近年，隨著台灣對2050年達到淨零碳排設定的策略目標，再生能源技術的研究與發展受到極大的重視。同時，國際間對於再生能源憑證的關注與需求亦日益增加，該憑證成為驗證再生能源使用與碳足跡減少的重要工具，且對供應鏈企業在應對上游業者需求具有特殊的商業價值。為進一步鼓勵再生能源的利用，並配合企業取得再生能源憑證的需求，《電業法》已著手推動電力交易平台的建置，其目的在於引導能源業者走向電能購售的新型態，且讓用戶不再只能跟台灣電力公司購電。
針對《電業法》及台灣電力公司的轉供辦法，當前台灣的購售電契約主要以一年為單位週期。此背景下，售電業者必須精準評估用戶和再生能源案場的長短期發電與用電量。短期資料的目的在於確認用戶用電時間曲線與再生能源發電曲線的匹配度，以及其有效媒合的可能性；而長期資料則為設定購售電契約參數提供關鍵依據。
本研究使用一套融合梯度下降和學習率的轉供契約參數設定方法。為增強媒合的準確性，採用了基於Transformer-LSTM的深度學習框架來估算案場的發電量和用戶的用電需求。透過這項研究，我們期望可以為售電業者提供一種更精確、高效的工具，以便實現最佳化的能源分配同時兼顧有效的再生能源憑證取得策略。

In recent years, with Taiwan setting a strategic goal to achieve net-zero carbon emissions by 2050, research and development in renewable energy technologies have garnered unprecedented attention. Simultaneously, the global focus and demand for renewable energy certificates are on the rise. These certificates have become crucial tools for verifying the use of renewable energy and reducing carbon footprints, offering unique commercial value to supply chain enterprises in addressing the requirements of upstream businesses. To further promote the use of renewable energy and align with corporate demands for renewable energy certificates, Taiwan has initiated the establishment of an electricity trading platform. The intention is to guide energy providers towards a new paradigm of electricity trading and to allow consumers to purchase electricity from providers other than the Taiwan Power Company.
Considering the Taiwan Power Company's reselling regulations, the prevalent electricity trading contracts primarily adopt a one-year cycle. Electricity sellers need to precisely evaluate the long-term and short-term electricity generation and consumption of consumers and renewable energy projects. Short-term data aims to verify the alignment of the consumer's electricity consumption curve with the renewable energy generation curve, as well as the feasibility of efficient matchmaking. In contrast, long-term data serves as a crucial foundation for setting electricity sale contract parameters.
This study introduces a contractual parameter optimization method that integrates gradient descent and learning rate concepts. To enhance the accuracy of matchmaking, a deep learning framework based on the Transformer-LSTM has been adopted to estimate the electricity generation of projects and the consumption demands of consumers. Through this research, we aspire to equip electricity sellers with a more accurate toolkit to achieve efficient energy allocation and acquire renewable energy certificates that better meet demand.

[1] (2023). 再生能源裝置容量. [Online] Available: https://www.geipc.tw/LiveEnergy.aspx
[2] (2021). 再生能源憑證制度及綠電交易介紹. [Online] Available: https://www.trec.org.tw/activity/2021-09-10%2011-54-37/get/j0kUZgUt7EsgNpZTqmKW?v=1631246527
[3] T. Georgitsioti, N. Pearsall, I. Forbes, and G. Pillai, "A combined model for PV system lifetime energy prediction and annual energy assessment," Solar Energy, vol. 183, pp. 738-744, 2019/05/01/ 2019, doi: https://doi.org/10.1016/j.solener.2019.03.055.
[4] D. Thevenard and S. Pelland, "Estimating the uncertainty in long-term photovoltaic yield predictions," Solar Energy, vol. 91, pp. 432-445, 2013/05/01/ 2013, doi: https://doi.org/10.1016/j.solener.2011.05.006.
[5] N. Cristianini and E. Ricci, "Support Vector Machines," in Encyclopedia of Algorithms, M.-Y. Kao Ed. Boston, MA: Springer US, 2008, pp. 928-932.
[6] S. Taylor and B. Letham, "Forecasting at Scale," The American Statistician, vol. 72, 09/29 2017, doi: 10.1080/00031305.2017.1380080.
[7] T. Chen and C. Guestrin, "XGBoost: A Scalable Tree Boosting System," p. arXiv:1603.02754doi: 10.48550/arXiv.1603.02754.
[8] S. Hochreiter and J. Schmidhuber, "Long Short-Term Memory," Neural Comput., vol. 9, no. 8, pp. 1735–1780, 1997, doi: 10.1162/neco.1997.9.8.1735.
[9] A. Vaswani et al., "Attention Is All You Need," p. arXiv:1706.03762doi: 10.48550/arXiv.1706.03762.
[10] D. P. Ananthu and K. N, "A study of 100kWp PV Plant Output Power Forecasting: A case study," in 2021 5th International Conference on Computing Methodologies and Communication (ICCMC), 8-10 April 2021 2021, pp. 723-728, doi: 10.1109/ICCMC51019.2021.9418418.
[11] M. Bouzerdoum, A. Mellit, and A. Massi Pavan, "A hybrid model (SARIMA–SVM) for short-term power forecasting of a small-scale grid-connected photovoltaic plant," Solar Energy, vol. 98, pp. 226-235, 2013/12/01/ 2013, doi: https://doi.org/10.1016/j.solener.2013.10.002.
[12] S. Ghosh and P. K. Gupta, "Forecasting of Solar Power Volatility using GJR-GARCH method," in 2021 IEEE Electrical Power and Energy Conference (EPEC), 22-31 Oct. 2021 2021, pp. 261-266, doi: 10.1109/EPEC52095.2021.9621364.
[13] R. Gupta, A. K. Yadav, S. Jha, and P. K. Pathak, "Time Series Forecasting of Solar Power Generation Using Facebook Prophet and XG Boost," in 2022 IEEE Delhi Section Conference (DELCON), 11-13 Feb. 2022 2022, pp. 1-5, doi: 10.1109/DELCON54057.2022.9752916.
[14] D. Kothona, K. Spyropoulos, C. Valelis, C. Sarigiannidis, K. C. Chatzisavvas, and G. C. Christoforidis, "Efficient 24-hour ahead PV energy production forecasting employing a transformer-based model," in 2022 2nd International Conference on Energy Transition in the Mediterranean Area (SyNERGY MED), 17-19 Oct. 2022 2022, pp. 1-6, doi: 10.1109/SyNERGYMED55767.2022.9941461.
[15] M. López Santos, X. García-Santiago, F. Echevarría Camarero, G. Blázquez Gil, and P. Carrasco Ortega, "Application of Temporal Fusion Transformer for Day-Ahead PV Power Forecasting," Energies, vol. 15, no. 14, doi: 10.3390/en15145232.
[16] (2022). 台灣電力股份有限公司電能轉供及併網型直供營運規章. [Online] Available: https://service.taipower.com.tw/powerwheeling/static/file/1120620%E8%BD%89%E7%9B%B4%E4%BE%9B%E7%87%9F%E9%81%8B%E8%A6%8F%E7%AB%A0.pdf