Efficient DoS Attack Defense Strategy for Blockchain Networks Based on Bayesian Attack Graph and Stackelberg Game

With their decentralized and immutable properties, blockchains are widely adopted across diverse domains. However, clients remain vulnerable to distributed denial-of-service (DDoS) attacks. Specifically, adversaries can further disrupt service availability and undermine the reputation by exhausting honest node resources. However, existing defenses are limited by idealized assumptions about attack paths and weak adaptability to dynamic attacker behaviors. To address this issue, we propose a DoS defense framework for blockchain networks integrating Bayesian Attack Graphs (BAG), Stackelberg game, and reinforcement learning. BAG models quantify node reachability probabilities via Bayesian inference, considering vulnerability severity, attack cost, and attack benefit, thus capturing attackers' motives and randomness in attack path selection. A Stackelberg game simulates attacker-defender strategies, while reinforcement learning dynamically refines defenses in real time. Extensive experiments demonstrate that our method outperforms both the baseline model and Q-MIND in terms of performance, specifically showing a 47% faster convergence speed and a 32% higher stable reward. Additionally, this method can achieve a 97% node survival rate, whereas this proportion is only 15% in the absence of defense.