Tree-Based Ensemble Machine Learning Techniques for Power System Static Security Assessment

Emerging machine learning techniques (MLTs) have proven to be effective in many applications including power system studies. In this paper, the performance of tree-based ensemble MLTs has been investigated to assess the static security of power system and compared with base-line methods. System wide composite security index is used to assess the severity of a contingency against bus voltage and power flow violations. The scope of study is limited to widely used base line estimators: multilayer perceptron model, support vector machine, K-nearest neighbors, decision tree and ensemble-based estimators: random forest, AdaBoost, gradient boost, and extra tree. Performance analysis is done in terms of accuracy, classification rate and training time. To further investigate the suitability of these approaches to time-constrained security environment at Energy Management Center, recursive feature elimination (RFE) method is used to contain the curse of dimensionality and performance evaluation is done on test systems extending from IEEE 14-bus to IEEE 118-bus including practical 75-bus Indian system. Ensemble-based methods in general and ETC in particular outperformed base line methods in terms of accuracy and computational time with RFE.