Assessing Bulk Power System Reliability by End-to-End Line Maintenance-Aware Learning

To overcome the slow-running drawback of the Monte Carlo simulation method for bulk power system reliability assessment, this paper develops an end-to-end machine learning approach to directly predict the targeted reliability index considering grid topology changes caused by emergent line maintenance. Three representative machine learning models, i.e., Support Vector Machine, Boosting Trees, and Graph Neural Network, are considered and compared. The grid topology information is embedded into the above models via two different feature engineering schemes. Details regarding dataset creation and data preprocessing are also described. Then, two case studies with different experimental settings and prediction targets are performed on the IEEE RTS-79 system to inspect the proposed approach's adaptability. Results demonstrate the proposed approach's effectiveness and speed improvement. Finally, an analysis is presented regarding the generalizability of the adopted machine learning model with respect to the varying dataset size based on the empirical-risk theory from the machine learning community.