在智慧電網與再生能源併網需求日益提升的背景下，太陽能光伏發電憑藉其成本低廉與維護簡便等優勢，已成為分散式發電系統的重要一環。然而，其受天氣變化影響劇烈，易造成電力供應波動，對電網調度與備載容量規劃構成挑戰。為此，建立高效且精準的太陽能發電預測模型，成為實現穩定供電與能源優化配置的關鍵技術。因此如何提升太陽能發電量預測的精準度，為本文研究之方向。
為克服傳統預測方法面臨的數據不足與氣象複雜性挑戰，本研究提出一套結合資料生成、天氣分類、混合模型設計與自動化超參數調整之深度學習預測架構。首先，採用TimeGAN與LSTM生成模型進行資料增強，並使用K-means演算法將氣象資料分為三種天氣類型以擴充特徵集。接著，將三種資料分別輸入六種深度學習模型，並使用TPE貝氏優化法自動調整超參數，確保模型效能達最佳化。
實驗結果顯示，透過生成式的模型進行資料增強能有效提升預測準確性。同時，TPE貝氏優化過程也展現穩定的收斂趨勢，表明TPE能有效找到最佳超參數組合。整體而言，混合模型皆優於單一模型，其中本研究所提出之CNN-BiGRU-MHSA模型在各評價指標上均表現最佳，驗證本研究架構在短期太陽能發電預測上的可行性與優勢。

With the growing demand for smart grids and the integration of renewable energy sources, solar photovoltaic (PV) power generation has become a crucial component of distributed energy systems due to its low cost and minimal maintenance requirements. However, its output is highly susceptible to weather variations, often resulting in power supply fluctuations and creating challenges for grid dispatching and reserve capacity planning. Consequently, developing an efficient and accurate solar power forecasting model has become a key technology for achieving stable power supply and optimized energy allocation. Enhancing the accuracy of solar power prediction is therefore the central focus of this study.
To address the limitations of traditional forecasting methods caused by insufficient data and the complexity of meteorological conditions, this study proposes a deep learning-based forecasting framework that integrates data generation, weather classification, hybrid model design, and automated hyperparameter tuning. Specifically, TimeGAN and LSTM models are employed to augment the training data, while the K-means algorithm is used to classify weather data into three categories to enrich the feature set. These datasets are then input into six different deep learning models, and hyperparameters are optimized using the Tree-structured Parzen Estimator (TPE) Bayesian optimization method to ensure optimal model performance.
Experimental results demonstrate that data augmentation using generative models significantly improves prediction accuracy. Furthermore, the TPE optimization process shows a stable convergence trend, confirming its effectiveness in identifying optimal hyperparameter combinations. Overall, hybrid models outperform single-architecture models, with the proposed CNN-BiGRU-MHSA model achieving the best performance across all evaluation metrics, validating the effectiveness and practicality of the proposed framework for short-term solar power forecasting.

[1] International Energy Agency. (2022). World Energy Outlook 2022.
[2] United Nations Environment Programme. (2021). Emissions Gap Report 2021.
[3] J. Antonanzas, N. Osorio, R. Escobar, R. Urraca, F. J. Martinez-de-Pison, and F. Antonanzas-Torres, “Review of photovoltaic power forecasting,” Solar Energy, vol. 136, pp. 78-111, Oct. 2016.
[4] J. Gaboitaolelwe, A. M. Zungeru, A. Yahya, C. K. Lebekwe, D. N. Vinod, and A. O. Salau, “Machine Learning Based Solar Photovoltaic Power Forecasting: A Review and Comparison,” IEEE Access, vol. 11, pp. 40820-40845, April. 2023.
[5] G. Graditi, S. Ferlito, and G. Adinolfi, “Comparison of Photovoltaic plant power production prediction methods using a large measured dataset,” Renewable Energy, vol. 90, pp. 513-519, May. 2016.
[6] T. DeVries, and G. W. Taylor, “Dataset Augmentation in Feature Space,” arXiv preprint arXiv: 1702.05538, Feb. 2017.
[7] Y. Gao, C. A. Ellis, V. D. Calhoun, and R. L. Miller, “Improving age prediction: Utilizing LSTM-based dynamic forecasting for data augmentation in multivariate time series analysis,” arXiv preprint arXiv: 2312.08383, Dec. 2023.
[8] Q. Li, X. Zhang, T. Ma, D. Liu, H. Wang, and W. Hu, “A Multi-step ahead photovoltaic power forecasting model based on TimeGAN, Soft DTW-based K-medoids clustering, and a CNN-GRU hybrid neural network, ”Energy Reports, vol. 8, pp. 10346-10362, Nov. 2022.
[9] L. M. Liu, X. Y. Ren, F. Zhang, L. Gao, and B. Hao, “Dual-dimension Time-GGAN data augmentation method for improving the performance of deep learning models for PV power forecasting,” Energy Reports, vol. 9, pp. 6419-6433, Dec. 2023.
[10] J. Liu, H. Zang, F. Zhang, L. Cheng, T. Ding, Z. Wei, and G. Sun, “A hybrid meteorological data simulation framework based on time-series generative adversarial network for global daily solar radiation estimation,” Renewable Energy, vol. 219, Dec. 2023.
[11] Z. Wang, I. Koprinska, and M. Rana, “Clustering based methods for solar power forecasting,” 2016 International Joint Conference on Neural Networks (IJCNN), pp. 1487-1494, Jul. 2016.
