Deep learning based power grid real-time operation risk automatic identification and grading method

By combining a causal inference-based dual-stream deep learning architecture with a power grid digital twin model, and using reinforcement learning optimization based on dispatcher feedback, the real-time and adaptive problems of power grid operation risk identification are solved, enabling accurate risk classification and rapid response to power grid risks.

CN122155384APending Publication Date: 2026-06-05GUIZHOU POWER GRID CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
GUIZHOU POWER GRID CO LTD
Filing Date
2026-02-03
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing technologies suffer from high computational complexity and poor real-time performance in identifying power grid operation risks, making it difficult to cover all fault combinations. Furthermore, data-driven methods have limited feature extraction capabilities, failing to delve into spatiotemporal correlations and causal logic, lacking the ability to quantitatively extrapolate the consequences of potential risk development, and the models cannot adaptively optimize.

Method used

A dual-stream deep learning architecture based on causal inference is used to extract spatiotemporal features, which are then combined with a digital twin model of the power grid for forward extrapolation. Reinforcement learning is then used to optimize the model based on dispatcher feedback to generate interpretable early warning reports.

Benefits of technology

It enables accurate, forward-looking, and adaptive automatic identification and classification of power grid operation risks, improving the accuracy of risk identification and the objectivity of decision-making, shortening response time, and increasing the safety margin of power grid operation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122155384A_ABST
    Figure CN122155384A_ABST
Patent Text Reader

Abstract

The application discloses a power grid real-time operation risk automatic identification and grading method based on deep learning, and belongs to the technical field of power system automation, which comprises the following steps: obtaining and fusing power grid real-time and historical data to generate a data set; performing double-flow feature extraction based on causal inference to generate a deep feature vector; using a power grid digital twin model to perform forward deduction to obtain a risk deduction consequence set; developing dynamic risk research and judgment to generate a risk grade and a disposal plan; generating an interpretable early warning report and updating network weights through reinforcement learning according to dispatcher feedback data. The application adopts the double-flow feature extraction based on causal inference and the digital twin deduction technology, can effectively handle the nonlinear relationship between risks and working conditions, realizes automatic identification, quantitative calculation and dynamic grading of risks, and improves the timeliness and accuracy of dispatchers in dealing with real-time operation risks and cascading failures.
Need to check novelty before this filing date? Find Prior Art