ATPSO: Adaptive Task Priority Scheduling and Offloading Optimization Scheme for vehicles in harsh environments

In harsh environments, vehicle driving risks significantly increase. Safety enhancing applications can mitigate these risks, but can also introduce delay-sensitive and computationally intensive tasks that challenge vehicle data processing and communication. This paper proposes an Adaptive Task Priority Scheduling and Offloading Optimization Scheme (ATPSO) based on the Vehicular Edge Computing (VEC) paradigm to improve Quality of Service (QoS) for safety tasks. Firstly, a multifactor prioritization mechanism using environment awareness is developed, leveraging SHapley Additive ExPlanations (SHAP) and XGBoost models to quantify safety priorities from real-time data and assess vehicle risks. A Transformer model dynamically adjusts priority weights to minimize risk. Secondly, the local queue stability problem is addressed as a Lyapunov optimization problem, proposing a multi-queue priority scheduling and distributed resource allocation scheme. Lastly, a Markov Decision Process (MDP) model is constructed to handle dynamic computational offloading, and the Entropy-Enhanced Multi-Agent Soft Actor–Critic (EE-MASAC) algorithm is introduced to optimize offloading strategies and resource allocation. Simulation results demonstrate that ATPSO effectively reduces safety task delays, improves task completion rates, and lowers vehicle risk scores, outperforming existing methods in adaptability and performance, offering a solid foundation for practical deployment of vehicle safety applications.