EcoWild: Reinforcement Learning for Energy-Aware Wildfire Detection in Remote Environments.

Area of Science:

• Cyber-physical systems
• Artificial intelligence
• Environmental monitoring

Background:

• Early wildfire detection is crucial for remote areas but challenging due to limited connectivity and energy.
• Existing systems struggle with energy constraints and require frequent maintenance or cloud access.
• Autonomous, long-term operation is essential for effective wildfire surveillance in inaccessible locations.

Purpose of the Study:

• To introduce EcoWild, an energy-adaptive cyber-physical system for autonomous wildfire detection.
• To enable sustainable operation of edge devices using solar power and reinforcement learning.
• To dynamically adjust sensing and communication strategies based on real-time environmental and energy data.

Main Methods:

• Developed a reinforcement learning agent to manage sensing and communication on solar-powered edge devices.
• Integrated a decision tree-based fire risk estimator and on-device smoke detection.
• Modeled solar energy harvesting, battery dynamics, and communication costs for realistic simulations.

Main Results:

• EcoWild maintained system responsiveness and avoided battery depletion across diverse conditions.
• Achieved 2.4× to 7.7× faster wildfire detection compared to static baseline systems.
• Demonstrated moderate energy consumption and system reliability in 125 simulated deployment scenarios.

Conclusions:

• EcoWild offers a sustainable and effective solution for autonomous wildfire detection in remote, energy-constrained environments.
• Reinforcement learning enables adaptive resource management for long-term operation of edge devices.
• The system significantly improves detection speed and reliability, crucial for mitigating wildfire impacts.