Options
SARIMA models for power evolution in photovoltaic systems
Journal
Salud, Ciencia y Tecnología
ISSN
27969711
Date Issued
2025-08-01
Author(s)
Juan Espinoza
Christian Reyes
Diana Campaña
Elsa Basantes
Sandra Chasiluisa
Abstract
Introduction.- The increasing use of renewable energy in power generation systems has highlighted the need for efficient schemes to predict model parameters. In particular, photovoltaic systems require accurate tools to model and forecast solar energy generation behavior.
Objective.-To formulate SARIMA models with high accuracy in fitting, explanation, and prediction of energy yields in solar photovoltaic systems, specifically focused on the plant located at Plaza del Duque de Béjar, Spain. Method.- A fitting strategy based on genetic algorithms was adopted to accelerate the estimation of the SARIMA model using hourly solar photovoltaic generation data.
The auto.arima package in RStudio was employed as a methodological tool, enabling automatic selection and optimization of the best model parameters. Results.- The selected model was SARIMA (5,0,0)(2,1,0)242424, characterized by a stationary stochastic process with a clear seasonal component. The model showed remarkable estimation accuracy, with low standard errors in the autoregressive coefficients. Additionally, the model residuals were well-adjusted, displaying independence and absence of serial autocorrelation.
Conclusions.- The proposed model demonstrated excellent predictive performance, supported by training error metrics (ME (Mean Error)= -1.344268 and MASE (Mean Absolute Scaled Error)= 0.7048786).
Its sound mathematical structure and strong fit make it a reliable tool for forecasting photovoltaic solar energy in systems with similar characteristics.
Objective.-To formulate SARIMA models with high accuracy in fitting, explanation, and prediction of energy yields in solar photovoltaic systems, specifically focused on the plant located at Plaza del Duque de Béjar, Spain. Method.- A fitting strategy based on genetic algorithms was adopted to accelerate the estimation of the SARIMA model using hourly solar photovoltaic generation data.
The auto.arima package in RStudio was employed as a methodological tool, enabling automatic selection and optimization of the best model parameters. Results.- The selected model was SARIMA (5,0,0)(2,1,0)242424, characterized by a stationary stochastic process with a clear seasonal component. The model showed remarkable estimation accuracy, with low standard errors in the autoregressive coefficients. Additionally, the model residuals were well-adjusted, displaying independence and absence of serial autocorrelation.
Conclusions.- The proposed model demonstrated excellent predictive performance, supported by training error metrics (ME (Mean Error)= -1.344268 and MASE (Mean Absolute Scaled Error)= 0.7048786).
Its sound mathematical structure and strong fit make it a reliable tool for forecasting photovoltaic solar energy in systems with similar characteristics.