AI Empowered Computing Resource Allocation in Vehicular Ad-Hoc Networks

The vehicular ad hoc network (VANET) has emerged as a heterogeneous network with no fixed infrastructure. In VANET, vehicles are the mobile nodes that communicate with one another and with the infrastructure. One of the major challenges in VANET is to allocate resources efficiently due to environmental characteristics such as highly mobile vehicles, latency-sensitive and short connection times. The conventional resource management mechanisms are not very effective in such an environment. The data-driven Artificial Intelligence (AI) based techniques are a very efficient alternative to traditional statistical techniques. This project proposes an AI-empowered task offloading and computing resource allocation model which can dynamically organize the computing resources in VANET. The model is divided into two layers. First, the task offloading layer, where the Random Forest (RF) algorithm is used to determine whether the vehicle’s computing tasks should be offloaded to the cloud computing (CC) server or mobile edge computing (MEC) server or to be processed locally (in-vehicle computing). Second, the resource allocation layer, where the deep deterministic policy gradient (DDPG) algorithm is used to determine the computing platform again when the task is determined to be offloaded to either MEC servers or the cloud servers. To evaluate the performance of the RF classifier, we applied the model to a real-world driving trajectory dataset then compared the results with a different set of machine learning algorithms namely, K-nearest neighbour (KNN), Multilayer Perceptron (MLP) and Support Vector Machine (SVM). The results show the RF model outperformed other models in classification accuracy score of 99.83% for task offloading decision, where the KNN, MLP and SVM achieved 98%, 94.81% and 90.94%, respectively. Moreover, the DDPG based resource allocation scheme converges within 150 episodes. Thus, the proposed model can find the optimal offload decision and the computational resource allocation simultaneously and achieves a high delay to quality-of-service (QoS) satisfaction ratio.