Optimal strategy selection of network defense based on evolutionary game and reinforcement learning

In the field of cybersecurity, the frequency of changes in the network environment is a critically important factor for the defender to consider. To address this challenge, a network attack and defense model based on evolutionary game theory and reinforcement learning has been proposed. Its primary aim is to analyze how the defender can optimally deploy defense strategies in a network environment that experiences both relative stability and highly dynamic changes. In the evolutionary game theory component of the attack-defense model, an attack-defense matrix is used to quantify the gains of both the attacker and defender. By examining the evolutionary stable states of both sides in relatively stable network environments, the optimal defense action for the defender is determined. In the reinforcement learning part of the attack-defense model, the Q-learning algorithm is employed, allowing the defender to interact with the constantly changing network environment. This enables the defender to acquire the best network defense strategies, especially when the network environment undergoes frequent fluctuations. Through a series of simulation experiments, the effectiveness of the proposed model and algorithms has been validated, significantly enhancing the efficiency of the defender in network security.