A Novel Ensemble Learning-Based Method for Online Total Transfer Capability Assessment of New Power System With Increased Stochastics

With the ever-increasing penetration level of renewable energy and emerging electricity demand, it is very important to ensure the safe and reliable operation of the modern transmission system with strong stochastics and dynamics. Exceeding transmission limits will lead to line overloading, equipment damage, voltage instability, and even cascading failures. Therefore, it is of great significance to accurately evaluate the total transfer capability (TTC) in real-time. This paper proposes a novel ensemble model-based method to learn the nonlinear mapping between online operating conditions and TTCs of key transmission corridors. Massive representative operational samples are generated using a high-fidelity power system security analysis engine involving a variety of operating conditions, and thermal stability, static voltage security, and transient stability are considered. After feature extraction and label labeling of these samples, the dataset was split into training and test sets. The samples were used to predict TTC using three types of models: Light Gradient Boosting Machine (LGBM), Extreme Gradient Boosting (XGBoost), and Random Forest (RF). The hyperparameters of these models were optimized through random search and 50% discounted cross-validation methods to enhance the accuracy of TTC estimation. Finally, the optimized models were combined using the stacking method. In this approach, LGBM, XGBoost, and RF served as base models, while LGBM functioned as the meta-model, leveraging the strengths of each model. The proposed method was validated on a 500-node network model with real-world operational characteristics, demonstrating improved TTC evaluation accuracy and higher efficiency.