Physics-Informed Learning Based Wind Farm Two-Machine Aggregation Model for Large-Scale Power System Stability Studies

Aggregation of wind turbines (WTs) in wind farms (WFs) can reduce modeling and computation burden, but it may also reduce accuracy. Furthermore, it may be difficult to accurately determine the dynamic behaviors of WTs under power system disturbances. This paper proposes a novel aggregation modeling method of WFs for power system transient analysis, based on a two-stage approach. In the first stage, a dendrogram algorithm generates a simple and generic model (GM), while in the second stage, the GM is refined using a WF-tailored partial differential equation functional identification of nonlinear dynamics (PDE-FIND) algorithm to improve the accuracy of the initial GM. The dynamic library of the PDE-FIND algorithm is reformulated to contain variables that are likely to be used in expressing the power error equations. A requirements-oriented algorithm is also proposed to extract the most critical variables and generate a precision-adjustable aggregation model (AM) that balances accuracy and simplicity. The effectiveness of the proposed method is validated by extensive comparisons between GMs and the proposed AM.