SolarNexus: A deep learning framework for adaptive photovoltaic power generation forecasting and scalable management

Hyunsik Min and Byeongjoon Noh

Applied Energy, 2025, vol. 391, issue C, No S0306261925005781

Abstract: Photovoltaic (PV) power forecasting plays a crucial role in managing renewable energy resources. However, traditional forecasting models often encounter difficulties in adapting to dynamic environmental conditions and scaling across diverse regions. In response to these challenges, we propose a deep learning framework that integrates a temporal convolutional network (TCN), multi-head attention (MHA), online learning, and transfer learning. To validate our approach, we utilized data from nine solar power plants in South Korea. The dataset, obtained from the Korea Open Data Portal and the Korea Meteorological Administration, encompasses hourly data on photovoltaic generation and meteorological parameters from January 1, 2017, to December 31, 2019, with two years for training and one for testing. We compared our TCN-MHA online learning model against specialized models such as ACL-RGRU and CNN-DeepESN under identical conditions, and we conducted an ablation study to assess the contribution of each component. Furthermore, we investigated cross-regional forecasting by implementing transfer learning using Sejong and Jeju as source regions. The experimental results demonstrate that our framework attained an average normalized root mean square error (NRMSE) of approximately 17.19 and a normalized mean absolute error (NMAE) of about 12.64, signifying about a 60 % error reduction compared to existing models. Notably, transfer learning reduced training time from approximately 265.61 to about 38.90 seconds (about 85 % reduction), graphic processing unit (GPU) utilization from about 75.18 % to about 16.14 % (about 78 % reduction), and power consumption from approximately 4,464.33 kW to about 35.01 kW (over 99 % reduction), all while maintaining high forecasting accuracy.

Keywords: Photovoltaic power forecasting; Model adaptability and scalability; Temporal-convolutional network; Multi-head attention; Online learning; Transfer learning (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0306261925005781
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:appene:v:391:y:2025:i:c:s0306261925005781

Ordering information: This journal article can be ordered from
http://www.elsevier.com/wps/find/journaldescription.cws_home/405891/bibliographic
http://www.elsevier. ... 405891/bibliographic

DOI: 10.1016/j.apenergy.2025.125848

Access Statistics for this article

Applied Energy is currently edited by J. Yan

More articles in Applied Energy from Elsevier
Bibliographic data for series maintained by Catherine Liu ().