Optimizing Vehicular Fog Computing With Novel Joint Task Scheduling Algorithm Based on Hybrid Genetic Metaheuristics

ABSTRACT To address the requirement of latency‐sensitive intelligent transportation applications, vehicular fog computing has proven to be a promising paradigm to push the computation towards the vehicles. However, efficient task offloading and scheduling is a challenging optimization problem because of the highly dynamic nature of vehicular environments, heterogeneous resource availability, task dependency, and varying network conditions. To address these issues, this paper proposes a joint task optimization framework that simultaneously optimizes task‐to‐node assignment and execution scheduling within a unified optimization model. The task scheduling problem is formulated as a two‐dimensional bin‐packing variant capable of capturing both temporal execution constraints and computational resource limitations. The proposed framework integrates a heuristic‐guided genetic algorithm for global exploration with simulated annealing for local refinement, enabling an effective balance between exploration and exploitation. Furthermore, a cost‐aware heuristic mutation strategy is introduced to improve workload balancing while reducing communication and computation costs. The framework also incorporates dynamic task adaptation to support real‐time task arrivals and changing vehicular network conditions. Extensive simulations are conducted under varying vehicular densities, workload intensities, queue stability conditions, and dynamic task arrival rates. Experimental results demonstrate that the proposed framework consistently outperforms existing approaches in terms of response time, energy consumption, task completion time, offloading cost, and load balancing efficiency. In particular, the proposed achieves an average response time of 0.42 s, an energy consumption of 1200 J, an offloading cost of 100 vehicles, and a task completion time of 620 s for 200 tasks. Scalability analysis also validates the stable performance up to 300 vehicles and 500 computational tasks, while the confidence interval analysis shows the robustness and reliability of the optimization process.