我國能源局於2016年提出能源轉型政策，並規劃於2025年達成再生能源發電占比20%為目標，再生能源發電占比中又以發展風力及太陽光發電為主，其中太陽光電裝置容量為20 GW。再生能源間歇性發電的特性對電力系統的調度將帶來明顯的衝擊，再加上為解決高佔比再生能源發電所帶來系統運轉之挑戰，發展再生能源發電預測及各項輔助服務為首要任務。
本論文應用長短期記憶(LSTM)法及極限梯度提升(XGBoost)法分別建立太陽光電系統發電預測模型，首先透過數據預處理及對數據進行特徵選擇，選出與太陽光電系統首要發電相關之特徵做為模型之輸入參數，將兩種不同方法建立之模型因季節不同分為不同模型訓練，最後將模型預測之輸出使用誤差分析指標比較其模型預測精度，同時比較當使用經特徵選擇之特徵進行訓練之模型，與僅學習每季節中每日發電趨勢之模型在預測準確度上差異，模擬結果顯示:於太陽光電系統之發電預測，預測模型中有考慮特徵選擇參數之預測準確度，會高於僅學習發電趨勢之預測方式，且XGBoost在時間序列預測表現上優於擅長處理時間序列之LSTM算法。

Bureau of Energy in Taiwan proposed an energy transition policy in 2016, and plans to achieve the goal of 20% of the total power generation from renewable energy in 2025. Renewable power generation is mainly based on the development of wind and solar photovoltaic (PV) generation, of which the installation capacity target of PV generation is 20 GW. The intermittent power generation characteristics of renewable energy will have a significant impact on the dispatch of the power system. In addition, in order to solve the challenge of system operation caused by the high proportion of renewable energy power generation, the development of renewable energy power generation forecasts and various auxiliary services is the primary task.
In this thesis, the long and short-term memory (LSTM) method and the extreme gradient boosting (XGBoost) method are used to establish new generation forecasting models for the PV generation system. First of all, forecasting model related data and key feature parameters must be preprocessed and selected in sequence. Next, the most closely dependent key feature parameters of the PV generation will be inputted to its forecasting model. Both forecasting models (LSTM method and XGBoost method) will be trained individually according to different seasons. Finally, the error analysis index is used to evaluate accuracies of the two proposed forecasting models. Moreover, generation forecasting models with/without key feature parameters used are also intensively surveyed. The simulation results show that the model with key feature parameters used in training is better than the other. Moreover, the XGBoost algorithm have better time saving ability than the LSTM algorithm. It also can be concluded that even though the LSTM algorithm is specialized in applications of time series but the XGBoost algorithm have better performance in terms of the time-series forecasting.

[1] NREL, “Best research-cell efficiencies,” National Renewable Energy Laboratory, Apr. 2021. [Online]. Available: https://www.nrel.gov/pv/cell-efficiency.html.
[2] IRENA, “Renewable capacity statistics 2021,” International Renewable Energy Agency, Abu Dhabi, United Arab Emirates, Technical Report, Mar. 2021.
[3] R. L. Arantegui and A. Jäger-Waldau, “Photovoltaics and wind status in the European Union after the Paris Agreement,” Renewable and Sustainable Energy Reviews, vol. 81, no. 2, pp.2460-2471, Jan. 2018.
[4] IRENA, “Renewable energy prospects for the European Union,” International Renewable Energy Agency, Abu Dhabi, United Arab Emirates, Technical Report, Feb. 2018.
[5] IEA, “European Union 2020 – Energy Policy Review,” International Energy Agency, France, Technical Report, Jan. 2020.
[6] S. Potrč, L. Čuček, M. Martin, and Z. Kravanja, “Sustainable renewable energy supply networks optimization – The gradual transition to a renewable energy system within the European Union by 2050,” Renewable and Sustainable Energy Reviews, vol. 146, Aug. 2021.
[7] United States Environmental Protection Agency, “Overview of the American Recovery and Reinvestment Act of 2009,” July 2013.
