Data-centric approach for online P-margin estimation from noisy phasor measurements

Abstract

A new estimation method for load P-margin of transmission systems is proposed by using machine learning techniques. The estimation solution uses a reduced number of features as inputs to the machine learning algorithm and does not rely on power flow measurements, avoiding using time-varying grid parameters. The method involves investigating the performance of several machine learning algorithms to undertake the estimation task and explore different data transformation processes, including an optimized feature selection scheme, enabling an enhanced performance of the machine learning algorithms. Moreover, the method comprises the use of different Explainable-AI approaches to better understand the behavior of the solution. The method’s performance for different noise levels is widely studied by employing a noise model available in the recent technical literature. The mean absolute percentage error - MAPE and the root mean square error - RMSE are calculated for performance assessment. Numerical examples of the proposed technique are presented using the IEEE 14-bus test system, considering normal and contingency (N-1,N-2) conditions for a wide range of load cases.