[1]WANG Dewen,MIAO Qingjian,LI Chenghao,et al.GPT4PF: fine-tuning pre-trained LLM for photovoltaic power prediction[J].CAAI Transactions on Intelligent Systems,2026,21(4):864-875.[doi:10.11992/tis.202507023]
Copy

GPT4PF: fine-tuning pre-trained LLM for photovoltaic power prediction

CAAI Transactions on Intelligent Systems[ISSN 1673-4785/CN 23-1538/TP] Volume: 21 Number of periods: 2026 4 Page number: 864-875 Column: 学术论文—机器学习 Public date: 2026-07-05
Title:
GPT4PF: fine-tuning pre-trained LLM for photovoltaic power prediction
Author(s):
WANG Dewen1; 2;  MIAO Qingjian1;  LI Chenghao1;  SUN Yuan1;  ZHAO Wenqing1; 3
1. Department of Computer, North China Electric Power University, Baoding 071003, China;
2. Hebei Key Laboratory of Knowledge Computing for Energy & Power, Baoding 071003, China;
3. Engineering Research Center of Intelligent Computing for Complex Energy Systems Ministry of Education, Baoding 071003, China
Keywords:
pre-trained large language model; photovoltaic forecast; large language model; time encoding; multi-stage fine-tuning; time-series segmentation; input encoding; linear probing
CLC:
TP183; TM615
DOI:
10.11992/tis.202507023
Abstract:
Accurate photovoltaic power forecasting is crucial for ensuring the security of power systems and improving the efficiency of solar energy utilization.This study proposed a photovoltaic power forecasting model named GPT4PF, which was developed by fine-tuning a pre-trained large language model (LLM). A highly adaptive and efficient three-stage fine-tuning strategy was proposed, which enhanced forecasting accuracy while limiting the number of training parameters to merely 0.49% of the total model parameters. An input encoding layer embedded with temporal information was constructed to strengthen the LLM’s capability of modeling temporal features. Ablation experiments showed that the mean absolute error (MAE) was reduced by approximately 1%. In addition, the issue that a single time point carries limited information and struggles to reflect the overall characteristics and patterns of photovoltaic data was addressed through time-series segmentation technology. Experimental results showed that compared with DLiner, PatchTST, and LLM4TS, the proposed model achieved a reduction in MAE by 18.9%, 4.1%, and 5.4% respectively.
References:
[1] 国家能源局. 国家能源局发布2024年全国电力工业统计数据[EB/OL]. (2025-01-21)[2025-07-08]. http://www.nea.gov.cn/20250121/097bfd7c1cd3498897639857d86d5dac/c.html. National Energy Administration. National energy administration releases 2024 national statistical data on the electric power industry[EB/OL]. (2025-01-21)[2025-07-08]. http://www.nea.gov.cn/20250121/097bfd7c1cd3498897639857d86d5dac/c.html.
[2] MURATA A, OHTAKE H, OOZEKI T. Modeling of uncertainty of solar irradiance forecasts on numerical weather predictions with the estimation of multiple confidence intervals[J]. Renewable energy, 2018, 117: 193-201
[3] HIRATA Y, AIHARA K. Improving time series prediction of solar irradiance after sunrise: comparison among three methods for time series prediction[J]. Solar energy, 2017, 149: 294-301
[4] AGOUA XG, GIRARD R, KARINIOTAKIS G. Short-term spatio-temporal forecasting of photovoltaic power production[J]. IEEE transactions sustainable energy, 2017, 9(2): 538-546
[5] 李承皓, 杨永标, 宋嘉启, 等. 基于IMVMD和BiLSTM-SARIMA组合模型的台区光伏短期发电功率预测[J]. 太阳能学报, 2025, 46(2): 433-440 LI Chenghao, YANG Yongbiao, SONG Jiaqi, et al. Short term photovoltaic power generation prediction based on imvmd and BiLSTM-SARIMA combination model in station area[J]. Acta energiae solaris sinica, 2025, 46(2): 433-440
[6] 朱琼锋, 李家腾, 乔骥, 等. 人工智能技术在新能源功率预测的应用及展望[J]. 中国电机工程学报, 2023, 43(8): 3027-3047 ZHU Qiongfeng, LI Jiateng, QIAO Ji, et al. Application and prospect of artificial intelligence technology in renewable energy forecasting[J]. Proceedings of the CSEE, 2023, 43(8): 3027-3047
[7] 张赟宁, 魏广军. 考虑特征选择的短期光伏功率组合预测模型[J]. 电力系统及其自动化学报, 2024, 36(8): 122-132 ZHANG Yunning, WEI Guangjun. Combined prediction model for short-term photovoltaic power considering feature selection[J]. Proceedings of the CSU-EPSA, 2024, 36(8): 122-132
[8] 马世龙, 乌尼日其其格, 李小平. 大数据与深度学习综述[J]. 智能系统学报, 2016, 11(6): 728-742 MA Shilong, WUNIRI Qiqige, LI Xiaoping. Deep learning with big data: state of the art and development[J]. CAAI transactions on intelligent systems, 2016, 11(6): 728-742
[9] LI Gangqiang, WANG Huaizhi, ZHANG Shengli, et al. Recurrent neural networks based photovoltaic power forecasting approach[J]. Energies, 2019, 12(13): 2538
