Detecting faults on power grids

This project aims to develop a framework for fault diagnosis in transmission lines through machine learning. A fault refers to a disturbance in the normal flow of electricity in the components of a power system, such as an increase in the current flow to one or more phases (Figure 1). In particular, this project focuses especially on two crucial tasks: fault classification and fault location

Figure 1 – Example of a real fault of type AG with inception around 0.08 seconds

The former task is responsible for determining which phases (A, B, C) are involved in the respective failure, with the action or not of the ground (G). The latter task is responsible for providing an approximate position where the failure occurred in the line. Both activities are critical aspects of transmission line maintenance and repair, as they enable power companies to quickly and accurately identify the type and the location of a fault to dispatch a repair staff to fix the issue. Figure 2 represents an overview of the proposed method.

Figure 2 – Operating environment of the proposed method for fault diagnosis in transmission lines.

The solution developed in this project uses oscillography data of voltage and current signals (e.g., Figure 1), extracting features from these waveforms to feed ensemble-based algorithms to perform the classification and location task. In particular, we developed a novel dynamic regressor selection scheme to perform the fault location task.

Current Results

We have already published some results in national and international conferences like KDMiLe (https://doi.org/10.5753/kdmile.2022.227805), SBBD (https://doi.org/10.5753/sbbd.2023.231718), and IECON (https://dx.doi.org/10.1109/IECON49645.2022.9968993), as well as a recent work in the international journal Neural Computing & Applications (https://doi.org/10.1007/s00521-023-09155-y). Moreover, all data created in our study and used to assess our framework are available to the scientific community: https://dx.doi.org/10.5281/zenodo.10275032

Main Technologies

deep learning, ensemble, dynamic regressor selection, denial constraints

Scroll to Top