Autoregressive Modeling of Time Series in Renewable Energy Systems
Year: 2024
Type: conference-paper
Abstract: Autoregressive (AR) models have been increasingly used to forecast solar irradiance and estimate state-of-charge (SOC) in energy storage systems (ESSs). This work aims to apply AR models to real SOC and solar irradiance measurements. We determine the AR model coefficients using the modified covariance method while we estimate the model order using three standard metrics: final prediction error (FPE), Akaike information criterion (AIC), and criterion autoregressive transfer function (CAT). Additionally, we also evaluate the model order that minimizes the spectral Kullback-Leibler divergence between the data periodogram and the AR power spectral density estimates corresponding to the three metrics. We assess the model accuracy using root mean squared error (RMSE) and demonstrate that the lowest RMSE is generally achieved with higher model orders. The comparative analysis of the data collected under different weather conditions reveals that partly cloudy conditions are the most challenging.
Authors: Kaan Icer, Showrov Rahman, Kaushallya Adhikari
Institution: University of Rhode Island
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Office of Naval Research
OpenAlex ID: W4404564725