Response-Based Adaptive Under-Voltage Load Shedding Using Oblique Energy Decision Tree

This paper develops a response-based under-voltage load shedding (UVLS) approach using a novel learner, i.e., energy decision tree (DT). Focusing on short-term voltage stability, post-contingency trajectories of U/P/Q quantities are acquired by PMUs as time series, which are further converted to tractable distance data via shapelet (representative sub-sequence) transform. Based on the oblique DT algorithm, an oblique hybrid DT involving U/P/Q quantities is built to reliably detect the unstable cases that need to be corrected by UVLS. Inspired by the physical significance of P time series based distance values, i.e., energy differences, a compact oblique energy DT is induced from multi-variable energy deviations. After that, the UVLS strategy is adaptively constructed by solving a simple nonlinear optimization problem whose constraints are specified by the energy DT. Numerical tests on the realistic Hong Kong power grid illustrate the effectiveness and reliability of the proposed approach.