Advanced clustering with high resolution in photovoltaic systems and its impact on prediction models

Fabián K.M.F.; Salazar D.G.P.; Sanchez Sutil F.J.; de la Casa Hernandez J.; Ortega A.C. · International Conference on Electrical Engineering, Computing Science and Automatic Control, CCE (2025) · DOI: 10.1109/cce67728.2025.11271920

Resumen

This study proposes a hybrid framework that integrates advanced clustering techniques with deep learning models, designed to exploit very high temporal resolution data in order to increase the accuracy of short- and medium-term forecasts. The main objective is to demonstrate that the incorporation of clustering processes makes it possible to identify daily atmospheric variability, seasonality or other dynamic transitions, which in turn allows the training of more robust specialized models. A database with 10-second resolution was used under varied weather conditions. The results show that the application of clustering optimizes data segmentation, facilitating more effective learning, improving the predictive capacity of the algorithms used. In terms of performance, it was observed that the clustered models outperformed the reference models (without clustering), reaching improvements in R 2 of up to 0.005 in short-term forecasts and an average of 0.031 for the medium term. Likewise, relative improvements were recorded in prediction errors: the MAE decreased on average by 18.79% for the short term and 42.97% for the medium term, while the RMSE showed average reductions of 8.37% and 41.88%, respectively, as well as a reduction in processing time of an average of 6.47% for shortterm predictions (robust method) and an average of 83% for medium-term forecasts (less complex method). These findings demonstrate that high granularity and clustering bring considerable practical value in critical applications such as: (i) power availability estimation by time slots, (ii) intelligent programming of battery storage systems, and (iii) energy planning in PV systems.

Instituciones: Universidad Nacional de San Agustín de Arequipa