Work

Impact of bad and partial input data on neural network-based low voltage state estimation

Year: 2026

Type: article

Abstract: In recent years, it has been shown that distribution networks can be monitored using learning-based approaches. Their potential is evident when applied to low-voltage (LV) networks, where traditional monitoring solutions are often lacking. However, despite being data-driven, these methodologies still rely on a key input: the three-phase model of the LV distribution grid, which is often difficult to obtain due to scarce, low-quality or partial information. This work aims to bridge the gap between theoretical development and practical application in three main directions. First, it extends the framework to account for distributed PV generation, broadening its scope and applicability. Second, it enhances model performance by incorporating new types of measurements that are increasingly available at secondary substations. Finally, the focus of the paper is centered around practical implementation challenges related to bad or partially correct offline and online input data, such as the grid model and the input measurements. The method demonstrates robustness to various grid model inaccuracies, with only the phase allocation of loads and generators having a notable impact. Errors in line parameters, load profiles, PV modeling, and minor topological modifications have limited effects. Regarding grid evolution, model retraining is required only under significant changes such as a doubling of PV penetration or a 9% increase in total load. Trust in the model’s behavior is further reinforced by a Local Sensitivity Analysis, which reveals high sensitivity to voltage inputs and robustness to moderate perturbations in other features. The proposed methodology is implemented using real grid data from a local distribution system operator, confirming its effectiveness under realistic conditions.

Source: Sustainable Energy Grids and Networks

DOI: 10.1016/j.segan.2026.102327

Authors: Andrea Bragantini, Andreas Sumper, Yamen Alsyoufi, Samuel Galceran-Arellano

Institutions: Universitat Politècnica de Catalunya, Universidad de Oviedo

Language: English

Topic: Optimal Power Flow Distribution

Subfield: Electrical and Electronic Engineering

Field: Engineering

Domain: Physical Sciences

SDGs: Affordable and clean energy, Industry, innovation and infrastructure

Open Access status: hybrid

Estimated APC paid: $2,760