[10] 姚蕊, 刘小芳. 线性分解和周期增强Informer的太阳辐射短临预报研究[J]. 太阳能学报, 2025, 46(2): 505-510 YAO Rui, LIU Xiaofang. Linera decomposition and periodic enhancement in short-term solar irradiance forecasting with informer[J]. Acta energiae solaris sinica, 2025, 46(2): 505-510
[11] 陈嘉俊, 刘波, 林伟伟, 等. 基于Transformer的时间序列预测方法综述[J]. 计算机科学, 2025, 52(6): 96-105 CHEN Jiajun, LIU Bo, LIN Weiwei, et al. Survey of Transformer-based time series forecasting methods[J]. Computer science, 2025, 52(6): 96-105
[12] ZHOU Tian, NIU Peisong, WANG Xue, et al. One fits all: Power general timeseries analysis by pretrained lm[J]. Advances in neural information processing systems, 2023, 36: 43322-43355
[13] SUN Chenxi, LI Hongyan, LI Yaliang, et al. TEST: text prototype aligned embedding to activate LLM’s ability for time series [C]//The Twelfth International Conference on Learning Representations. Vienna: ICLR, 2024.
[14] SHI Xiaoming, WANG Shiyu, NIE Yuqi, et al. Time-MoE: billion-scale time series foundation models with mixture of experts[C]//The Thirteenth International Conference on Learning Representations. Singapore: ICLR, 2025.
[15] CHAI Minkang, XIA Fei, HAO Shuotao, et al. PV power prediction based on LSTM with adaptive hyperparameter adjustment[J]. IEEE Access, 2019, 7: 115473
[16] KUMARI P, TOSHNIWAL D. Long short term memory-convolutional neural network based deep hybrid approach for solar irradiance forecastin-g[J]. Appl energy 2021: 295.
[17] 王德文, 焦天媛. 基于二次分解的不同太阳辐射下光伏功率预测[J]. 太阳能学报, 2024, 45(9): 360-368 WANG Dewen, JIAO Tianyuan. Prediction of photovoltaic power under different solar radiation based on secondary decomposition[J]. Acta energiae solaris sinica, 2024, 45(9): 360-368
[18] 范航舟, 梅红岩, 赵勤, 等. 融合双注意力机制的GNN多维时间序列预测[J]. 智能系统学报, 2024, 19(5): 1277-1286 FAN Hangzhou, MEI Hongyan, ZHAO Qin, et al. Multivariate time series forecasting with a graph neural network and dual attention mechanism[J]. CAAI transactions on intelligent systems, 2024, 19(5): 1277-1286
[19] ZHOU Haoyi, ZHANG Shanghang, PENG Jieqi, et al. Informer: beyond efficient transformer for long sequence time-series forecasting[C]//The Thirty-Fifth AAAI Conference on Artificial Intelligence. Online: AAAI, 2021: 11106.
[20] WU Haixu, XU Jiehui, WANG Jianmin, et al. Autoformer: decomposition transformers with auto-correlation for long-term series forecasting[J]. Advances in neural information processing systems, 2021, 34: 22419
[21] NIE Yuqi, NGUYEN Nam H, SINTHONG Phanwadee, et al. A time series is worth 64 words: long-term forecasting with transformers[EB/OL]. (2022-11-27)[2025-07-21]. http://arxiv.org/abs/2211.14730.
[22] XUE Hao, SALIM Flora D. Promptcast: a new prompt-based learning paradigm for time series forecasting[J]. IEEE transactions on knowledge and data engineering, 2023, 36(11): 6851-6864
[23] JIN Ming, WANG Shiyu, MA Lintao, et al. Time-LLM: time series forecasting by reprogramming large language models [C]//The Twelfth International Conference on Learning Representations. Vienna: ICLR, 2024.
[24] GARZA A, CHALLU C, MERGENTHALER C M. TimeGPT-1[EB/OL]. (2023-10-05)[2025-07-21]. https://arxiv.org/abs/2310.03589.
[25] RASUL K, ASHOK A, WILLIAMS A R, et al. Lag-Llama: towards foundation models for time series forecasting[EB/OL]. (2023-10-12)[2025-07-21]. https://arxiv.org/abs/2310.08278.
[26] CHANG Ching, PENG W C, CHEN T F. LLM4TS: two-stage fine-tuning for time-series forecasting with pre-trained LLMs[EB/OL]. (2023-08-16)[2025-07-21]. https://arxiv.org/abs/2308.08469v1.
[27] LAI Zefeng, WU Tangjie, FEI Xihong, et al. BERT4ST: fine-tuning pre-trained large language model for wind power forecasting[J]. Energy conversion and management, 2024, 307: 118331
[28] WU Tangjie, LING Qiang. STELLM: spatio-temporal enhanced pre-trained large language model for wind speed forecasting[J]. Applied energy, 2024, 375: 124034
[29] XIE Zhe, LI Zeyan, HE Xiao, et al. ChatTS: aligning time series with LLMs via synthetic data for enhanced understanding and reasonin[EB/OL]. (2024-12-04)[2025-07-21]. https://arxiv.org/abs/2412.03104.
[30] KIM T, KIM J, TAE Y, et al. Reversible instance normalization for accurate time-series forecasting against distribution shift[C]//The Ninth International Conference on Learning Representations. Online: ICLR, 2021.
[31] TOMIHARI A, SATO I. Understanding linear probing then fine-tuning language models from ntk perspective[EB/OL]. (2024-05-27)[2025-07-21]. https://arxiv.org/abs/2405.16747v2.
Similar References:

Memo

-

Last Update: 1900-01-01

Copyright © CAAI Transactions on Intelligent Systems