Application of multi-linear regression models and machine learning techniques for online voltage stability margin estimation

This paper investigates the use of multi-linear regression models (MLRMs) and machine learning techniques for online voltage stability margin prediction. The methodology relies upon the relationship between system wide reactive power reserves and voltage stability margin. A comprehensive voltage stability assessment considering an extensive contingency list and several load increase directions is performed. Data regarding reactive power reserves and voltage stability margin are stored for further MLRM development. Once properly designed and validated, the MLRMs are ready to be used in the online environment. As a few models are necessary to represent all contingencies in the list, an identification tool named MLRM-IDtool is necessary to identify what model to use based on current system conditions. Decision trees and neural networks are tested as classification tools to identify which multi-linear regression model to use. The methodology is tested in the IEEE 30 bus system with promising results. It will be shown that the two-stage proposed approach can successfully estimate voltage stability margin in the online environment and also handle uncertainty related to load behavior.