Method and device for intrusion detection of power information system based on multi-dimensional situation
By constructing a multi-dimensional intrusion detection method for power information systems, and utilizing Transformer and graph neural network encoders, the problem of insufficient multimodal data fusion in power grid information security is solved, achieving high-precision detection of complex attacks and identification of low-frequency attacks.
Patent Information
- Application Number
- CN202610754375.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-28
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, power grid information security detection methods rely on a single data source, making it difficult to detect complex attack behaviors. Furthermore, the lack of feature interaction modeling during multimodal data fusion leads to a high false negative rate, making it difficult to cope with the complex and ever-changing security threat situation of the power grid.
A multi-dimensional situational intrusion detection method for power information systems is constructed. Through multi-source data preprocessing and feature extraction, a dedicated encoder is built using Transformer and graph neural network to encode network behavior, system operating status and security device context features, respectively. Feature recalibration and dimensionality reduction are performed through a fusion network, and the model is trained by combining weighted cross-entropy loss.
It achieves high-precision detection of complex attacks on the power grid, improves the ability to fuse multimodal information, reduces the risk of overfitting, increases the detection rate of low-frequency attacks, and supports online updates and integration into existing power grid monitoring platforms.
Smart Images
Figure CN122419955A_ABST