Date of Award

5-2026

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering (Holcomb Dept. of)

Committee Chair/Advisor

Dingrui Li

Committee Member

Christopher Edrington

Committee Member

Behnaz Papari

Abstract

Renewable energy production has grown significantly in recent years and continues to expand, resulting in an increasing number of inverter-based resources (IBRs) in the grid. As the number of IBRs in the grid continues to grow, grid-forming (GFM) inverters are becoming increasingly popular due to their ability to regulate voltage and frequency, providing increased grid stability and enabling islanding. As GFM inverters become more widely used in power systems, accurate and efficient models of their behavior are needed for system design purposes.

Emerging advances in neural networks have led to research on computationally efficient neural-network-based (NN-based) modeling approaches for inverters. These modeling approaches are highly accurate and very computationally efficient. However, much of this research has focused on grid-following (GFL) inverters, while NN-based modeling of GFM inverters is lacking. This work seeks to fill this gap in the literature by proposing a NN-based GFM inverter modeling approach.

The proposed modeling approach uses black-box modeling, where the following variables are the input and output of the model:

  • Input: Three-phase load current, three-phase filter inductor current, and positive and negative sequence load currents
  • Output: Three-phase load voltage 

The training data covers a wide range of power conditions and many different behaviors, including steady-state, load step, no-load, and fault conditions, both balanced and unbalanced. The training approach utilizes domain expansion to first teach the model steady-state and transient behaviors before attempting to teach fault behaviors. In the training set, the phase order of the current and voltage variables is presented in three orders (ABC, BCA, and CAB) to ensure that the model learns the symmetry between phases.

The proposed modeling approach is validated by building a NN-based model. This model can very accurately replicate behaviors that it has seen in the training set, even when it has not seen that exact power case. The model achieves accuracy greater than 97% while simulating approximately 200 times faster than a comparable average model. Since the model is a black box, it can be built, utilized, and shared without compromising the confidential internal structure of the inverter. Future research directions are provided.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.