734. Deep Reinforcement Learning for Fire Prevention
Invited abstract in session TE-36: Analytics for Forest Fires Prevention and Risk Analysis, cluster Use of Analytics in Forest Fires Management.
Tuesday, 16:15-17:45Room: FENH309
Authors (first author is the speaker)
| 1. | Lucas Murray
|
| Department of Industrial Engineering, Universidad de Chile | |
| 2. | Andrés Weintraub
|
| Industrial engineering, University of Chile | |
| 3. | Jaime Carrasco
|
| University of Chile | |
| 4. | Tatiana Andrea Castillo Jaimes
|
| Department of Industrial Engineering, Universidad de Chile |
Abstract
Wildfires have increased in severity and frequency, primarily due to climate change, turning them into one of the most destructive forces in nature. Designing and constructing landscapes to withstand fires has become a critical practice, and, therefore, the development of sophisticated tools to support decision-making in this area is increasingly pressing. In this regard, we propose using one of the most novel AI paradigms, Reinforcement Learning, to solve the firebreak allocation problem. Through the use of the fire-spread simulator Cell2Fire and Deep Neural Networks, we programmed an intelligent agent to learn how to position firebreaks in a small forest, achieving near-optimal performance. Furthermore, we generated relatively good solutions using known heuristics and pre-trained the agent with them, resulting in equal or superior performance, suggesting significant potential in the use of intelligent agents to solve OR-tasks, particularly in settings such as fire-prevention. We could solve instances of up to 20x20 cells, demonstrating convergence with good results. This is one of the first times Reinforcement Learning has been used to solve the problem.
Keywords
- Artificial Intelligence
- Climate and Disaster Risk Management
- Agent Systems
Status: accepted
Back to the list of papers