Banca de QUALIFICAÇÃO: ÁLISSON DE OLIVEIRA ALVES

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : ÁLISSON DE OLIVEIRA ALVES
DATE: 27/07/2026
TIME: 09:00
LOCAL: Videoconferência na plataforma Google Meet
TITLE:

Integration of remote sensing, deep learning, and GIS databases for prioritizing areas susceptible to non-technical losses in electric power distribution networks.

KEY WORDS:

Non-technical losses; Remote sensing; BDGD; Deep learning; Semantic segmentation; Distribution grids; Geoprocessing.

PAGES: 185
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUBÁREA: Sistemas Elétricos de Potência
SUMMARY:

This research investigates the integration of remote sensing, deep learning, and georeferenced electrical infrastructure data to support the identification of areas susceptible to non-technical losses in distribution grids. The problem is addressed as a territorial prioritization task, rather than as an automatic fraud classification problem. The proposed methodology combines operational data from the utility, records of confirmed irregularity reports, the Geographic Database of the Distribution Utility (BDGD), the Notable Points layer (PONNOT), and multi-resolution satellite imagery. The pipeline includes data acquisition and preprocessing, semantic segmentation of urban occupation, modeling of formal electrical coverage, generation of cadastral shadow areas, and structuring of georeferenced alerts to support field inspections.

 

The study area focuses on the state of Minas Gerais, with a detailed analysis of the municipality of Sabará and the area supplied by the SBAU substation, selected due to the convergence of electrical criticality, cadastral data availability, and confirmed irregularity reports. To detect urban occupation, high-resolution images associated with Open Buildings and Landsat 8 images associated with the MapBiomas urbanized-area class were used. Semantic segmentation architectures from the LinkNet and D-LinkNet families were evaluated to produce binary occupation masks. In parallel, formal electrical infrastructure was modeled from BDGD/PONNOT through asset filtering, generation of influence areas, morphological closing, and rasterization. The spatial difference between detected occupation and registered electrical coverage produced the cadastral shadow layer, interpreted as a territorial indicator of susceptibility, not as direct confirmation of irregularity.

 

The preliminary results demonstrate the technical feasibility of integrating satellite imagery, segmentation models, and georeferenced electrical infrastructure to reduce the operational search area. The approach enabled the comparison between urban expansion and grid evolution, the identification of regions with possible cadastral underrepresentation, and the organization of alerts associated with municipalities, feeders, and priority spatial patches. Comparative operational validation, including field accuracy rate, inspection cost, recovered energy, and financial return, remains a subsequent stage, dependent on a controlled field inspection campaign. The main contribution of this research lies in the construction of a reproducible geospatial pipeline capable of transforming dispersed evidence into more precise inspection targets in the context of the electric power distribution sector.

COMMITTEE MEMBERS:
Presidente - 1242315 - PABLO JAVIER ALSINA
Interno - 2885532 - IVANOVITCH MEDEIROS DANTAS DA SILVA
Externo à Instituição - ANDOUGLAS GONÇALVES DA SILVA JÚNIOR - IFRN
Externa à Instituição - MARIA DE FATIMA ALVES DE MATOS - ISI-ER
Externo à Instituição - RAUL BENITES PARADEDA - UERN