A Simulated Annealing-Based Task Offloading and Delay Optimization Method for Vehicular Fog Computing

Abstract

With the rapid increase in real-time computational demands from in-vehicle applications, traditional cloud computing is often unable to meet the millisecond-level response requirements of the Internet of Vehicles due to transmission delays. Vehicular fog computing, which integrates edge infrastructures and idle resources from nearby vehicles, presents an effective paradigm for enabling low-latency services. However, optimizing task offloading in such dynamic environments remains challenging due to the high mobility of vehicles, time-varying wireless channel states, and the highly heterogeneous computational capabilities of available fog nodes. This paper addresses the delay minimization problem in task offloading for Vehicular Fog Computing (VFC) by constructing a fine-grained system model. Tasks are modeled as divisible and heterogeneous subtasks, with factors such as distance, channel state, and computational resources comprehensively considered. The problem is formulated as a mixed-integer nonlinear programming problem. To solve it, an intelligent offloading algorithm based on Simulated Annealing (SA) is proposed. The algorithm regulates the search process using temperature parameters—conducting extensive exploration of the solution space at high temperatures to avoid local optima, and gradually refining the search as temperature decreases to converge to a near-globally optimal task allocation strategy. Simulation experiments are conducted under various task scale scenarios, and the results demonstrate that the SA-based algorithm consistently achieves lower latency across different configurations. Compared with greedy and random algorithms, the proposed method significantly reduces delay and offers greater improvements over local execution. In the scenario with a task data size of 100 Mb and a computational workload of 250 TFLOPS, the latency of the SA algorithm is only 59.0% of that of the greedy algorithm and 44.8% of that of the random algorithm. This study validates the effectiveness of the SA algorithm for task offloading optimization in VFC environments, providing a low-latency solution for real-time Internet of Vehicles applications and offering meaningful insights for the development of intelligent transportation systems.