Machine Learning-Based Workload Prediction in Vehicular Platooning Systems

As the demand for self-driving vehicles increases, efficiently scheduling vehicular tasks to maximize battery life and ensure timely completion is crucial. Vehicle platooning, where multiple vehicles travel together as a unit, enhances fuel efficiency, reduces congestion, and increases safety by maintaining optimal distances between vehicles. In this work, we categorize each vehicle's tasks into the Delay-Sensitive, Bounded-Delay, and Delay-Tolerant categories. However, platooning introduces significant computational challenges, such as real-time data handling, vehicle communication, and adapting to dynamic environments. These challenges necessitate advanced computational methods for effective task management. Machine Learning (ML) techniques are employed to determine the most suitable server for offloading tasks. This decision is supported by a workload prediction algorithm that forecasts the next slot's workload. We develop a task offloading approach using Random Forest models to predict workload, considering current server attributes and historical data. This enables us to schedule tasks on the best server in a dynamic environment. Our simulation results demonstrate significant improvements (up to 90%) in task success rates. Results show that our work can contribute towards developing more reliable and efficient autonomous driving systems by optimizing resource utilization and ensuring timely task completion in dynamic environments.