[8] W. J. Cole, B. A. Frew, P. J. Gagnon, J. Richards, Y. Sun, R. M. Margolis, and M. A. Woodhouse, “SunShot 2030 for photovoltaics (PV): envisioning a low-cost PV future,” National Renewable Energy Laboratory (NREL), Golden, USA, Technical Report, Sept. 2017.
[9] EIA, “Renewable energy explained,” U.S. Energy Information Administration (EIA), May 2021. [Online]. Available: https://www.eia.gov/energyexplained/renewable-sources/.
[10] EIA, “Annual Energy Outlook 2021,” U.S. Energy Information Administration (EIA), Washington, USA, Technical Report, Fab. 2021.
[11] 胡仕儀，2012，「台灣能不能?!?!政策簇擁日本太陽能發電2030年衝100GW目標更明確」，第102期，第12-16頁。
[12] 林祥輝，2017，「日本再生能源政策論點與目標 ─加強成本競爭力、強化自立環境、改善電力系統確保適當調整力」
[13] 張瓊之，2020，「國家能源政策評析報告：日本」
[14] Z. A. Liza and M. R. Islam, “Solar park: the next generation energy source in Bangladesh,” Journal of Energy Research and Reviews, vol. 4. no. 2, pp. 9-19, Feb. 2020.
[15] 2020_2021產業技術白皮書-我國產業創新研發政策推動。
[16] 2020_2021產業技術白皮書-我國重點產業需求與技術發展課題。
[17] R. Inman, H. Pedro, and C. Coimbra, "Solar forecasting methods for renewable energy integration," Progress in Energy and Combustion Science, vol. 39, pp. 535–576,2013.
[18] 何清，李寧，羅文娟，史忠植，「大數據下的機器學習算法綜述」，2014年4月。
[19] 曾婉菁，「機器學習探究」，第三十四卷第二期。
[20] 余凱，賈磊，陳雨強，徐偉伟，「深度學習的昨天、今天和明天」，2013年。
[21] J. Antonanzas, N. Osorio, R. Escobar, R. Urraca, F. J. Ascacibar, and F. Antonanzas, "Review of photovoltaic power forecasting," Solar Energy, vol. 136, June 2016.
[22] Sobrina Sobria, Sam Koohi-Kamalia, Nasrudin Abd, Rahim, “Solar photovoltaic generation forecasting methods: A review”, Energy Conversion and Management, Vol.156 , pp.459-497,2018
[23] Shahzad Muzaffar and Afshin Afshari, “Short-Term Load Forecasts Using LSTM Networks”, Energy Procedia, Vol.158, pp.2922-2927.2019.
[24] 王鑫，吳際，劉超，楊海燕，杜艳麗，牛文生，「基于LSTM循环神经网络的故障时间序列预测」，2018年4月。
[25] A. Alzahrani, P. Shamsi, C. Dagli, and M. Ferdowsi, "Solar Irradiance Forecasting Using Deep Neural Networks," Procedia Computer Science, vol. 114, pp. 304-313, January 2017.
[26] Y. Wu and B. Wang, “Research on EA-Xgboost hybrid model for building energy prediction”, Journal of Physics: Conference Series, vol. 1518, article no. 012082, 2020.
[27] Huan Zheng and Yanghui Wu, “A XGBoost Model with Weather Similarity Analysis and Feature Engineering for Short-Term Wind Power Forecasting”, July 2019.
[28] RENEWABLES, “GLOBAL STATUS REPORT”.2020.
[29] Director (Joint Secretary), Renewable Energy Development Sustainable and Renewable Energy Development Authority (SREDA) Power Division, Ministry of Power, Energy and Mineral Resources Government of Bangladesh , “Sustainable Energy Development”, December 2020.
[30] Tetra Tech ES, “Challenges in the Development of Variable Renewable Energy in Bangladesh”, June 2020.
