A Lightweight Deep Learning Framework for Intelligent and Secure IoT Network Management in Smart City Environments

Smart cities need to depend on large-scale Internet of Things (IoT) networks for critical urban functions like traffic management, air quality monitoring, smart grid, waste management and surveillance by the public. The need to keep these networks energy-efficient, secure, and reliable and to manage them on a large scale, however, poses significant challenges due to the heterogeneous and dynamic nature of these networks. In this paper, the concept of lightweight deep learning framework for intelligent clustering and routing in smart city IoT environment is proposed, which is termed as OptiMobileGAT-Net. They combine MobileNet-V2 for efficient spatial feature extraction, Graph Attention Networks (GATs) for topology-aware representation learning, and Gated Recurrent Units (GRUs) for modeling temporal dynamics of the network conditions to make adaptive routing decisions. The review of recent IoT network management approaches reveals four research gaps: high computational overhead optimization algorithms, lack of attention to energy efficiency issues, lack of focus on trust and security, and instability of RL in dynamic environments, limited adaptability of existing deep learning models for resource-limited devices.

To overcome these limitations, we introduced multi-factor node feature representation, lightweight spatial feature, topology learning based on graph attention, temporal modelling with GRUs, and a feed-forward attention optimization module, which collectively considers energy, trust, and security without iterative optimization. The framework is intended to be evaluated in a simulated 100-node smart city IoT network with malicious-node scenario. The identified research gaps and proposed methodology are presented in this paper and experimental validation and performance evaluation are presented in the subsequent stage.