Curiosity Driven Resource Allocation for 5G and Beyond Vehicular Networks

Abstract

With the rapid advancement of the Internet of Vehicles (IoV), there arises an increasing demand for efficient connectivity and communication mechanisms between vehicles and infrastructures, wherein resource allocation assumes paramount importance. The primary objective of a resource allocation algorithm is to distribute limited resources, including power and spectrum, to mobile devices within the network while catering to the diverse requirements of users. In this thesis, we introduce a novel approach called the Intrinsic Curiosity Module (ICM) based Double Q Learning (DQL) for resource allocation, denoted as ICM-DQRA, aimed at addressing resource allocation challenges in IoV network. We integrate the ICM into the DQL algorithm to incorporate an intrinsic reward to the agent. This intrinsic reward, absent in most reinforcement learning algorithms, serves to incentivize the agent to explore the environment further and make decisions conducive to better rewards. Through comprehensive simulations, our proposed method outperforms other approaches, such as the greedy method and DQL method. Specifically, the ICM-DQRA algorithm achieves a more efficient resource allocation among vehicles, leading to a substantial reduction in energy consumption across the network, ranging from 20% to 27%.