A game theory-reinforcement learning model for optimal multi-stage resource allocation
Xiaoxiong Zhang, Xiaolei Zhou, Xinyao Xu, Hao Yan · 2025
Resource allocation plays an important role in the offense-defense game process. This problem is complex, especially involving multiple stages, attackers, and defense measures. To address the above-mentioned problem, a game theory-reinforcement learning model is used to determine the optimal resource allocation for both players. In each stage, the defender can deploy false targets (FTs) and strengthen the target, whereas the attacker can choose to identify the FTs or attack the targets randomly. Each player aims to maximize their utility by choosing appropriate actions. In specific, the Q-learning method is adopted in the game theoretic model, exploring the best resource allocation strategy over the entire planning horizon. Comparative studies demonstrate the effectiveness and feasibility of the proposed model. The model can provide certain support for the resource allocation problem.