Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Using Drone Swarm to Stop Wildfire: A Predict-then-optimize Approach

Shijie Pan, Aoran Cheng, Yiqi Sun, Kai Kang, Cristobal Pais, Yulun Zhou, Zuo-Jun Max Shen

TL;DR

This work proposes a mixed-integer programming (MIP) model coupled with dynamic programming (DP) to enable efficient drone swarm task planning and uses chance-constrained robust optimization (CCRO) to ensure robust firefighting performances under varying situations.

Abstract

Drone swarms coupled with data intelligence can be the future of wildfire fighting. However, drone swarm firefighting faces enormous challenges, such as the highly complex environmental conditions in wildfire scenes, the highly dynamic nature of wildfire spread, and the significant computational complexity of drone swarm operations. We develop a predict-then-optimize approach to address these challenges to enable effective drone swarm firefighting. First, we construct wildfire spread prediction convex neural network (Convex-NN) models based on real wildfire data. Then, we propose a mixed-integer programming (MIP) model coupled with dynamic programming (DP) to enable efficient drone swarm task planning. We further use chance-constrained robust optimization (CCRO) to ensure robust firefighting performances under varying situations. The formulated model is solved efficiently using Benders Decomposition and Branch-and-Cut algorithms. After 75 simulated wildfire environments training, the MIP+CCRO approach shows the best performance among several testing sets, reducing movements by 37.3\% compared to the plain MIP. It also significantly outperformed the GA baseline, which often failed to fully extinguish the fire. Eventually, we will conduct real-world fire spread and quenching experiments in the next stage for further validation.

Using Drone Swarm to Stop Wildfire: A Predict-then-optimize Approach

TL;DR

This work proposes a mixed-integer programming (MIP) model coupled with dynamic programming (DP) to enable efficient drone swarm task planning and uses chance-constrained robust optimization (CCRO) to ensure robust firefighting performances under varying situations.

Abstract

Drone swarms coupled with data intelligence can be the future of wildfire fighting. However, drone swarm firefighting faces enormous challenges, such as the highly complex environmental conditions in wildfire scenes, the highly dynamic nature of wildfire spread, and the significant computational complexity of drone swarm operations. We develop a predict-then-optimize approach to address these challenges to enable effective drone swarm firefighting. First, we construct wildfire spread prediction convex neural network (Convex-NN) models based on real wildfire data. Then, we propose a mixed-integer programming (MIP) model coupled with dynamic programming (DP) to enable efficient drone swarm task planning. We further use chance-constrained robust optimization (CCRO) to ensure robust firefighting performances under varying situations. The formulated model is solved efficiently using Benders Decomposition and Branch-and-Cut algorithms. After 75 simulated wildfire environments training, the MIP+CCRO approach shows the best performance among several testing sets, reducing movements by 37.3\% compared to the plain MIP. It also significantly outperformed the GA baseline, which often failed to fully extinguish the fire. Eventually, we will conduct real-world fire spread and quenching experiments in the next stage for further validation.

Paper Structure

This paper contains 13 sections, 1 theorem, 8 equations, 6 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methods
  3. Wildfire Spread Prediction using Convex-NN
  4. Optimal Drone Swarm Task Allocation
  5. Mixed-Integer Programming.
  6. Myopic Dynamic Programming.
  7. Chance-Constraint Robust Optimization.
  8. A Two-Stage Algorithm: Benders Decomposition and Branch-and-Cut
  9. Data and Experiments
  10. Wildfire Spread Prediction Performance
  11. Drone Swarm Quenching Algorithm Performance
  12. Path to Deployment
  13. Conclusion & Future Work

Key Result

Proposition 1

In time slot $t$, the robust chance constraint (C6) in MIP is equivalent to the following constraint sets, where $\bm{x}$, $\bm{m}$ and $\bm{D}$, are vectors with dimension $I_tJ$, for decision variable $x_{ijlt}$, auxiliary variable $m_{ijlt}$ and distance between bases and fire points $D_{ijt}$, respectively. $\bm{\mu}$ is the vector of the mean of each distribution of firefighting time at fire

Figures (6)

  • Figure 1: A conceptual diagram of drone swarm firefighting.
  • Figure 2: Real-world air-drone firefighting.
  • Figure 3: Predict-then-optimize approach flowchart.
  • Figure 4: Actual (top) and Convex-NN-S predicted (bottom) wildfire spread process for two example forest environments.
  • Figure 5: Three steps of Convex-NN-SQ predicted wildfire spread in a 40x40 forest with drone quenching: intervention on the left, prediction in the middle, and original spread on the right.
  • ...and 1 more figures

Theorems & Definitions (1)

  • Proposition 1