Forecast of operational data in electric energy plants using adaptive algorithm

Traditional time series methods offer models whose parameters remain constant over time. However, industrial supply and demand processes require timely decisions based on a dynamic reality. A change in configuration, turning off, or on a production line or process, modifies the problem and the varia...

Descripción completa

Detalles Bibliográficos
Autores: Viloria, Amelec, García Guiliany, Jesús Enrique, Hernandez-P, Hugo, CABAS VASQUEZ, LUIS CARLOS, Pineda, Omar
Tipo de recurso: artículo
Estado:Versión aceptada para publicación
Fecha de publicación:2020
País:Colombia
Institución:Corporación Universidad de la Costa
Repositorio:Repositorio REDICUC
Idioma:inglés
OAI Identifier:oai:repositorio.cuc.edu.co:11323/7726
Acceso en línea:https://hdl.handle.net/11323/7726
https://doi.org/10.1007/978-981-15-3125-5_48
https://repositorio.cuc.edu.co/
Access Level:acceso abierto
Palabra clave:Time series models
Estimation
Forecasts
Data analysis
Data mining
Statistical learning
Decision trees
Descripción
Sumario:Traditional time series methods offer models whose parameters remain constant over time. However, industrial supply and demand processes require timely decisions based on a dynamic reality. A change in configuration, turning off, or on a production line or process, modifies the problem and the variables to be predicted. Decision support systems must dynamically adapt in order to respond quickly and appropriately to operations and their processes. This methodology is based on obtaining, for each period, the model that best fits the data, evaluating many alternatives and using statistical learning techniques. In this way, the model will adapt to the data in practice and make decisions based on experience. With three months of testing for the estimation of variables associated with supply and demand processes, predictions that differ less than 8 hundredths (less than 0.08) or 0.1% of the measured value were obtained. This indicates that data science and statistical learning represent an important area of research for variable prediction and process optimization.