Optimal Patrol Planning for Green Security Games with Black-Box Attackers

摘要: Motivated by the problem of protecting endangered animals, there has been a surge interests in optimizing patrol planning for conservation area protection. Previous efforts these domains have mostly focused on routes against specific boundedly rational poacher behavior model that describes poachers’ choices areas to attack. However, algorithms do not apply other poaching prediction models, particularly, those complex machine learning models which are recently shown provide better than traditional bounded-rationality-based models. Moreover, previous handle important concern whereby poachers infer partially monitoring rangers’ movements. In this paper, we propose OPERA, general framework that: (1) generates optimal implementable patrolling black-box attacker can represent wide range models; (2) incorporates entropy maximization ensure generated more unpredictable and robust partial monitoring. Our experiments real-world dataset from Uganda’s Queen Elizabeth Protected Area (QEPA) show OPERA results defender utility, efficient coverage unpredictability benchmark past used rangers at QEPA.