QoS-Aware Intelligence Information Sharing Requests Scheduling in IoV: CPO-Based Modeling and Solution

Abstract

With the accelerated development of autonomous driving and large language model, blockchain-supported data interaction and artificial intelligence (AI)-assisted performance optimization is the current mainstream research in the Internet of Vehicles (IoV). However, the trial-and-error behavior of the AI algorithm during the training process is a threat to road safety. Therefore, this paper proposes a general constrained policy optimization (CPO)-based modeling and solution for high-dimensional constrained optimization problems. We focus on intelligent driving information sharing in blockchain-enhanced IoV and optimize the service rewards in the sharing requests scheduling problem while ensuring the frequency resource limitation, service quality constraint, and road safety constraint. The constrained state space (CSS) is innovatively proposed to abstract the environment mathematically with the definition of constraint hyperplanes and distance. Accordingly, the constrained Markov Decision process (CMDP) and the optimization problem are formulated. With the practical implementation of the CPO theory, the constrained sharing requests scheduling (CSRS) algorithm is proposed. Ablation experiments are deep reinforcement learning-based methods without using the CSS-based constraint modeling or without using the CPO-based constrained problem solving process. Results show the effectiveness of CSS and CSRS algorithm in improving the policy training efficiency, and the testing results shows excellent generalization ability.