[12] A. Bajpai, and M. Duchon, “A Hybrid Approach of Solar Power Forecasting Using Machine Learning,” 2019 3rd International Conference on Smart Grid and Smart Cities (ICSGSC), pp. 108-113, Jun. 2019.
[13] S. Cui, S. Lyu, Y. Ma, and K. Wang, “Improved informer PV power short-term prediction model based on weather typing and AHA-VMD-MPE,” Energy, vol. 307, Oct. 2024.
[14] Q. T. Phan, Y. K. Wu, Q. D. Phan, and H. Y. Lo, “A Novel Forecasting Model for Solar Power Generation by a Deep Learning Framework With Data Preprocessing and Postprocessing,” IEEE Transactions on Industry Applications, vol. 59, pp. 220-231, Jan.-Feb. 2023.
[15] X. Wang, Z. Wu, J. Ge, Z. Zhang, L. Han, and S. Wang, “Grid Load Forecasting Based on Dual Attention BiGRU and DILATE Loss Function,” IEEE Access, vol. 10, pp. 64569-64579, Jun. 2022.
[16] X. Wang, J. Dai, and Y. Liang, “Ultra-short-term photovoltaic power prediction model based on EMD-BiGRU,” 2024 5th International Conference on Computer Engineering and Application (ICCEA), pp. 1638-1642, Apr. 2024.
[17] P. Kumari, and D. Toshniwal, “Long short term memory–convolutional neural network based deep hybrid approach for solar irradiance forecasting,” Applied Energy, vol. 295, Aug. 2021.
[18] Y. Dai, W. Yu, and M. Leng, “A hybrid ensemble optimized BiGRU method for short-term photovoltaic generation forecasting,” Energy, vol. 299, Jul. 2024.
[19] A.L. Jonathan, D. Cai, C. C. Ukwuoma, N.J. J. Nkou, Q. Huang, and O. Bamisile, “A radiant shift: Attention-embedded CNNs for accurate solar irradiance forecasting and prediction from sky images,” Renewable Energy, vol. 234, Nov. 2024.
[20] Y. Zhou, N. Zhou, L. Gong, and M. Jiang, “Prediction of photovoltaic power output based on similar day analysis, genetic algorithm and extreme learning machine,” Energy, vol. 204, Aug. 2020.
[21] M. F. Tahir, M. Z. Yousaf, A. Tzes, M. S. E. Moursi, and T. H.M. El-Fouly, “Enhanced solar photovoltaic power prediction using diverse machine learning algorithms with hyperparameter optimization,” Renewable and Sustainable Energy Reviews, vol. 200, Aug. 2024.
[22] N. E. Michael, S. Hasan, A. Al-Durra, and M. Mishra, “Short-term solar irradiance forecasting based on a novel Bayesian optimized deep Long Short-Term Memory neural network,” Applied Energy, vol. 324, Oct. 2022.
[23] T. O. Omotehinwa, M. O. Lawrence, D. O. Oyewola, and E. G. Dada, “Bayesian optimization of one-dimensional convolutional neural networks (1D CNN) for early diagnosis of Autistic Spectrum Disorder,” Journal of Computational Mathematics and Data Science, vol. 13, p. 100105, Dec. 2024.
[24] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735-1780, Nov. 1997.
[25] K. Cho, B. Van Merriënboer, D. Bahdanau, and Y. Bengio, "On the properties of neural machine translation: Encoder-decoder approaches," arXiv preprint arXiv:1409.1259, Oct. 2014.
[26] M. Schuster and K. K. Paliwal, “Bidirectional recurrent neural networks,” IEEE Transactions on Signal Processing, vol. 45, no. 11, pp. 2673-2681, Nov. 1997.
[27] D. Bahdanau, K. Cho, and Y. Bengio, “Neural machine translation by jointly learning to align and translate,” arXiv preprint arXiv:1409.0473, Sep. 2014.
[28] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in Neural Information Processing Systems, vol. 30, pp. 5998-6008, Dec. 2017.
[29] J. Bergstra, R. Bardenet, Y. Bengio, and B. Kégl, ”Algorithms for hyper-parameter optimization,” Proceedings of the 25th International Conference on Neural Information Processing Systems (NIPS'11), pp. 2546–2554, Dec. 2011.
[30] 觀測氣象資料查詢系統，2024，台中氣象站(467490)日報表。[Online] Available：https://codis.cwa.gov.tw/StationData.
[31] The Weather Channel，2024，台中市大里區每小時天氣預報。[Online]. Available: https://www.weather.com.
[32] M. B. Kursa and W. R. Rudnicki, "Feature Selection with the Boruta Package," Journal of Statistical Software, vol. 36, no. 11, Sep. 2010.
[33] A. Rafie, P. Moradi, and A. Ghaderzadeh, “A Multi-Objective online streaming Multi-Label feature selection using mutual information,” Expert Systems with Applications, vol. 216, p. 119428, Apr. 2023.
[34] H. Gong, Y. Li, J. Zhang, B. Zhang, and X. Wang, “A new filter feature selection algorithm for classification task by ensembling pearson correlation coefficient and mutual information,” Engineering Applications of Artificial Intelligence, vol. 131, p. 107865, May. 2024.
[35] J. Yoon, D. Jarrett, and M. van der Schaar, “Time-series generative adversarial networks,” Proceedings of the 33rd International Conference on Neural Information Processing Systems, pp. 5508-5518, Dec. 2019.