Two-Stage Intelligent Identification of Dominant Instability Modes in Large Power Systems Based on Minimum Random Convolution Kernel Transformation and Extreme Gradient Boosting

In large power systems, large disturbance rotor angle instability and large disturbance voltage instability often occur in an intertwined and coupled manner. Accurately identifying the dominant one, i.e., the dominant instability mode (DIM), during simulation analysis is crucial for developing effective control measures. This paper proposes a two-stage intelligent identification framework for dominant instability modes in large power systems, based on the indirect mapping concept of “high-dimensional electrical features” $\rightarrow$ “lowdimensional key features” $\rightarrow$ “dominant instability model.” In the first stage, the framework uses the minimally random convolutional kernel transform (MiniRocket) to extract key features from simulation data, converting high-dimensional timeseries simulation data of large power systems into low-dimensional vector features, thereby addressing the dimensionality disaster problem of large power systems. In the second stage, the framework employs the ensemble method of extreme gradient boosting (XGBoost) to classify based on low-dimensional vector features, accurately identifying the stability and dominant instability mode of large power systems. A case study of an actual regional power grid in China (2131 nodes) demonstrates that the proposed two-stage intelligent identification method outperforms existing machine learning and deep learning methods in terms of accuracy. Additionally, the proposed method has a faster training speed and better adaptability to large power systems.