A Hybrid Neural Network and Graph Theory based Clustering Protocol for Dynamic IoT Networks

Internet of Things has emerged as a revolutionary technology that holds promise in a wide range of applications. However, its deployment presents some difficulties since IoT networks are based on battery-empowered devices. Clustering techniques were introduced to conserve the energy of network devices. Recently, Machine Learning-neural network-based clustering techniques have proven their efficiency for topology management and network lifetime. Graph theory is a powerful mathematical and computational discipline that studies graphs, which are abstract models of network designs connecting objects and allowing significant decisions to be made that affect network performance. In this paper, a combination of neural networks and graph theory is done in order to perform cluster based routing in dynamic IoT networks called SOM-FW. The proposed method clusters sensor nodes using the artificial neural network method named Self-Organizing Maps, based on their distribution, providing a balanced CH distribution. The election of the cluster-head is based on the Floyd-Warshall algorithm, which computes the center of the cluster graph. A dynamic re-clustering strategy is adopted to handle novel incoming nodes, faulty or dead nodes, and CH premature death. The superiority of the protocol is extensively demonstrated in energy efficiency, and network lifetime.