為了降低全球暖化與氣候變遷對社會帶來的衝擊，各國政府積極發展可再生能源電 力系統，其中以太陽能發展較為成熟，其可及性和儲電效率相較其他再生能源更有優勢。 與太陽能電廠產電量最相關的自然資源就是日照時數，因此本研究主要目的為探討日照 時數的預測模型，比較日照預測模型的配適與預測表現，以利太陽能電廠未來的規劃與 價值評估。資料方面，除了日照時數本身，也考量其他氣象因子與日照時數的相關度， 並對資料進行季節性分解。訓練模型方面，採用三個類型的多種模型:第一類時間序列 模型選擇 AR(1)、ARMA(1,1)、ARMAX(1,1)和 GARCH(1,1)模型，第二類機器學習模型 選擇線性迴歸、迴歸樹和 LSTM 模型，第三類混合模型就是將時間序列模型與機器學習 模型混合，像是 AR(1)- LSTM 和 ARMAX(1,1)-LSTM 的混合模型。研究結果發現，取去 除季節性後的資料，訓練出的模型預測誤差相較保留季節性趨勢資料來的小，多種模型 中，含有季節性分解的 ARMAX(1,1)-LSTM 混合模型預測誤差最小，其次預測誤差最小 的是 ARMAX(1,1)模型，這兩種模型都有將其他氣象因子作為特徵變數訓練模型，說明 加入更多特徵變數資料使模型能更準確進行預測。整體而言，混合模型進行預測比單一 模型預測表現更為優異。

To reduce the impact of global warming and climate impact on society, governments worldwide are activity developing renewable energy power systems. Among these, solar energy is relatively mature and offers advantages in accessibility and storage efficiency compared to other kind of renewable energy. The most closely relate natural resource to the power generation of solar power plants is sunshine duration. Therefore, the main objective of this study is to explore prediction models for sunshine duration, compare the fitting and prediction performance of different models, and thereby facilitate the planning and value assessment of solar power stations.
Regarding the data, besides sunshine duration itself, other meteorological factors correlated with sunshine duration are also considered, and perform seasonal decomposition on the data. For model training, three types of models are used: Time Series models including AR(1), ARMA(1,1), ARMAX(1,1), and GARCH(1,1) models; Machine Learning models including Linear Regression, Regression Tree, and LSTM models; and hybrid models that combine Time Series model and Machine Learning model, such as AR(1)-LSTM and ARMAX(1,1)-LSTM hybrid models.
The study found that models trained on residual data after removing seasonality had smaller prediction errors compared to those retaining seasonal trend data. Among the various models, the ARMAX(1,1)-LSTM hybrid model with seasonal decomposition had the smallest prediction error, followed by the ARMAX(1,1) model. Both models used other meteorological factors as feature variables, indicating that incorporating more feature variables enables more accurate predictions. Overall, hybrid models perform better in prediction compared to single models.

[1] 徐天佑、曾鴻陽 (2007)。
55-72.
參考文獻
台灣地區有關太陽能日照量之環境時空因素研究探討。環
境教育學刊，N. 6，2007。
[2] 柴德美、劉鹏 (2018)。定陶日照時數變化特徵及影響因子分析。Climate Change
Research Letters，7，110。
[3] Ahmad, M. W., Mourshed, M., & Rezgui, Y. (2018). Tree-based ensemble methods for
predicting PV power generation and their comparison with support vector regression.
Energy, 164, 465-474.
[4] Alfadda, A., Rahman, S., & Pipattanasomporn, M. (2018). Solar irradiance forecast using
aerosols measurements: A data driven approach. Solar Energy, 170, 924-939.
[5] Bartoszek, K., Matuszko, D., & Węglarczyk, S. (2021). Trends in sunshine duration in
Poland (1971–2018). International Journal of Climatology, 41(1), 73-91.
[6] Breiman, L. (1996). Heuristics of instability and stabilization in model selection. The
Annals of Statistics, 24(6), 2350-2383.
[7] Chandola, D., Gupta, H., Tikkiwal, V. A., & Bohra, M. K. (2020). Multi-step ahead
forecasting of global solar radiation for arid zones using deep learning. Procedia
Computer Science, 167, 626-635.
[8] Curry, B., & Morgan, P. H. (2010). Seasonality and neural networks: A new approach.
International Journal of Metaheuristics, 1(2), 181-197.
[9] David, M., Ramahatana, F., Trombe, P. J., & Lauret, P. (2016). Probabilistic forecasting
of the solar irradiance with recursive ARMA and GARCH models. Solar Energy, 133,
[10] Jonsson, P., & Eklundh, L. (2003). Seasonality extraction from time-series of satellite
sensor data. IEEE transactions on Geoscience and Remote Sensing, 40(8), 1824-1832.
[11] Kannan, N., & Vakeesan, D. (2016). Solar energy for future world: A review. Renewable
and Sustainable Energy Reviews, 62, 1092-1105.
[12] Li, Y., Su, Y., & Shu, L. (2014). An ARMAX model for forecasting the power output of a
grid connected photovoltaic system. Renewable Energy, 66, 78-89.
[13] Mayer, K., Schmid, T., & Weber, F. (2015). Modeling electricity spot prices: combining
mean reversion, spikes, and stochastic volatility. The European Journal of Finance,
21(4), 292-315.
[14] Mubiru, J. (2008). Predicting total solar irradiation values using artificial neural
networks. Renewable Energy, 33(10), 2329-2332.
[15] Reikard, G. (2009). Predicting solar radiation at high resolutions: A comparison of time
series forecasts. Solar Energy, 83(3), 342-349.
[16] Sharma, V., Yang, D., Walsh, W., & Reindl, T. (2016). Short term solar irradiance
forecasting using a mixed wavelet neural network. Renewable Energy, 90, 481-492.
[17] Wang, Z., Wang, Y., Zeng, R., Srinivasan, R. S., & Ahrentzen, S. (2018). Random forest
based hourly building energy prediction. Energy and Buildings, 171, 11-25.
[18] Yadav, A. K., & Chandel, S. S. (2014). Solar radiation prediction using Artificial Neural
Network techniques: A review. Renewable and Sustainable Energy Reviews, 33, 772-
781.
[19] Yang, D., Jirutitijaroen, P., & Walsh, W. M. (2012). Hourly solar irradiance time series
forecasting using cloud cover index. Solar Energy, 86(12), 3531-3543.
[20] Zendehboudi, A., Baseer, M. A., & Saidur, R. (2018). Application of support vector
machine models for forecasting solar and wind energy resources: A review. Journal of
Cleaner Production, 199, 272-285.
[21] Zhang, G. P. (2003). Time series forecasting using a hybrid ARIMA and neural network
model. Neurocomputing, 50, 159-175.