太陽光能系統是智慧永續基礎建設的關鍵產業。一場從傳統能源轉向投資再生能源的全球浪潮正風起雲湧。太陽光能系統扮演的角色是將傳統集中式能源系統轉向分散式能源系統。代表能源供應的形式將轉型成更直接面對使用者，有按用戶需求直接就地生產以供應能源的性質，惟分散式能源系統對於使用單位的技術要求比簡單使用大電網供電高，需要相應的技術與文化環境。另方面，因為數據採集的過程或是設備更新維修都會導致數據缺失的問題，這是我們要建立智慧永續基礎建設的時代背景下亟需改善的問題。因此，建立考量數據缺失的智慧型太陽光能系統輸出預測的研究，是現在重要且具有挑戰性的議題。過去有使用深度學習的方式對太陽光能的發電量進行預測，也有針對太陽光能缺失數據填補的研究，但是沒有整合考量缺失數據與深度學習的系統決策架構。因此，本研究提出一個決策架構，考量了從判斷數據缺失類型到選擇數據填補方法，並且進行預測的完整決策過程。另方面，本研究還考量了另外兩種數據缺失的情形，提出了一種新的太陽光能的近鄰數據之填補方式，且考量使用自編碼器作為特徵提取的方式應用在長短期記憶模型做系統輸出的預測實驗。其中，本研究比較了兩種不同自編碼器和四種不同結構的長短期記憶模型對預測能力的差異。另方面，在缺失率占20%的填補實驗裡，使用本研究提出的填補方法，再將自編碼器的長短期記憶模型進行系統輸出預測可以將誤差從RMSE 7.19降到6.86，RMSLE可以從0.37降到0.35。最後，本研究重複測試了兩個季節的數據，並根據結果提出一個整合不同缺失數據情景下，該使用何種缺失數據填補方式及預測模型的決策過程。最後，本論文的結尾為未來的研究方向提出了四項建議。

The solar energy system is a crucial industry for the construction of intelligent and sustainable infrastructure. The role of it is to shift the traditional centralized energy system to a distributed energy system. Represents the transformation of the form of energy supply to be more directly facing users and has the nature of on-demand producing to the user to supply energy. However, the technical requirements of the distributed system for localization users are higher than simply using large-scale power grids. Besides, during the data collection or equipment update and maintenance will cause data missing, this problem demanding need to be improved in the state-of-the-art works. Therefore, research on the intelligent decision-making system that considers missing solar energy value is now an important and challenging topic. In previous research, deep learning techniques were used to predict solar power generation, and there were also researches on the imputation of missing solar energy data. However, no system decision-making architecture integrated consideration of missing data and deep learning to make a prediction before. Therefore, this research proposes a decision-making framework that considers the complete decision-making process from judging the data missing to selecting the data imputation method and making predictions. This research also considers the other two types of missing data, proposes a new solar system neighboring area imputation method, and considers using Autoencoders as a feature extraction approach that fits the LSTM models experiment. This work compared the prediction ability of two different autoencoders and four different structures of LSTM models. Besides, in the imputation research with a missing rate of 20%, using the imputation method proposed in this study, using the Autoencoder-LSTM model to predict the system output can reduce the RMSE 7.19 to 6.86, RMSLE is reduced from 0.37 to 0.35. Finally, this study repeatedly tested the two seasons' data, and based on the results, proposed a decision-making process of which imputation and prediction model should be used under different missing data scenarios. Furthermore, it gave four recommendations for future research at the end of this thesis.

AlSkaif, T., Dev, S., Visser, L., Hossari, M., & van Sark, W. (2020). A systematic analysis of meteorological variables for PV output power estimation. Renewable Energy, 153, 12-22.
Banks, D., House, L., Arabie, P., McMorris, F., & Gaul, W. (2004). Classification, Cluster Analysis, and Data Mining. Springer-Verlag, Berlin.
Batista, G. E., & Monard, M. C. (2002). A study of k-nearest neighbour as an imputation method. Applied artificial intelligence, 87(251-260), 48.
Batista, G. E., & Monard, M. C. (2003). An analysis of four missing data treatment methods for supervised learning. Applied artificial intelligence, 17(5-6), 519-533.
Central Weather Bureau(2015) , CWB Observation Data Inquire System. Retrieved Mar. 01, 2021, from https://e-service.cwb.gov.tw/HistoryDataQuery/index.jsp.
Chollet, F. (2018). Deep learning with Python (Vol. 361): New York : Manning.
Das, U. K., Tey, K. S., Seyedmahmoudian, M., Idna Idris, M. Y., Mekhilef, S., Horan, B., & Stojcevski, A. (2017). SVR-based model to forecast PV power generation under different weather conditions. Energies, 10(7), 876.
Dwyer, S., & Teske, S. (2018). Renewables 2018 Global Status Report. Renewables Global Status Report.
García-Laencina, P. J., Sancho-Gómez, J.-L., Figueiras-Vidal, A. R., & Verleysen, M. (2009). K nearest neighbours with mutual information for simultaneous classification and missing data imputation. Neurocomputing, 72(7-9), 1483-1493.
Gensler, A., Henze, J., Sick, B., & Raabe, N. (2016). Deep Learning for solar power forecasting—An approach using AutoEncoder and LSTM Neural Networks. Paper presented at the 2016 IEEE international conference on systems, man, and cybernetics (SMC), Budapest.
Gilani, S. I.-u.-H., Dimas, F. A. R., & Shiraz, M. (2011). Hourly solar radiation estimation using ambient temperature and relative humidity data. International Journal of Environmental Science Development, 2(3), 188-193.
