NEURAL NETWORK-BASED DATA COMMUNICATION SYSTEM IN INTERNET OF VEHICLES

This paper proposes a communication system for Internet of Vehicles based on a neural network, to minimize the transferred data amount and energy consumption. Preliminary simulations justify the role of such systems and show that vehicle classification may reduce the network load significantly if the classification process is performed fast enough considering the number of vehicles in the communication area and their velocity. A flexible and scalable framework for a neural network implementation using Computer Unified Device Architecture technology is illustrated. The performance evaluation (computing time and speed-up) in neural network normal operation and training phase, comparing with similar CPU implementation, shows good results for average characteristics of commercial computers. This communication system can be used also in Internet of Things.