Selección del modelo óptimo de predicción de la relación de desempeño de una planta solar fotovoltaica. Un enfoque multicriterio basado en algoritmos de aprendizaje automático
Date
Authors
Subject
Clasificadores
comparaciones pareadas
criterios de decisión
métricas de evaluación
TOPSIS
Classifiers
pairwise comparisons
decision criteria
evaluation metrics
TOPSIS
comparaciones pareadas
criterios de decisión
métricas de evaluación
TOPSIS
Classifiers
pairwise comparisons
decision criteria
evaluation metrics
TOPSIS
Language:
Journal Title
Journal ISSN
Volume Title
Publisher
Instituto Tecnológico de Santo Domingo (INTEC)
La producción de energía eléctrica a partir de las plantas solares fotovoltaicas se ha intensificado en los últimos años con el fin de disminuir el uso de los combustibles fósiles. Sin embargo, este tipo de plantas no está exenta de sufrir pérdidas de energía, reduciendo en consecuencia su rendimiento. La Comisión Electrotécnica Internacional, a través de sus estándares, ha diseñado una serie de indicadores de desempeño clave para estas plantas, uno de los cuales es la relación de desempeño. El objetivo de esta investigación es presentar una metodología multicriterio para seleccionar el mejor modelo de clasificación para predecir la clase de la relación de desempeño de plantas solares fotovoltaicas. Se ilustra la metodología, utilizando los datos de una planta comercial ubicada en la zona central de Chile, considerando la técnica de análisis multicriterio TOPSIS, y los algoritmos de K vecinos más cercanos, máquinas de soporte vectorial, bosques aleatorios, y regresión logística, como alternativas del problema de decisión. Los criterios de decisión son las siguientes métricas: exactitud, precisión, f1-score, recall, y ROC-AUC. Como resultado se obtuvo que el mejor modelo correspondió al obtenido con regresión logística, con un puntaje del 100%, seguido del modelo de bosques aleatorios con 82,86%. Se recomienda incorporar nuevos modelos de clasificación a la metodología, y probarla con los datos de otra planta solar fotovoltaica.
The production of electrical energy from photovoltaic solar plants has intensified in recent years to reduce the use of fossil fuels. However, this type of plant is not exempt from suffering energy losses, consequently reducing its performance. The International Electrotechnical Commission, through its standards, has designed a series of key performance indicators for these plants, one of which is the Performance Ratio. The objective of this research is to present a multicriteria methodology to select the best classification model to predict the class of the Performance Ratio of photovoltaic solar plants. The methodology is illustrated, using data from a commercial plant located in the central area of Chile, considering the TOPSIS multicriteria analysis technique, and the K nearest neighbors, support vector machines, random forest, and logistic regression algorithms, as alternatives to the decision problem. The decision criteria are the metrics: accuracy, precision, f1-score, recall, and ROC-AUC. As a result, it was obtained that the best model corresponded to the one obtained with logistic regression, with a score of 100%, followed by the random forest model with 82.86%. It is recommended to incorporate new classification models to the methodology and test it with data from another plant.
The production of electrical energy from photovoltaic solar plants has intensified in recent years to reduce the use of fossil fuels. However, this type of plant is not exempt from suffering energy losses, consequently reducing its performance. The International Electrotechnical Commission, through its standards, has designed a series of key performance indicators for these plants, one of which is the Performance Ratio. The objective of this research is to present a multicriteria methodology to select the best classification model to predict the class of the Performance Ratio of photovoltaic solar plants. The methodology is illustrated, using data from a commercial plant located in the central area of Chile, considering the TOPSIS multicriteria analysis technique, and the K nearest neighbors, support vector machines, random forest, and logistic regression algorithms, as alternatives to the decision problem. The decision criteria are the metrics: accuracy, precision, f1-score, recall, and ROC-AUC. As a result, it was obtained that the best model corresponded to the one obtained with logistic regression, with a score of 100%, followed by the random forest model with 82.86%. It is recommended to incorporate new classification models to the methodology and test it with data from another plant.
Description
Type
info:eu-repo/semantics/article
info:eu-repo/semantics/publishedVersion
Art.
info:eu-repo/semantics/publishedVersion
Art.
Source
Science, Engineering and Applications; Vol. 6 No. 2 (2023): Science, Engineering and Applications; 7-29
Ciencia, Ingenierías y Aplicaciones; Vol. 6 Núm. 2 (2023): Ciencia, Ingenierías y Aplicaciones; 7-29
2636-2171
2636-218X
10.22206/cyap.2023.v6i2
Ciencia, Ingenierías y Aplicaciones; Vol. 6 Núm. 2 (2023): Ciencia, Ingenierías y Aplicaciones; 7-29
2636-2171
2636-218X
10.22206/cyap.2023.v6i2