Research on Dynamic Risk Assessment Methods for the New Power System Based on Deep Reinforcement Learning

With the advancement of digital transformation of the power grid, the deep coupling characteristics of information and physics pose complex network security threats to the new power system. This article proposes a dynamic network security risk quantification and evaluation model based on DQN, and combines joint simulation and network attack chain analysis to reveal the multi-dimensional network security risk propagation mechanism of the new power system. The experiment shows that the model has a risk prediction error of less than 5% for network attacks such as denial of service. Based on simulation results, this article also proposes optimization suggestions for dynamic defense strategies, providing theoretical support for the security protection of the new power system.