AI THROUGH LINK BASED ROUTING IN EV NETWORKS

Abstract

Electric Vehicles today depend a lot on wireless communication to share information with nearby vehicles
and road-side units. Small sensor modules inside EVs send details like battery level, movement, and surrounding conditions, but these sensors often run on limited energy and their connection quality changes frequently because the vehicles are always moving. Many traditional routing approaches do not perform well in such situations since they follow fixed rules and cannot adjust when the network becomes unstable.
The main idea of this project is to build a routing method that can adapt on its own. To do this, a Meta-Reinforcement Learning approach is used, where the system learns from different situations and becomes capable of choosing better routes even when the link quality or available energy keeps changing. The EV network is first simulated in MATLAB, where nodes are given three energy-harvesting abilities—solar, RF-based, and regenerative braking—so that they can gain some energy back during the simulation. The learning part of the routing is created using Python (PyTorch), and MATLAB communicates with the trained model to make routing choices in real time.
The results from the simulation show that the Meta-RL method performs noticeably better than the basic distance based or energy-aware routing. It manages to deliver more packets, uses energy more sensibly, and responds better to link fluctuations. The system consistently achieved around 83% successful packet delivery, which makes it a promising approach for future EV communication systems.