SECURE AND PRIVACY-PRESERVING WSNS IN SMART CITY VIA BLOCKCHAIN-ENABLED FEDERATED LEARNING AND HOMOMORPHIC ENCRYPTION

Abstract

The rapid exponential growth of WSN and IoT devices in smart cities has revolutionized urban services, including traffic management, environmental monitoring, and automation of major infrastructures. Voluminous data generated by heterogeneous sensors, however, present significant challenges to privacy, security, and resource efficiency. Traditional solutions have been plagued by high computation costs, vulnerabilities to malicious nodes, as well as a lack of adequate security for secretive information during model aggregation. In a quest to overcome such vulnerabilities, this work presents a novel framework for secure and private WSNs under blockchain-enabled federated learning as well as homomorphic encryption. In this proposed approach, two novelties are implemented, including Lightweight Local Model Training via Edge-MGTNet, utilizing MobileNet-V3, Tiny-GNN, Micro-TCN, edge pruning, as well as knowledge distillation for lightweight local intelligence extraction, and Encrypted Model Packaging via HashEnc-SparseNet, utilizing sparsified gradient encoding, Paillier homomorphic encryption, as well as hash-based integrity verification (HIV) for encrypting transmission of models with tamper-proof blockchain registration. Robust experiments on merged smart city IoT datasets reveal exemplary performance compared to prior solutions, with Accuracy = 99.22%, Precision = 98.32%, Sensitivity = 96.9%, and Specificity = 96.56%, indicating the effectiveness of such a framework for offering resilient, trustworthy, as well as private intelligence across heterogeneous WSN nodes.