ISSN: 2182-2069 (printed) / ISSN: 2182-2077 (online)
A Deep Reinforcement Learning Approach for Traffic Control with the FESD-CVT Model in Internet of Connected Vehicles
Road congestion has become a serious problem in many parts of the world. The Intelligent Transportation System (ITS) can deliver high-quality services for cars due to recent advancements in networks. ITS needs help to optimally distribute resources due to the many uses of vehicles and the constantly changing state of the network. To address these issues, networking for ITS is made possible by artificial intelligence algorithms that possess the cognitive aptitude for various and time-varying aspects of the Internet of Connected Vehicles (IoCVs). This research explores Deep, dynamically coordinated edge computing and content caching, boosting Mobile Network Operator (MNO) revenues. At first, an arithmetical multi-class label, with classes ranging from C1 to Cn, is added to the pristine data set. The best data model is then generated from the labelled clean data and applied to the class of samples in data. The suggested model, dubbed DRL-FESD-CVT, employs a Feature Extraction of Shallow and Deep properties (FESD) architecture with two modules Convolutional Neural Networks (CNNs) and Vision-Transformer (V.T.) to mine the spatial-spectral properties of labelled data. To maintain spatial and spectral correlations simultaneously, we next build a Shallow Spatial-Spectral Feature Extraction (S3FE) unit using a two-layer 3D Convolutional Neural Network (CNN). Waiting time and other critical performance metrics are reduced compared to conventional methods in SUMO simulations, demonstrating the superiority of the suggested strategy. When the training data is 60%, the proposed model achieved 99.87% of accuracy and 99.86% of sensitivity, where the VT achieved 98.42% of accuracy and 98.52% of sensitivity. When the missing ratio is 20, the proposed model achieved 0.82 of accuracy and the existing VT model achieved 0.87 of accuracy.