Title: Global-local temporal attention network for short-term solar irradiance prediction

Authors: Jian Wang; Huiyuan Liu; Zhichong Dou; Jieshan Shan; Zhanxi Zhang; Fu Shen; Hongchun Shu; Yiming Han; Zilong Cai; Kaizheng Wang; Meng Wang; Dongkai Zhang; Lei Kou; Huiyuan Nie

Addresses: Faculty of Electric Power Engineering, Kunming University of Science and Technology, Kunming, 650500, China ' Faculty of Electric Power Engineering, Kunming University of Science and Technology, Kunming, 650500, China ' Inspur Electronic Information Industry Co., Ltd., Jinan, 250101, China ' Faculty of Electric Power Engineering, Kunming University of Science and Technology, Kunming, 650500, China ' Southern Power Grid Electric Vehicle Service Co., Ltd., Shenzhen, 518000, China ' Faculty of Electric Power Engineering, Kunming University of Science and Technology, Kunming, 650500, China ' Faculty of Electric Power Engineering, Kunming University of Science and Technology, Kunming, 650500, China ' Faculty of Electric Power Engineering, Kunming University of Science and Technology, Kunming, 650500, China ' Faculty of Electric Power Engineering, Kunming University of Science and Technology, Kunming, 650500, China ' Faculty of Electric Power Engineering, Kunming University of Science and Technology, Kunming, 650500, China ' State Grid Zhejiang Electric Power Co., Ltd., Kaihua Power Supply Company, Quzhou, 330801, China ' College of Information Engineering, Henan University of Science and Technology, Luoyang, 471023, China ' State Key Laboratory of Physical Oceanography, Institute of Oceanographic Instrumentation, Qilu University of Technology (Shandong Academy of Sciences), Qingdao, 266061, China ' Yalong River Hdropower Development Company Ltd., Chengdu, 610051, China

Abstract: The intermittent and fluctuating nature of solar radiation poses significant challenges to power system stability, as it can lead to unpredictable fluctuations in energy generation, thereby complicating grid management. To address this, we propose a solar irradiance prediction method based on the global-local temporal attention network (GLTAN). The GLTAN comprises three key components: the global-local feature extraction (GLFE) module, the temporal attention mechanism (TAM), and the gated recurrent unit (GRU). The GLFE module uses a dual-layer transformer and temporal convolutional network (TCN) to extract both global and local features, capturing short-term and long-term trends. The TAM selectively highlights relevant information, while the GRU captures short-term dependencies. Experimental results for four different prediction time steps show that GLTAN outperforms the other models, with an average of 7.3% improvement in R2 over nine prediction steps.

Keywords: solar irradiance; short-term prediction; GLFE; global-local feature extraction; self-attention mechanism.

DOI: 10.1504/IJCSM.2025.151299

International Journal of Computing Science and Mathematics, 2025 Vol.22 No.4, pp.332 - 349

Published online: 22 Jan 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article