Research on Dynamic Model of Security Situation of Power Information Network in Smart Grid

Abstract

With the rapid development of smart grid, power information network faces increasingly severe security threats. In smart grid, real-time security situation awareness is of great significance for preventing potential attacks. Most of the existing security situation assessment methods are based on rules or statistical analysis, lacking the deep understanding of time series data and the ability to dynamically adjust. This paper proposes a time series deep reinforcement learning (TDRL) algorithm to dynamically evaluate the security situation of power network. The TDRL algorithm constructs a reinforcement learning framework, regards the security state of power information network as the environment state, defines attack and defense strategies as actions, and combines historical data of security events for learning and optimization. Through training, the model can predict the changes in security situation in real time, and adjust according to the actual risks to improve the anti-attack capability of the power grid. The experimental part uses a real smart grid data set for simulation. The model performs better than the traditional rule-based evaluation method in multiple attack scenarios. In the comparison experiment, the TDRL algorithm has significantly improved in accuracy, detection rate and F1 value, with an accuracy rate of more than 95% and a false alarm rate of less than 5%.