volume | PIER Journals

Abstract

Photovoltaic (PV) power sequences are highly susceptible to high-frequency stochastic noise under complex micro-meteorological conditions. Furthermore, existing forecasting models struggle with isolated multi-scale physical features and insufficient nonlinear mapping capabilities. To address these limitations, this paper proposes an improved TimeMixer-based PV power forecasting method. First, the macroscopic trend and microscopic seasonal components are extracted via a past-decomposable-mixing architecture. Second, an adaptive gated feature fusion mechanism is introduced as a physically motivated feature-level filter to attenuate high-frequency noise channels through dynamic attention masks, effectively blocking the cross-scale propagation of invalid meteorological interference. Finally, a cross-scale joint nonlinear network is constructed to capture nonlinear interactions among multi-band components through state matrix aggregation and activation operators. Case studies utilizing operational data from a 50 MW PV power plant, in Xinjiang, China, demonstrate that the proposed architecture effectively overcomes smoothing degradation and phase lag under complex scenarios, such as abrupt cloud cover. Compared with the original baseline, the proposed method reduces the forecasting mean squared error by 10.80%, significantly enhancing both global fitting accuracy and dynamic extreme-value tracking capability.

1. Kang, C. and L. Yao, "Key scientific issues and theoretical research framework for power systems with high proportion of renewable energy," Automation of Electric Power Systems, Vol. 41, No. 9, 1-11, 2017.
doi:10.7500/AEPS20170120004 Google Scholar

2. Antonanzas, J., N. Osorio, R. Escobar, R. Urraca, F. J. Martinez-de-Pison, and F. Antonanzas-Torres, "Review of photovoltaic power forecasting," Solar Energy, Vol. 136, 78-111, 2016.
doi:10.1016/j.solener.2016.06.069 Google Scholar

3. Zhu, Yongqiang and Jun Tian, "Application of least square support vector machine in photovoltaic power forecasting," Power System Technology, Vol. 35, No. 7, 54-59, 2011. Google Scholar

4. Polo, Alessandro, "A two-step learning-by-examples method for photovoltaic power forecasting," Progress In Electromagnetics Research C, Vol. 125, 35-49, 2022.
doi:10.2528/pierc22061003 Google Scholar

5. Zhou, Hangxia, Yujin Zhang, Lingfan Yang, Qian Liu, Ke Yan, and Yang Du, "Short-term photovoltaic power forecasting based on long short term memory neural network and attention mechanism," IEEE Access, Vol. 7, 78063-78074, 2019.
doi:10.1109/access.2019.2923006 Google Scholar

6. Zhou, Haoyi, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wancai Zhang, "Informer: Beyond efficient transformer for long sequence time-series forecasting," Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35, No. 12, 11106-11115, 2021.

7. Wu, Haixu, Jiehui Xu, Jianmin Wang, and Mingsheng Long, "Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting," Advances in Neural Information Processing Systems, Vol. 34, 22419-22430, 2021. Google Scholar

8. Cai, Ziwen, Zhukui Tan, Yun Zhao, Yuelang Zhang, Xipeng Liu, and Houyi Zhang, "Multiformer-TSA-based photovoltaic power forecasting method," Journal of Electric Power Science and Technology, Vol. 41, No. 1, 130-139, 2026.
doi:10.19781/j.issn.1673-9140.2026.01.013 Google Scholar

9. Bak, Seongho, Sowon Choi, Donguk Yang, Doyoon Kim, Heeseon Rho, and Kyoobin Lee, "Transfer learning for photovoltaic power forecasting across regions using large-scale datasets," IEEE Access, Vol. 13, 136175-136190, 2025.
doi:10.1109/access.2025.3591040 Google Scholar

10. Li, Z., Y. Wang, R. Zhang, et al. "A multi-timescale photovoltaic power prediction method based on SE-CNN-BiLSTM and improved Transformer," Zhejiang Electric Power, Vol. 45, No. 3, 120-130, 2026.
doi:10.19585/j.zjdl.202603011 Google Scholar

11. Lin, Huapeng, Liyuan Gao, Mingtao Cui, Hengchao Liu, Chunyang Li, and Miao Yu, "Short-term distributed photovoltaic power prediction based on temporal self-attention mechanism and advanced signal decomposition techniques with feature fusion," Energy, Vol. 315, 134395, 2025.
doi:10.1016/j.energy.2025.134395 Google Scholar

12. Wang, Shiyu, Haixu Wu, Xiaoming Shi, Tengge Hu, Huakun Luo, Lintao Ma, James Y. Zhang, and Jun Zhou, "Timemixer: Decomposable multiscale mixing for time series forecasting," arXiv preprint arXiv:2405.14616, 2024.
doi:10.48550/arXiv.2405.14616 Google Scholar

13. Li, Z., Xinghua Wang, Chenyang Fu, et al. "Planning-state PV-load joint scenario generation considering decomposable multi-scale temporal feature fusion," Power System Protection and Control, Vol. 53, No. 12, 152-164, 2025.
doi:10.19783/j.cnki.pspc.241042 Google Scholar

14. Qiu, Zhibo, Changyou Fu, Beijing Liang, and Chao Cao, "Power load forecasting based on the improved TimeMixer," International Conference on Image, Signal Processing, and Pattern Recognition (ISPP 2025), Vol. 13664, 1094-1100, Nanjing, China, 2025.
doi:10.1117/12.3070579

15. Gao, Shuang, Chenhao Li, Zeyu Li, et al. "Fault detection of offshore floating photovoltaic power system considering wave motion and time-sequence characteristics," Transactions of China Electrotechnical Society, Vol. 41, No. 3, 999-1011, 2026.
doi:10.19595/j.cnki.1000-6753.tces.250113 Google Scholar

16. Piantadosi, Gabriele, Sofia Dutto, Antonio Galli, Saverio De Vito, Carlo Sansone, and Girolamo Di Francia, "Photovoltaic power forecasting: A Transformer based framework," Energy and AI, Vol. 18, 100444, 2024.
doi:10.1016/j.egyai.2024.100444 Google Scholar

17. Huang, Xiaohong, Xiuzhen Ding, Yating Han, Qi Sima, Xiaokang Li, and Yukun Bao, "Day-ahead photovoltaic power forecasting based on SN-transformer-bimixer," Energies, Vol. 18, No. 16, 4406, 2025.
doi:10.3390/en18164406 Google Scholar

Dr. Chao Wang

Dr. Xinyuan Xie

Dr. Fengsheng Chen

Dr. Pengyi Fan

Dr. Zhengning Pan

Dr. Tao Yu

Dr. Zhongan Yu