Herramienta computacional para el entrenamiento y validación de dispositivos electrónicos inteligentes como dispositivos de protección

In our project we identified the problems faced by electrical protection schemes in active distribution networks (ADNs) and microgrids (MGs), in which conventional protection systems are not a viable option anymore. As a consequence of this, new adaptive protection schemes have been employed. However, the implementation of some of these new protection schemes requires extensive data analysis and data preparation times in order to train the protections present in the system. For this reason, a computational tool was developed for the training and validation of intelligent electronic devices (IEDs) used as protective devices, which guarantee the correct functioning of the protective schemes used in ADNs and MGs. Using for this purpose, different techniques of Machine Learning (ML). In order to validate the correct functioning of the developed tool, we worked with the modified IEEE 34 Nodes system, to which different distributed generation (DG) sources were added in order to simulate the functioning of ADNs. In this way, the training of the IEDs was carried out with a database containing a total of 16 failure scenarios and 112 non-failure scenarios, presenting these scenarios with different changes in topology and generation, so as to obtain better models of ML from the techniques used. In total, 3 different types of ML techniques were analyzed, these being the techniques of Decision Trees (DT), Support Vector Machines (SVM) and Neural Networks (NN). The results obtained showed that the technique with the best results turned out to be the DT technique, with which precision and reliability values of 100% were obtained for all the IEDs under study. Because of the above, we believe that it was possible to meet the overall objective of the project.

• Autores: Benavides Bolaño, Lina María; Coba Jaramillo, Duiristt de Jesús
• Fecha de publicación: 2021
• Institución: Universidad del Norte
• Repositorio: Repositorio Uninorte
• Idioma: spa
• OAI Identifier: oai:manglar.uninorte.edu.co:10584/9567
• Acceso en línea: http://hdl.handle.net/10584/9567
• Palabra clave: Protección adaptativa; Microrred; Red de distribución activa; Técnicas de aprendizaje de máquina; Machine learning; Microgrid; Active distribution network; Adaptive protection
• License: Universidad del Norte