A computational offloading optimization scheme based on deep reinforcement learning in perceptual network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Network Engineering

Background:

• The integration of Internet of Things (IoT) and edge computing enhances the computational power of the IoT perception layer.
• Current edge computing offloading techniques often lack dynamic policy adjustment, leading to inefficiencies.
• Solidified offloading policies struggle to adapt to the fluctuating network conditions inherent in the IoT perception layer.

Purpose of the Study:

• To investigate an optimized computation offloading scheme for the IoT perception layer.
• To address the limitations of static offloading policies in dynamic edge computing environments.
• To leverage deep reinforcement learning for adaptive task offloading in IoT networks.

Main Methods:

• Developed a computation offloading optimization scheme utilizing deep reinforcement learning.
• Implemented an algorithm capable of adaptively adjusting computational task offloading policies for IoT terminals.
• Simulated and experimented with the proposed scheme under varying network conditions within the perception layer.

Main Results:

• The proposed algorithm demonstrated adaptive adjustment of offloading policies based on real-time network changes.
• Significant improvements in the operational efficiency of the IoT perceptual layer were observed.
• A notable reduction in average task delay was achieved compared to existing offloading algorithms.

Conclusions:

• Deep reinforcement learning provides an effective framework for adaptive computation offloading in the IoT perception layer.
• The developed scheme enhances IoT system performance by dynamically optimizing task distribution.
• This approach offers a viable solution for managing computational tasks in complex and evolving IoT environments.