Integrating autoencoders to improve fault classification with PV system insertion
The extensive integration of distributed generation (DG) units leads to significant changes in the operation of power distribution systems (PDS). Integrating DG units contributes to meeting the growing energy demand, diversifying the energy matrix, and reducing power grid losses. In contrast, they c...
| Autores: | , , , , , |
|---|---|
| Tipo de recurso: | artículo |
| Estado: | Versión publicada |
| Fecha de publicación: | 2025 |
| País: | Brasil |
| Institución: | Universidade Estadual Paulista (UNESP) |
| Repositorio: | Repositório Institucional da UNESP |
| Idioma: | inglés |
| OAI Identifier: | oai:repositorio.unesp.br:11449/307384 |
| Acceso en línea: | http://dx.doi.org/10.1016/j.epsr.2025.111426 https://hdl.handle.net/11449/307384 |
| Access Level: | acceso abierto |
| Palabra clave: | Artificial neural networks Autoencoders Fault classification Photovoltaic systems Power distribution systems |
| Sumario: | The extensive integration of distributed generation (DG) units leads to significant changes in the operation of power distribution systems (PDS). Integrating DG units contributes to meeting the growing energy demand, diversifying the energy matrix, and reducing power grid losses. In contrast, they can affect conventional protection systems in power grids by altering current flow, which affects the characteristics, direction, and amplitude of short-circuit currents. Consequently, improper operation of protection equipment can cause false positives, negatively affecting the detection, classification, and reliability of the power grid. This study addresses fault classification in PDS, considering the extensive integration of DG units, specifically PV systems. PDS is evaluated at various levels of PV insertion using different fault scenarios modeled in the IEEE 34-bus test system. This includes five scenarios with variations in the PV system insertion. Autoencoders are applied during the pre-processing phase, while eleven different algorithms are used in the classification stage to identify fault types. They can improve the performance of the classification system by reducing the size of input signals and extracting the most relevant features. The results reveal that the K-nearest neighbor (KNN) and random forest (RF) algorithms demonstrate the best performance, maintaining a minimum accuracy of 95.42% in all scenarios. |
|---|