Attention-Shared Multi-Agent Actor-Critic-Based Deep Reinforcement Learning Approach for Mobile Charging Dynamic...

Area of Science:

• Computer Science
• Electrical Engineering
• Network Engineering

Background:

• Wireless energy transmission (WET) is crucial for wireless rechargeable sensor networks (WRSNs).
• Mobile charging via WET is a key strategy to address energy constraints in WRSNs.
• Existing research often overlooks joint optimization of charging sequence and ratio control.

Purpose of the Study:

• To propose a novel deep reinforcement learning approach for joint charging sequence scheduling and charging ratio control (JSSRC) in WRSNs.
• To address the limitations of current mobile charging scheduling methods that ignore JSSRC.
• To enhance the operational efficiency and longevity of WRSNs.

Main Methods:

• Developed an attention-shared multi-agent actor-critic-based deep reinforcement learning approach (AMADRL-JSSRC).
• Employed two heterogeneous agents: a charging sequence scheduler and a charging ratio controller.
• Designed specific reward functions considering tour length and the number of dead sensors for each agent.
• Utilized a centralized critic network with an attention mechanism for decentralized policy training.

Main Results:

• The proposed AMADRL-JSSRC approach demonstrated superior performance compared to baseline algorithms.
• Simulation results showed significant improvements in network lifetime.
• A notable reduction in the number of dead sensors was achieved.

Conclusions:

• AMADRL-JSSRC effectively optimizes JSSRC in dynamic charging environments for WRSNs.
• The method offers a promising solution for extending WRSN operational duration and reliability.
• This deep reinforcement learning framework provides a robust strategy for mobile charging management.