Vehicle Trajectory Prediction Method for Task Offloading in Vehicular Edge Computing.

Area of Science:

• Computer Science
• Artificial Intelligence
• Vehicular Computing

Background:

• Vehicular Edge Computing (VEC) offers real-time computation for vehicles, but task offloading efficiency impacts Quality of Service (QoS).
• Predictive task offloading in VEC faces limitations due to resource constraints and the need for timely vehicle trajectory data.
• Deploying machine learning models on edge servers for trajectory prediction presents significant challenges.

Purpose of the Study:

• To propose a novel vehicle trajectory prediction method for enhancing task offloading in VEC.
• To improve the accuracy and efficiency of trajectory prediction for VEC task offloading.
• To design task offloading strategies and optimization algorithms that minimize energy consumption within time constraints.

Main Methods:

• Constructing a T-pattern prediction tree (TPPT) using historical vehicle trajectory data.
• Identifying the vehicle frequent itemset with the largest support within the TPPT for trajectory prediction.
• Real-time updating of the TPPT with predicted trajectory results and designing energy-efficient task offloading strategies.

Main Results:

• The proposed prediction method demonstrated over 10% improvement in accuracy compared to the baseline T-pattern method.
• Prediction efficiency was enhanced by over 6.5 times, significantly outperforming the baseline.
• Experimental validation on real-vehicle and Capital Bikeshare datasets confirmed the method's effectiveness.

Conclusions:

• The vehicle trajectory prediction method based on frequent patterns offers high accuracy and efficiency for VEC task offloading.
• This approach effectively addresses the challenge of vehicle trajectory prediction in dynamic VEC environments.
• The optimized task offloading strategies contribute to reduced energy consumption while meeting time constraints.