Risk-Based Machine Learning Approaches for Probabilistic Transient Stability

Authors

Umair Shahzad, University of Nebraska-LincolnFollow

First Advisor

Sohrab Asgarpoor

Date of this Version

12-2021

Citation

A dissertation presented to the faculty of the Graduate College at the University of Nebraska in partial fulfillment of requirements for the degree of Doctor of Philosophy

Major: Electrical Engineering

Under the supervision of Professor Sohrab Asgarpoor

Lincoln, Nebraska, United States, December 2021

Comments

Copyright 2021, Umair Shahzad

Abstract

Power systems are getting more complex than ever and are consequently operating close to their limit of stability. Moreover, with the increasing demand of renewable wind generation, and the requirement to maintain a secure power system, the importance of transient stability cannot be overestimated. Considering its significance in power system security, it is important to propose a different approach for enhancing the transient stability, considering uncertainties. Current deterministic industry practices of transient stability assessment ignore the probabilistic nature of variables (fault type, fault location, fault clearing time, etc.). These approaches typically provide a conservative criterion and can result in expensive expansion plans or conservative operating limits. With the increasing system uncertainties and widespread electricity market deregulation, there is a strong inevitability to incorporate probabilistic transient stability (PTS) analysis. Moreover, the time-domain simulation approach, for transient stability evaluation, involving differential-algebraic equations, can be very computationally intensive, especially for a large-scale system, and for online dynamic security assessment (DSA).

The impact of wind penetration on transient stability is critical to investigate, as it does not possess the inherent inertia of synchronous generators. Thus, this research proposes risk-based, machine learning (ML) approaches, for PTS enhancement by replacing circuit breakers, including the impact of wind generation. Artificial Neural Network (ANN) was used for predicting the benefit-cost ratio (BCR) to reduce the computation effort. Moreover, both ANN and support vector machine (SVM) were used and consequently, were compared, for PTS classification, for online DSA. The training of the ANN and SVM was accomplished using suitable system features as inputs, and PTS status indicator as the output. DIgSILENT PowerFactory and MATLAB was utilized for transient stability simulations (for obtaining training data for ML algorithms), and applying ML algorithms, respectively. Results obtained for the IEEE 14-bus test system demonstrated that the proposed ML methods offer a fast approach for PTS prediction with a fairly high accuracy, and thereby, signifying a strong possibility for ML application in probabilistic DSA.

Advisor: Sohrab Asgarpoor