid energy aware clustering algorithm for increase scalability and eliminate redundant transmissions. The framework is implemented and evaluated on a Python-based simulation stack, which uses NumPy, SciPy and Scikit-learn for data processing, and Matplotlib for drawing the results. We employ a lightweight AES-128 encryption module based on PyCryptodome to secure the model parameter exchange, which ensures that users' privacy can be protected without sacrificing performance. Our results conclude that the protocol is suitable for real-time and privacy-preserving energy-efficient deployment of smart application like health monitoring, industrial automation and environmental sensing.