Robust time series analysis for forecasting photovoltaic energy yield

¹ South-West University “Neofit Rilski”, Department of Communication and Computer Engineering, Faculty of Engineering, 2700 Blagoevgrad, Bulgaria
² Bulgarian Academy of Sciences, Institute of Mathematics and Informatics, Department of Information Modeling, 1113 Sofia, Bulgaria
³ University of Ruse “Angel Kanchev”, Faculty of Natural Sciences and Education, Department of Applied Mathematics and Statistics, 7004 Ruse, Bulgaria
⁴ Bulgarian Academy of Sciences, Institute of Information and Communication Technologies, Department of Parallel Algorithms and Machine Learning with Neurotechnology Laboratory, 1113 Sofia, Bulgaria

^* Corresponding authors: sapundzhi@swu.bg, sggeorgiev@uni-ruse.bg

Abstract

This study introduces an approach to forecasting the power output of a photovoltaic (PV) system by employing an ARIMA-based algorithm. Two distinct ARIMA models were designed – one generated via SPSS and one selected by the researchers. Their effectiveness is gauged using various goodness-of-fit metrics, which provide a detailed evaluation of each model’s precision. In addition, the autocorrelation (ACF) and partial autocorrelation (PACF) functions of the residuals are analysed to confirm the models’ soundness, while confidence intervals for these residuals are calculated to further substantiate their validity. The analysis proceeds with the generation of monthly predictions for the dataset, complete with their own confidence bounds, thereby showcasing the forecasting strength of the models. The findings underscore the utility of ARIMA techniques in projecting PV energy yields, delivering critical insights that can be leveraged to enhance system performance and strategic planning. Overall, this work aims to contribute to renewable energy forecasting by demonstrating that ARIMA models are a viable tool for predicting the monthly operational outcomes of photovoltaic systems.