Combination with Continual Learning Update Scheme for Power System Transient Stability Assessment

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Power system transient stability assessment (TSA) increasingly relies on data-driven methods.
• Frequent changes in power system topology and operation necessitate real-time model updates for online applications.
• Existing models struggle with new scenarios due to time-varying system characteristics and limited resources for continuous retraining.

Purpose of the Study:

• To develop a sustainable and scalable method for updating TSA models in response to evolving power system conditions.
• To enhance the accuracy and adaptability of prediction models without requiring complete retraining on all historical data.
• To ensure TSA models can effectively handle new operational scenarios while retaining performance on previously encountered ones.

Main Methods:

• Introduction of the continual learning Sliced Cramér Preservation (SCP) algorithm for model updating.
• Integration of a deep residual shrinkage network (DRSN) as the classifier within the SCP framework, forming the SCP-DRSN model.
• Utilizing only new scenario data for model extension and updates, preserving knowledge of older scenarios.

Main Results:

• The SCP-DRSN model demonstrated effective updating and extension capabilities using only new scenario data on benchmark systems (New England 10-machine 39-bus and IEEE 118-bus).
• The updated model successfully improved prediction accuracy for new scenarios while maintaining performance for previously learned scenarios.
• The method showed improved coverage of new operational scenarios, addressing the challenge of dynamic power system environments.

Conclusions:

• The proposed SCP-DRSN method offers a sustainable and scalable solution for real-time TSA model updates in dynamic power systems.
• Continual learning with SCP enables adaptive TSA models that efficiently incorporate new operational data without catastrophic forgetting.
• This approach enhances the reliability and applicability of data-driven TSA for modern, complex power grids.