Task offloading decision making for IoV based on deep reinforcement learning.

Area of Science:

• Intelligent Transportation Systems
• Edge Computing
• Machine Learning

Background:

• In-vehicle applications face resource limitations (computing, storage, energy).
• Cloud-edge collaborative computing is a key solution, but existing schemes lack partial offloading and task prioritization.
• Determining optimal offloading rates and balancing resource allocation remain challenges.

Purpose of the Study:

• To address limitations in current cloud-edge collaborative task offloading schemes.
• To develop a novel task offloading decision scheme for dynamic vehicle environments.
• To enable efficient partial offloading and rational resource allocation based on task priorities.

Main Methods:

• Design of communication, energy consumption, cost, priority, and task offloading models.
• Implementation of a task offloading decision scheme using deep reinforcement learning algorithms.
• Development of an Improved Deep Deterministic Policy Gradient (IDDPG)-based scheme.

Main Results:

• The proposed IDDPG-based scheme significantly optimizes performance compared to existing methods.
• Achieved a 59.46% latency reduction compared to DQN and 67.39% compared to DDPG.
• Reduced energy consumption by 18.37% compared to DQN and 11.76% compared to DDPG.

Conclusions:

• The IDDPG-based scheme effectively handles partial offloading and task prioritization in cloud-edge environments.
• The proposed approach offers superior latency and energy efficiency for in-vehicle computing.
• This research provides a robust solution for optimizing resource utilization in connected vehicles.