← All work
The Moving Firefighter Problem
My doctoral research: a novel graph-theory model of how fire spreads and how to contain it — solved two different ways, exact optimization and deep reinforcement learning, then compared head to head.
Read the published paperPeer-reviewed · open access (MDPI)
View paper →The problem
Wildfire containment is a race: limited resources, a fire spreading across a landscape, and firefighters who must move to reach the next defensible point. I formalized this as the Moving Firefighter Problem (MFP) — a graph model that, unlike classic firefighter problems, accounts for the time it takes defenders to travel between nodes.
What I did
- An exact solver — modeled the MFP as a mixed-integer quadratically constrained program (MIQCP) that computes provably optimal defense strategies.
- A learning solver — trained deep reinforcement-learning agents to contain fire in a dynamic cellular-automata environment, exploring how reward design changes strategy.
- A fair comparison — built an instance translator so the exact and learned approaches could be measured on identical problems, with their strengths and trade-offs made explicit.
- Published — the model and results appear in a peer-reviewed, open-access paper.
The work, in repositories
The Moving Firefighter Problem →The core model — a mixed-integer quadratically constrained program (MIQCP) that finds optimal fire-defense strategies on a graph. Basis of the published paper.Firefighter-DRL →Deep reinforcement learning for the same problem, learning containment strategies in the dynamic “Bulldozer” environment under different reward shapings.Instance Translator →The bridge between the two worlds — translates MFP instances into the Bulldozer environment so optimization and DRL can be compared on identical problems.Gym Cellular Automata →The Bulldozer benchmark itself — a cellular-automata fire-spread environment for reinforcement learning (contributed work).
Methods
PythonReinforcement LearningMathematical OptimizationMIQCPCellular AutomataGraph Theory