Scalable Impedance Identification of Diverse IBRs via Cluster-Specialized Neural Networks

Quang Manh Hoang 1 , Guilherme Vieira Hollweg 1 , Bang Nguyen 3 , Akhtar Hussain 2 , Wencong Su 1,* , Van-Hai Bui 1,* · 2026

Read Scalable Impedance Identification of Diverse IBRs via Cluster-Specialized Neural Networks by Quang Manh Hoang 1 , Guilherme Vieira Hollweg 1 , Bang Nguyen 3 , Akhtar Hussain 2 , Wencong Su 1,* , Van-Hai Bui 1,* (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-23203.txt.

Introduction

Hoang and co-authors tackle impedance identification for inverter-based resources in modern power grids. These devices use varied controls, so their impedance behavior differs a lot from unit to unit. Training one model per device does not scale well. The team groups devices with K-means clustering, then trains a small feed-forward network for each cluster. In use, a few measurements predict impedance across many operating points. Tests on six devices, plus one unseen device with just ten points, show high accuracy.

Catalog record

CREATORQuang Manh Hoang 1 , Guilherme Vieira Hollweg 1 , Bang Nguyen 3 , Akhtar Hussain 2 , Wencong Su 1,* , Van-Hai Bui 1,*
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20196
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.23203

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Scalable Impedance Identification of Diverse IBRs via Cluster-Specialized Neural Networks Corresponding authors: V.H. Bui ([email protected]) and W. Su ([email protected]). Quang Manh Hoang 1 , Guilherme Vieira Hollweg 1 , Bang Nguyen 3 , Akhtar Hussain 2 , Wencong Su 1,* , Van-Hai Bui 1,* Abstract Modern machine learning approaches typically identify the impedance of a single inverter-based resource (IBR) and assume similar impedance characteristics across devices. In modern power systems, however, IBRs will employ diverse control topologies and algorithms, leading to highly heterogeneous impedance behaviors. Training one model per IBR is inefficient and does not scale. This paper proposes a scalable impedance identification framework for diverse IBRs via cluster-specialized neural networks. First, the dataset is partitioned into multiple clusters with similar feature profiles using the K-means clustering method. Then, each cluster is assigned a specialized feed-forward neural network (FNN) tailored to its characteristics, improving both accuracy and computational efficiency. In deployment, only a small number of measurements are required to predict impedance over a wide range of operating points. The framework is validated on six IBRs with varying control bandwidths, control structures, and operating conditions, and further tested on a previously unseen IBR using only ten measurement points. The results demonstrate high accuracy in both the clustering and prediction stages, confirming the effectiveness and scalability of the proposed method. † † aftertitle: I Introduction Modern power grids are expected to integrate an increasing number of inverter-based resources, which makes the system more vulnerable and reduces the applicability of traditional control and stability analysis methods. One of the major challenges is the complex interaction among multiple IBRs due to their wide range of control bandwidths, topologies, and algorithms, together with the reduced inertia compared to traditional synchronous-generator (SGs)-based power systems. These interactions can lead to small-signal stability problems such as sub-synchronous oscillation, near-synchronous oscillation, and super-synchronous resonance [ 10 ] . In traditional power systems dominated by SGs, small-signal stability can be studied using white-box state-space models, because SGs have standardized structures and known parameters; therefore, eigenvalue analysis can be directly applied to investigate the system stability [ 6 ] . However, this approach is not applicable to inverter-dominated power systems, where the control structures and parameters of IBRs are often kept confidential by power converter manufacturers due to security and intellectual property considerations [ 12 ] . As a result, impedance-based stability analysis, which is well studied in the literature, has emerged as a powerful tool for assessing the small-signal stability of black-box models in the frequency domain [ 10 , 15 , 7 ] . In this method, the converter is treated as a black-box impedance model that relates voltage and current perturbations, typically obtained by injecting current or voltage at selected frequencies and measuring the corresponding responses [ 2 ] . Various impedance measurement methods based on frequency scanning have been developed for dc and ac converters. However, measurement-based methods are known to be tedious and time-consuming, especially for large inverter-based power systems and wide frequency ranges. Even