Cybersecurity risk assessment and defense strategies for AI-based power trading systems

Abstract

To ensure the safe, stable, and reliable operation of the artificial intelligence power trading system, and to overcome the shortcomings of existing risk assessment methods such as relying on labeled data, difficulty tracking risk causality, and poor real-time performance, this study proposes an evaluation method that combines the Time Graph Neural Network (TGNN) with hierarchical defense. This method first clarifies the four layer architecture and security sensitive points of the system, collects multiple sources of data input into the model to capture the spatiotemporal characteristics of node risks, and then quantifies the risk probability through various mechanisms to assess network security risks. According to the evaluation results, the medium and low-risk areas adopt a layered reinforcement strategy, while the high-risk areas adopt dynamic blocking measures to achieve hierarchical defense. Experiments show that the proposed method effectively evaluates security risks, maintains high spatial correlation in graph convolution, defends against diverse attack types, and achieves a high interception rate, verifying its advantages in risk assessment and defense, ensuring business continuity and system security. Clinical trial number Not applicable.