A Transient Angle Stability Assessment Method for DFIG Grid-Connected Systems Based on Heterogeneous Graph Attention Networks

This paper presents a novel approach to assessing transient stability by utilizing a heterogeneous graph attention network (HAN). This method is proposed in response to the limitations of existing deep learning-based methods which fail to consider the intricate dynamic responses of power electronic equipment control on the transient angle stability of power systems with large-scale new energy. The proposed approach involves joint simulation modeling using the transient security assessment tool (TSAT) and real-time digital simulation system (RTDS) to divide the power grid into electromagnetic transient time scale and electromechanical transient time scale, obtaining measurement data for each generator node and variables of doubly fed induction generator (DFIG) inverter control, and establishing the generator graph topology. Then, the HAN model is applied to perform weighted aggregation of nodes, constructing the mapping relationship between the original input features and transient angle stability states. Finally, the proposed method is validated using a modified IEEE 39-bus system to demonstrate its effectiveness in transient angle stability assessment accuracy, adaptability to topological changes, and overall interpretability.