for a single IBR with detailed control structure, it may take several hours to obtain wide-frequency-range impedance characteristics at multiple operating points, and highly intermittent inverter-dominated systems require many such operating scenarios to be analyzed, which further increases the computational burden and complexity. To overcome these drawbacks, several studies have focused on using data-driven methods to identify black-box IBR impedance models at multiple operating points [ 5 , 14 , 9 , 7 ] . In [ 14 ] , an artificial neural network (ANN) was used to identify the impedance model of a single grid-following inverter (GFLI) at different operating points; however, this model cannot predict impedance for converters with different control topologies, such as grid-forming inverters (GFMIs), without collecting a large amount of additional data. A related work in [ 7 ] proposed a machine learning end-to-end framework based on transfer learning to identify GFLI impedance models at multiple operating points and with different control bandwidths. Nevertheless, the prediction accuracy of that model degrades significantly when applied to GFMIs, whose impedance characteristics differ from those of GFLIs. In [ 9 ] , the authors used a support vector machine algorithm to predict the impedance of current-control GFLI, power-control GFLI, and virtual-synchronous-generator-based GFMI, but the model is retrained separately for each IBR rather than trained jointly. Notably, many existing machine-learning-based impedance modeling approaches struggle to generalize across highly heterogeneous IBR topologies. Although different IBRs may share identical input structures and operating variables, their fundamentally different dynamic behaviors lead to distinct impedance characteristics, which often results in degraded prediction accuracy when a single unified model is applied. To address these limitations, this paper proposes a scalable impedance identification framework for diverse IBRs via cluster-specialized neural networks. The primary contribution lies in a system-level framework that addresses a practical and increasingly important challenge in large-scale power-electronic-dominated grids: scalable impedance identification under device heterogeneity. First, the dataset of six IBRs (three GFMIs and three GFLIs) with varying control bandwidths, control structures, and operating conditions is generated from analytical models. This dataset is then partitioned into multiple clusters with similar features using the K-means clustering method. Each cluster is subsequently assigned and trained with a specialized feed-forward neural network tailored to its feature profile, improving both accuracy and computational efficiency. In deployment, an unseen GFLI4 is tested using only ten measurement points at a single operating point. The results demonstrate high accuracy in both the clustering and prediction stages, confirming the effectiveness and scalability of the proposed method. II Proposed Framework Fig. 1: A clustering-driven framework for impedance identification in multi-IBR grids. Figure 1 depicts the proposed framework, consisting of an offline phase (clustering and model training) and an online prediction phase. In the offline phase, impedance datasets for multiple IBRs are obtained either via measurement-based identification or from analytical models over a wide range of operating points (see Subsection II-A ). After assembling the dataset, one or more salient features (e.g., the Y_{dd} component) are used for clustering (Subsection II-B ). Since the resulting clusters exhibit distinct characteristics, each cluster is assigned a dedicated feed-forward neural network architecture, as detailed in Subsection II-C . During deployment, the impedance of an unseen IBR can be identified from only a few measurements: first assign the device to its most likely cluster, then use the corresponding network to predict impedance across the broader operating range. II-A Data Acquisition II-A 1 Generate dataset For model training, datasets can be obtained from offline or hardware-in-the-loop simulations; from field devices using measurement-based methods [ 2 ] ; or from analytical models of IBRs [ 7 ] . In this paper, the analytical approach is employed for both GFLI and GFMI impedance modeling, using typical configurations reported in the literature. The accuracy of this approach has been validated against measurement-based identification in [ 13 , 15 ] . Analytical impedance models of both GFMIs and GFLIs have been extensively investigated [ 11 , 15 , 13 , 3 ] . Among the available techniques, the component-connection method is widely adopted due to its simplicity and ease of implementation [ 3 , 15 , 13 ] . In this method, linear subsystems, each described by a state-space model, are combined

Read the complete text