A decomposition approach for the stochastic asset protection problem. (February 2022)

• Record Type: Journal Article
• Authors: Nuraiman, Dian; Ozlen, Melih; Hearne, John
• Is Part Of: Computers & operations research. Volume 138(2022)
• Publication Date: 2022-02
• Subjects: Wildfires -- Asset protection problem -- Team orienteering problem -- Two-stage stochastic programming -- Math-heuristic -- Spatial decomposition; Operations research -- Periodicals; Electronic digital computers -- Periodicals; 004.05
• DOI: 10.1016/j.cor.2021.105591
• Languages: English
• ISSNs: 0305-0548

Abstract

Abstract: The deterministic Asset Protection Problem (APP) involves deploying firefighting resources of various capabilities to service as many community assets as possible within time windows determined by an advancing wildfire. A common situation arising during these situations is a wind change. Forecasts of changes in wind velocity (i.e. direction and speed) are reasonably accurate but there is some uncertainty around the time of a wind change. The timing has implications for which areas, and hence which assets, will be impacted by the wildfire. This presents a difficult problem for an Incident Management Team (IMT) operating under severe time pressure as the wildfire sweeps across the landscape. In this study, we extend the spatial decomposition-based math-heuristic originally developed for the deterministic APP to handle large real life sized stochastic APPs in operational time. This is achieved by regarding the deterministic and stochastic components as a coupled system. A two-stage stochastic programming (TSSP) model is thus only used for a smaller sub-problem around the uncertainty. We found this approach outperformed other methods when tested on a new set of benchmarks. More importantly, the accuracy and solution times make the method suitable for operational purposes. Highlights: This work studies stochastic asset protection problem with uncertainty around the time of a wind change. A new solution approach is proposed to handle large real life sized stochastic APPs in operational time using a commercial MIP solver. The method is an extension from the spatial decomposition based math-heuristic approach by coupling the deterministic and stochastic models. The proposed method outperforms other methods tested using extensive experiments.