Safety-Constrained Reinforcement Learning for Energy-Aware Transmission Scheduling in Seismic Wireless Sensor...

Area of Science:

• Wireless Sensor Networks (WSNs)
• Reinforcement Learning (RL)
• Energy Harvesting Systems
• Seismic Monitoring

Background:

• Seismic monitoring WSNs face energy constraints, leading to node failure and reduced reliability.
• Existing methods struggle to balance long-term operation, energy preservation, and data transmission.
• Premature node failure in WSNs compromises spatial coverage and seismic detection accuracy.

Purpose of the Study:

• To develop a safety-constrained RL framework for transmission scheduling in energy-harvesting seismic WSNs.
• To improve node survival and transmission success rates while adhering to energy constraints.
• To integrate a runtime safety guard layer without retraining the RL policy.

Main Methods:

• Proximal Policy Optimisation (PPO) integrated with action masking.
• A runtime guard-layer safety filter enforcing battery-preservation and load-balancing constraints.
• A scoring function combining battery headroom and network-wide load equity for safety enforcement.

Main Results:

• Guard-enhanced PPO achieved 99.46% transmission success and 66.47% node survival at 30 nodes.
• Demonstrated a 58.3% improvement in node survival over baseline methods with minimal reward reduction.
• Outperformed unconstrained PPO by 11.4% in cumulative reward, 0.8 pp in transmission success, and 15.4% in node survival.

Conclusions:

• Safety constraints, when aligned with the energy model, enhance both performance and safety in seismic WSNs.
• The guard layer effectively enforces critical constraints, proving beneficial across various network scales and event rates.
• Learned policies and safety filtering are crucial for sustained WSN operation, especially in larger networks.