Utilizing Hybrid Deep Reinforcement Learning Approaches for Congestion Control in Vehicular Ad-Hoc Networks (VANETs)

Abstract

In the field of wireless communication, Vehicular Ad-Hoc Networks (VANETs) have recently garnered a lot of attention. Through the use of Vehicular Ad Hoc Networks (VANETs), automobiles are able to share vital information like location, direction, and speed with roadside equipment. However, traffic congestion is a common result of several vehicles using popular routes at once, therefore effective strategies for anticipating and avoiding congestion are necessary. The authors of this study propose enhancing traffic congestion forecasting capabilities in VANETs via the use of Hybrid Deep Reinforcement Learning (HDRL). This approach makes use of state-of-the-art deep learning methods including O-BiLSTM, Fuzzy Interface Systems, and Recurrent Capsule Networks (CapsRNN). For testing purposes before incorporating them into the deep reinforcement learning model, the SUMO platform is used to mimic traffic conditions. The methods of simulation and the approaches to programming are described in detail. Two critical metrics—mean travel time delay and mean vehicle waiting time delay—form the basis of the evaluation of the proposed method. Our method shows that these measures may be improved over time by running several simulations that take environmental input into account. The outcomes achieved with and without the integration of the CapsRNN, FIS, and O-BiLSTM algorithms are compared in a comparative study. Particularly in cases with moderate traffic density, the findings show that the deep reinforcement learning model is effective. Also, the article compares the algorithms' abilities to forecast traffic jams and concludes that CapsRNN is superior than FIS and O-BiLSTM.