Existing research on moving target defense (MTD) decision-making methods predominantly relies on game models under the assumption of complete rationality, failing to adequately account for the bounded rationality of attackers and defenders. Additionally, traditional MTD strategies face issues of rigidity and resource inefficiency in dynamic environments, rendering them ineffective against continuously evolving adversarial scenarios. To address these limitations, this paper proposes a dynamically self-adaptive decision-making method based on supermodular games under bounded rationality. First, the adversarial process is modeled using a supermodular game framework that incorporates prospect theory to capture bounded rational behavior. Furthermore, a constrained payoff function that balances MTD security, performance, and affordability is designed. Second, an innovative Prospect-Theoretic Multi-Agent Advantage Actor-Critic (PT-MAA2C) algorithm is proposed to solve the game, leveraging its adaptive and efficient strategy exploration and exploitation capabilities to derive optimal equilibrium solutions. Finally, through experiments simulating MTD methods against DDoS attacks in a network topology deception scenario, the proposed method is demonstrated with significantly enhanced defense efficiency, system robustness, and resource optimization, offering both theoretical and practical value.