Deep-Reinforcement-Learning-Based Joint Energy Replenishment and Data Collection Scheme for WRSN

Area of Science:

• Computer Science
• Electrical Engineering
• Network Engineering

Background:

• Wireless rechargeable sensor networks (WRSNs) enable continuous operation through mobile charging vehicles (MCVs).
• Existing WRSN strategies often fail to integrate energy replenishment and data collection efficiently.
• Unmanned aerial vehicles (UAVs) offer high mobility for versatile WRSN applications.

Purpose of the Study:

• To propose a joint energy replenishment and data collection scheme (D-JERDG) for WRSNs using deep reinforcement learning.
• To optimize the integration of data collection and wireless charging processes in WRSNs.
• To leverage UAV mobility for efficient network operation.

Main Methods:

• Network partitioning using K-means clustering.
• Cluster head selection based on remaining energy and location via an improved dynamic routing protocol.
• Shortest flight path determination using simulated annealing (SA).
• UAV control and hover point optimization using a multiobjective deep deterministic policy gradient (MODDPG) model.
• Reward function redesign for joint optimization of node death rate, UAV throughput, and energy consumption.

Main Results:

• The D-JERDG scheme effectively integrates data collection and energy replenishment.
• The MODDPG model successfully optimizes UAV flight paths and hover points.
• The proposed scheme demonstrates significant improvements in throughput, time utilization, and reduced charging costs compared to baseline methods.
• Joint optimization of multiple objectives including node survival, data throughput, and energy efficiency was achieved.

Conclusions:

• The D-JERDG scheme offers a robust solution for optimizing WRSN performance.
• Deep reinforcement learning, specifically MODDPG, is effective for complex UAV-based WRSN management.
• The proposed approach provides substantial advantages in efficiency and cost-effectiveness for WRSNs.