[31] 經濟部，「能源轉型白皮書核定本」，2020年11月。
[32] 經濟部，「109年太陽光電6.5GW達標計畫核定本」，2019年10月。
[33] 熊谷秀，「太陽光電在台灣-陽光屋頂百萬座」，電力技術專欄，第18期，2015年12月。
[34] 吳進忠，「再生能源併聯運轉對電力調度的挑戰與機會」，2018年8月。
[35] 顧鴻濤，「太陽能電池元件導論」，全威圖書有限公司，2008年。
[36] 吳財福，「綠色能源直流供電系統介紹」，國立中正大學，2009年 11月。
[37] 中央氣象局氣象預報中心，
file:///C:/Users/USER/Desktop/%E8%AB%96%E6%96%87%E5%8F%83%E8%80%83%E6%96%87%E7%8D%BB/%E6%B0%A3%E8%B1%A1%E9%A0%90%E5%A0%B1.pdf
[38] 交通部中央氣象局，
https://www.cwb.gov.tw/V8/C/K/Encyclopedia/nous/index.html#introduction-05
[39] 洪丽，「太阳能光伏电站短期发电预测研究」，東南大學，2013年，4月。
[40] D. Locke, “Mq telemetry transport (mqtt) v3. 1 protocol specification,”IBMdeveloperWorksTechnicalLibrary, August 2010,
https://www.ibm.com/developerworks/webservices/library/ws-mqtt/
[41] U. Hunkeler, H. L. Truong, and A. Stanford-Clark, “MQTT-S—Apublish/subscribe protocol for wireless sensor networks,” In Communication systems software and middleware and workshops, Vol.138 pp.791-798,2008.
[42] Standard, “OASIS Advanced Message Queuing Protocol (AMQP) Version 1.0,” OASIS, October, 2012.
[43] 毕艳冰、蒋林、张大海、王新军、崔立真，「基於實時數據分發服務的智能變電站IEC 61850的實現方法」，中國電機工程學報，第33卷，第7期，第149-155頁，2013年3月。
[44] H. León, C. Montez, and M. Stemmer, “Simulation models for IEC 61850 communication in electrical substations using GOOSE and SMV time-critical messages,” in Proc. IEEE World Conference on Factory Communication Systems (WFCS), Aveiro, Portugal, May 2016.
[45] IEC 61850-7-4, Communication Networks and Systems in Substations Part 7-4: Basic Communication Structure for Substation and Feeder Equipment Compatible Logical Node Classes and Data Classes.
[46] IEC 61850-7-420, Communication Networks and Systems for Power Utility Automation for Distributed Energy Resources(DER) Part 7-420.
[47] IEC 57 (2011). IEC 61850-7-1 edition 2.0, Communication networks and systems for power utility automation Part 7-1 ： Basiccommunication structure Principles and models, p. 70.
[48] 李永亮，等，「IEC61850第2版簡介及其在智能電網中的應用展望」電網技術期刊，第34卷，第4期，第11-16頁，2010年。
[49] 任雁铭，等，「IEC61850新動向和新應用」電力系統自動化期刊，第37卷，第2期，第1-6頁，2013年1月。
[50] IEC 61850-8-2 Ed1, Communication networks and systemsfor power utility automation Part 8-2: Specific communication service mapping (SCSM) -Mapping to Extensible Messaging Presence Protocol (XMPP), March, 2017.
[51] Saint-Andre,P. “Extensible Messaging and Presence Protocol(XMPP): Core,” Internet Engineering Task Force (IETF). Available: https://tools.ietf.org/html/rfc6120, March, 2011.
[52] Maria Malvoni,Maria Grazia De Giorgi,Paolo Marria Congedo, “Forecasting of PV Power Generation using weather input data-preprocessing techniques”,Energy Procedia,Vol.126,pp.651-658, September 2017.
[53] Tianqi Chen and Carlos Guestrin , “XGBoost A Scalable Tree Boosting System”, June 2016.
[54] 劉俊宏，「應用機器學習於IEC 61850 XMPP 分散式能源雲平台之太陽光伏發電預測」，2019年12月。