Coevolution of players strategies in security games

Stackelberg Security Games (SSGs) gained recently a lot of attention and popularity due to a bunch of successful practical applications in the field of security maintenance. SSGs model real-life security scenarios as non-cooperative games between the security forces (e.g. secret service, police) and the attackers (e.g. terrorists, military groups). The paper proposes a novel coevolutionary method (CoEvoSG) for solving SSGs that develops two competing populations of player strategies, in the process inspired by biological evolution, so as to approximate the Stackelberg Equilibrium (game solution). CoEvoSG is experimentally evaluated on over 800 test instances of three types of games with various characteristics. The results and their detailed analysis presented in the paper prove the CoEvoSG ability to repetitively find optimal or close to optimal solutions with time scalability excelling the state-of-the-art methods. Consequently, CoEvoSG is capable of calculating solutions for games bigger and more complex than ever before. This study extends our previously published conference paper Zychowski and Mandziuk (2022).