Towards Deep Q-Network Based Resource Allocation in Industrial Internet of Things

Area of Science:

• Computer Science
• Engineering
• Artificial Intelligence

Background:

• Industrial Internet of Things (IIoT) systems face complex resource allocation challenges due to dynamic environments and data limitations.
• Traditional optimization methods struggle with the high dimensionality and variability inherent in IIoT resource management.
• Effective resource allocation is critical for optimizing networking, computing, and energy usage in IIoT infrastructures.

Purpose of the Study:

• To propose a novel Deep Q-Network (DQN) based scheme for efficient resource allocation in Industrial Internet of Things (IIoT) systems.
• To address the dual objectives of improving bandwidth utilization and energy efficiency within IIoT environments.
• To develop a robust solution for resource management in complex, time-varying IIoT operational settings.

Main Methods:

• A Deep Q-Network (DQN) model was designed, integrating two deep neural networks (DNNs) with a Q-learning model.
• DNNs were utilized to abstract features from high-dimensional inputs and approximate the Q-function.
• The Q-learning model generated Q-tables and reward functions based on the approximated Q-function for agent decision-making.

Main Results:

• The proposed DQN scheme demonstrated significant improvements in both bandwidth utilization and energy efficiency.
• Simulations in a smart warehouse robotics scenario validated the efficacy of the DQN model.
• Experimental results showed superior performance compared to other representative resource allocation schemes.

Conclusions:

• The developed DQN-based scheme effectively optimizes resource allocation in IIoT systems.
• The approach successfully enhances bandwidth utilization and energy efficiency in practical IIoT applications like smart warehouses.
• This method offers a promising solution for managing complex resources in dynamic Industrial Internet of Things environments.