A Review of Short-Term Photovoltaic Power Forecasting Driven by Big Data: From LSTM to Hybrid Intelligent Models

DOI: https://doi.org/10.62517/jbdc.202501435

Author(s)

Ruijia Huang

Affiliation(s)

Chongqing University of Posts and Telecommunications, International college, ChongQing, China

Abstract

The dual carbon goals are compelling power systems to accelerate the integration of high proportions of renewable energy. However, photovoltaic output fluctuates dramatically due to weather randomness, introducing significant uncertainty into dispatch, trading, and energy storage deployment. Over the past five years, deep learning models represented by LSTM have rapidly become the mainstream tool for short-term photovoltaic power forecasting (STPF) due to their superior ability to capture long-range dependencies in time series data. Concurrently, the big data environment has fostered a new generation of hybrid forecasting frameworks integrating "signal decomposition, feature extraction, and parameter optimization." This paper systematically reviews the latest global advancements in data infrastructure, model evolution, algorithm optimization, and engineering implementation. It distills a universal paradigm-“VMD/CNN/Attention-LSTM combined with Population-Based Intelligence"-while emphasizing that challenges such as the depth of multi-source data fusion, model interpretability, transferability across sites, and the implementation of online learning remain critical areas for future research.

Keywords

Photovoltaic Power Generation; Short-Term Power Forecasting; LSTM; Population-based Optimization Algorithms; Big Data

References

[1] Wang Biyuan, Deng Xingye. Research on LSTM-Based Big Data Analysis and Forecasting Methodology [J]. Power System Technology, 2021. [2] State Grid. New Energy Cloud White Paper [R]. 2023. [3] Sun Wenlei, et al. A Framework for Short-Term PV Power Generation Forecasting Based on Weather Clustering and Linear Regression [J]. Automation of Electric Power Systems, 2020, 44(8): 177-183. [4] Li Junhui, Wang Chengshan, Li Zhenkun, et al. Adaptive Prediction of Photovoltaic Power Scenarios Based on DTW-HDBSCAN Clustering [J]. Automation of Electric Power Systems, 2022, 46(15): 102-109. [5] Yang Ting, Zhao Qian, Zhang Pei. Ultra-Short-Term Photovoltaic Power Forecasting Integrating Satellite Cloud Images and ConvLSTM [J]. Journal of Tianjin University (Natural Science and Engineering Technology Edition), 2022, 55(7): 701-708. [6] Chen Qingming, et al. A Photovoltaic Power Forecasting Model Based on VMD-GWO-LSTM [J]. Acta Energiae Solaris Sinica, 2023, 44(2): 315-322. [7] Zhang Chenghui, Wang Rui, Liu Yang. Short-Term Photovoltaic Power Forecasting Based on Hierarchically Reconstructed VMD-LSTM [J]. Acta Energiae Solaris Sinica, 2022, 43(9): 308-315. [8] Li Zhenkun, Wang Chengshan, Li Junhui. VMD-LSTM Photovoltaic Power Forecasting Model Based on DDPG Adaptive Parameter Selection [J]. Transactions of the Chinese Society for Electrical Engineering, 2023, 43(18): 7021-7029. [9] Pi Linlin, Liguo Tian. Short-Term Photovoltaic Power Forecasting Method Based on KNN-IDBO-LSTM [J]. Transactions of the Chinese Society for Electrical Engineering, 2023, 43(5): 1658-1667. [10] Lv Lin, Liu Chang, Chen Minyou. Soft Correction Method for Photovoltaic Power Anomalies Based on LOF and Multi-Objective IDBO [J]. Power System Technology, 2023, 47(10): 3892-3900. [11] Wang Taihua, Zheng Wenshuang. Short-term photovoltaic power prediction based on STSV-CNN-BiLSTM [J]. Journal of Hunan University (Natural Science Edition), 2025,52(10):193-204. [12] Yang J, Xu Y, Xia Y. Spatio-temporal dual attention network for PV power forecasting using satellite cloud images[J]. IEEE Transactions on Sustainable Energy, 2023, 14(3): 1685-1695. [13] Lim S C, et al. Sensor-less PV power forecasting using CNN-LSTM with sky image classification[J]. Applied Energy, 2022, 308: 118374. [14] Pedersen J K, Nielsen H A, Madsen H. Swin-UNet cloud segmentation for high-latitude PV power prediction without on-site sensors[J]. Renewable Energy, 2024, 218: 119220. [15] Antonanzas J, et al. CEEMDAN-CNN-LSTM for hourly solar irradiation forecasting[J]. Solar Energy, 2021, 219: 595-605. [16] D’Amico A, Cuomo F, Fontanella A. Clear-sky residual learning for satellite-free rooftop PV forecasting[J]. Solar Energy, 2023, 251: 123-135. [17] Bin J, Xiaosong Z, Jiewen L, Yang Z.Blockchain-Enabled Federated Learning Data Protection Aggregation Scheme With Differential Privacy and Homomorphic Encryption in IIoT[J]. IEEE Transactions on Industrial Informatics 2021,PP(99):1-1 [18] Southern Regional Office of the Energy Administration. Implementation Rules for Grid Connection Operation and Ancillary Services of Photovoltaic Power Generation in the Southern Region[R]. February 2025. [19] Deep Learning: Online Learning and Adaptation [OL]. Alibaba Cloud Developer Community, August 2024. [20] Han Shuang, North China Electric Power University: Multi-Scenario Wind Farm Power Forecasting Technology-2025 Smart Power Plant Forum[OL]. WeChat Official Account (Energy Outlook), June 18, 2025. [21] Chen Y, Liu Z, Wang H. Cross-regional federated learning platform for renewable energy forecasting based on hierarchical PBFT+DPoS consensus[J]. Applied Energy, 2023, 335: 120720. [22] Li Z, Guo B, Huang X. Model pruning and Paillier homomorphic encryption for lightweight federated learning in provincial new energy forecasting[J]. IEEE Trans. Industrial Informatics, 2023, 19(8): 8422-8432. [23] Wu H, Xu J, Wang J. Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting [C]//Proc. of NeurIPS, 2021. [24] Bifet A, Gavaldà R. ADWIN drift detector for real-time photovoltaic forecasting[J]. Knowledge-Based Systems, 2022, 242. [25] Liu Yongqian, Han Shuang, Wang Chengshan. Application Research of Multi-Task Joint Learning in County-Level “Photovoltaic-Storage-Charging-Load” Scenarios [J]. Transactions of the Chinese Society for Electrical Engineering, 2025, 45(3): 789-797.