Enhancing Energy Efficiency in Connected Vehicles through Multi-armed Bandits-Integrated Digital Twins

Task offloading and resource management in Vehicular Social Networks (VSNs) are critical for optimizing system performance and enhancing energy efficiency. This paper proposes a digital twin-based task offloading framework that offloads computationally intensive tasks to virtual counterparts within the network. The vehicle's mobility, fluctuating network conditions, and availability of variable devices create many challenges for task offloading in VSNs. To address these challenges, we introduce a deep learning-based algorithm designed to minimize system latency and maximize task execution efficiency. The system uses a Multi-armed Bandits algorithm with ϵ-greedy approach to establish a reward-based mechanism examining system latency and energy consumption. The algorithm learns through multiple iterations and assigns higher rewards to the DTs with lower waiting time and energy usage. This enables the Digital Twins (DTs) to make informed offloading decisions favoring nodes. The simulations demonstrate that our framework reduces system delays, improves the efficiency of task execution, and promotes energy conservation by optimizing the resource utilization and consistently distributing the energy consumption among all nodes. Unlike the traditional use of classical MAB strategies, this research extends the task offloading framework to the VSN domain, incorporating energy efficiency with a CO2 emission model offering real-time adaptability.