Hinton, G. E., & Salakhutdinov, R. R. (2006). Reducing the dimensionality of data with neural networks. Science, 313(5786), 504-507.
Hruschka, E. R., Hruschka, E. R., & Ebecken, N. F. (2004). Towards efficient imputation by nearest-neighbors: A clustering-based approach. Paper presented at the Australasian Joint Conference on Artificial Intelligence, Berlin, Heidelberg.
Hsu, Lin(2016) 開放資料、大數據及資料探勘之研究再生能源數據倉儲之建置與應用。Retrieved Mar. 01, 2021, from http://140.120.49.84/
Kardakos, E. G., Alexiadis, M. C., Vagropoulos, S. I., Simoglou, C. K., Biskas, P. N., & Bakirtzis, A. G. (2013). Application of time series and artificial neural network models in short-term forecasting of PV power generation. Paper presented at the 2013 48th International Universities' Power Engineering Conference (UPEC), Dublin, Ireland.
Kim, T., Ko, W., & Kim, J. (2019). Analysis and impact evaluation of missing data imputation in day-ahead PV generation forecasting. Applied Sciences, 9(1), 204.
Kwak, S. G., & Kim, J. H. (2017). Central limit theorem: the cornerstone of modern statistics. Korean journal of anesthesiology, 70(2), 144.
Layanun, V., Suksamosorn, S., & Songsiri, J. (2017). Missing-data imputation for solar irradiance forecasting in Thailand. Paper presented at the 2017 56th Annual Conference of the Society of Instrument and Control Engineers of Japan (SICE), Kanazawa, Japan.
Lee, D., & Kim, K. (2019). Recurrent neural network-based hourly prediction of photovoltaic power output using meteorological information. Energies, 12(2), 215.
Li, L.-L., Cheng, P., Lin, H.-C., & Dong, H. (2017). Short-term output power forecasting of photovoltaic systems based on the deep belief net. Advances in mechanical engineering, 9(9), doi: 1687814017715983.
Li, Y., He, Y., Su, Y., & Shu, L. (2016). Forecasting the daily power output of a grid-connected photovoltaic system based on multivariate adaptive regression splines. Applied Energy, 180, 392-401.
Luengo, J., García, S., & Herrera, F. (2010). A study on the use of imputation methods for experimentation with Radial Basis Function Network classifiers handling missing attribute values: the good synergy between RBFNs and EventCovering method. Neural Networks, 23(3), 406-418.
Massidda, L., & Marrocu, M. (2017). Use of Multilinear Adaptive Regression Splines and numerical weather prediction to forecast the power output of a PV plant in Borkum, Germany. Solar Energy, 146, 141-149.
Neo, Y., Teo, T. T., Woo, W. L., Logenthiran, T., & Sharma, A. (2017). Forecasting of photovoltaic power using deep belief network. Paper presented at the Tencon 2017-2017 IEEE Region 10 Conference, Penang, Malaysia.
Panapakidis, I. P., Bouhouras, A. S., & Christoforidis, G. C. (2018). A missing data treatment method for photovoltaic installations. Paper presented at the 2018 IEEE International Energy Conference (ENERGYCON), Limassol, Cyprus.
Shi, J., Lee, W.-J., Liu, Y., Yang, Y., & Wang, P. (2012). Forecasting power output of photovoltaic systems based on weather classification and support vector machines. IEEE Transactions on Industry Applications, 48(3), 1064-1069.
Sovilj, D., Eirola, E., Miche, Y., Björk, K.-M., Nian, R., Akusok, A., & Lendasse, A. (2016). Extreme learning machine for missing data using multiple imputations. Neurocomputing, 174, 220-231.
Troyanskaya, O., Cantor, M., Sherlock, G., Brown, P., Hastie, T., Tibshirani, R., . . . Altman, R. B. (2001). Missing value estimation methods for DNA microarrays. Bioinformatics, 17(6), 520-525.
Turrado, C. C., López, M. d. C. M., Lasheras, F. S., Gómez, B. A. R., Rollé, J. L. C., & Juez, F. J. d. C. (2014). Missing data imputation of solar radiation data under different atmospheric conditions. Sensors, 14(11), doi: 20382-20399.
Voyant, C., Notton, G., Kalogirou, S., Nivet, M.-L., Paoli, C., Motte, F., & Fouilloy, A. J. R. E. (2017). Machine learning methods for solar radiation forecasting: A review. 105, 569-582.
Xu, X., Chong, W., Li, S., Arabo, A., & Xiao, J. (2018). MIAEC: Missing data imputation based on the evidence chain. IEEE Access, 6, doi: 12983-12992.
Yozgatligil, C., Aslan, S., Iyigun, C., & Batmaz, I. (2013). Comparison of missing value imputation methods in time series: the case of Turkish meteorological data. Theoretical applied climatology, 112(1), 143-167.
Zhang, Y., Jin, C., Sharma, R. K., & Srivastava, A. K. (2019). Data-driven day-ahead PV estimation using hybrid deep learning. Paper presented at the 2019 IEEE Industry Applications Society Annual Meeting, Baltimore, Maryland.
Zhang, Y., Qin, C., Srivastava, A. K., Jin, C., & Sharma, R. K. (2020). Data-Driven Day-Ahead PV Estimation Using Autoencoder-LSTM and Persistence Model. IEEE Transactions on Industry Applications, 56(6), doi: 7185-7192.
Zheng, J., Zhang, H., Dai, Y., Wang, B., Zheng, T., Liao, Q., . . . Song, X. (2020). Time series prediction for output of multi-region solar power plants. Applied Energy, 257, doi: 114001.