臺灣博碩士論文加值系統

因應全球各地高度重視氣候變遷與追求永續發展的議題，急需乾淨且可再生的綠色能源應用於電力系統中，在這些可再生能源中，風力發電更是不可或缺的一環。風力發電一直受到全世界的關注，加上台灣在西半部地區有良好的地域以及條件、政策的極力推廣、所需成本逐年下降等諸多原因，使台灣越來越適合發展風力發電，以及進行相關的研究與評估其效益。

因此，本研究提出了以深度學習為基礎，將卷積神經網路和長短期記憶網路兩種方法作結合，來對風力發電量作預測以及評估準確性。這樣的結合技術取用了卷積神經網路擅長於特徵擷取和長短期記憶網路擅於處理時間序列之數據兩者各自的優點，並將預測結果進行點預測、概率預測評估，其中也使用多步預測的方式來輸出並觀察不同時間點的預測結果表現，而概率預測部分則使用非參數型的經驗累積分佈函數(Empirical Cumulative Distribution Function, ECDF)來定義預測區間的上下限。

本文為了證明所提方法於風力發電量預測準確性，將卷積神經網路模型與長短期記憶模型也一併進行預測，驗證所提模型是否完整將數據中的有效特徵擷取出來以提升學習精準度並進行預測。以此方式，最後以點預測、概率預測、多步點預測、多步概率預測以及不同預測區間之概率預測結果來進行三方比較，驗證所提模型表現和其穩定性優於另外兩種方法。

關鍵詞：深度學習、機器學習、點預測、概率預測、多步預測、卷積神經網路、長短期記憶網路、特徵擷取、非參數型概率、ECDF

In response to the issues of climate change and the pursuit of sustainable development in the world. In renewable energy, wind power is an indispensable part. Taiwan has a good geographical area in the west, as well as the promotion of conditions and policies, and the cost reduction year by year. They are all making Taiwan more suitable for developing wind power, as well as conducting related research and evaluating its benefits.

This paper proposes to combine the two methods of Convolutional Neural Network (CNN) and Long Short Term Memory Network (LSTM) based on deep learning. These techniques take advantages of the CNN's expertise in feature extraction and LSTM are good at processing time series data. Then, the results are subjected to point forecast and probabilistic forecast evaluation. In them also uses multi-step forecast to output and observe the forecast results at different time points. The probabilistic forecast uses Empirical Cumulative Distribution Function (ECDF) to define the upper and lower limits of the prediction interval.

In order to prove the accuracy of the proposed method for wind power generation, the CNN model and LSTM models are also predicted to verify whether the proposed model is complete and the effective features in the data are extracted to improve learning accuracy and predict. In this way, the three-party comparison is carried out with point forecast, probabilistic forecast, multi-step forecast, multi-step probabilistic forecast and probabilistic forecast results of different prediction intervals to verify that the proposed model performance and its stability are better than the other two methods.

Index Terms- Deep learning, machine learning, point forecast, probabilistic forecast, multi-step forecast, CNN, LSTM, feature extraction, ECDF

誌謝 i
摘要 ii
Abstract iii
目錄 iv
圖目錄 vii
表目錄 ix
第一章 緒論 1
1.1 研究動機與背景 1
1.2 文獻回顧 2
1.3 研究目的 3
1.4 研究貢獻 4
1.5 章節內容 4
第二章 風力發電產業現況與發展趨勢 7
2.1 全球風力發電發展趨勢 7
2.2 台灣現況 9
第三章 深度學習方法 13
3.1 卷積神經網路 13
3.1.1 卷積層 14
3.1.2 池化層 15
3.1.3 卷積神經網路模型 17
3.2 長短期記憶網路 18
第四章 風力發電量預測模型 21
4.1 風力發電數據集說明 21
4.2 風力發電數據集與發電量相關性分析 25
4.3 CNN-LSTM架構 29
4.4 時間序列連續多步預測方法 31
4.4.1 遞迴預測 32
4.4.2 直接預測 32
4.5概率預測方法 33
4.5.1 機率密度函數 33
4.5.2 累積分佈函數 34
4.5.3 信賴區間 34
4.5.4 預測區間 35
4.6非參數型概率預測方法 37
第五章 模擬結果與分析 43
5.1 風力發電訓練和測試資料 43
5.2 點預測模擬結果 44
5.2.1 點預測評估指標 44
5.2.2 點預測結果比較 45
5.2.3 多步點預測結果比較 48
5.3 概率預測模擬結果 52
5.3.1 概率預測結果呈現 52
5.3.2 多步概率預測結果比較 55
第六章 結論 59
6.1 結論 59
6.2 未來工作 59
參考文獻 61
 

[1]A.M. Foley, A. Marvuglia, P.G. Leahy and E.J. McKeogh, “A Review of Current Methods and Advances in Wind Power Prediction and Forecasting,” Renewable Energy, vol. 7, no. 1, pp. 1-8, 2012.
[2]Y. Zhan, Q.P. Zheng, J. Wang, P. Pinson, “Decision-dependent stochastic generation expansion planning with large amounts of wind power,” IEEE Trans. on Power Sys, vol. 32, no. 4, pp. 3015-3026, 2017.
[3]I. Okumus and A. Dinler, “Current status of wind energy forecasting and a hybrid method for hourly predictions,” Energy Conversion and Management, vol. 123, pp. 362–371, Sep. 2016.
[4]Mukund R. Patel, Wind and Solar Power Systems: Design, Analysis, and Operation, Second Edition. 2005.
[5]E. B. Ssekulima, M. B. Anwar, A. A. Hinai, and M. S. E. Moursi, “Wind speed and solar irradiance forecasting techniques for enhanced renewable energy integration with the grid: a review,” IET Renewable Power Generation, vol. 10, no. 7, pp. 885–989, Aug. 2016.
[6]W.-Y. Chang, “A Literature Review of Wind Forecasting Methods,” Journal of Power and Energy Engineering, vol. 2, no. 4, p. 161, Apr. 2014.
[7]S.-E. Thor and P. Weis-Taylor, “Long-term research and development needs for wind energy for the time frame 2000–2020,” Wind Energy, vol. 5, no. 1, pp. 73–75, 2002.
[8]A. Botterud, Z. Zhou, J. Wang, J. Sumaili, H. Keko, J. Mendes, R. Bessa and V. Miranda, "Demand dispatch and probabilistic wind power forecasting in unit commitment and economic dispatch: A case study of Illinois," IEEE Trans on Sustain. Energy, vol. 4, no. 1, pp. 250-261, 2013.
[9]J. B. Bremnes, "Probabilistic wind power forecasts using local quantile regression." Wind Energy, vol.7, no.1, pp. 47-54, 2004
[10]Bengio Y.“Deep Learning of Representations: Looking Forward,”arXiv preprint arXiv: 1305.0445, 2013.
[11]Hinton G E, Osindero S, and Teh Y-W. “A Fast Learning Algorithm for Deep Belief Nets,”Neural Computation, vol. 18, pp. 1527–1554, 2006.
[12]WAN Jie, CHEN Ning, QIAN Minhui, GUO Yufeng, Hu Qinghua, and YU Daren.“Day-ahead Wind Speed Prediction based on Hybrid Deep Belief Network,”Energy Conservation Technology, 2016.
[13]Y. LeCun, K. Kavukcuoglu, and C. Farabet. “Convolutional Networks and Applications in vision,”In Circuits and Systems (ISCAS), Proceedings of 2010 IEEE International Symposium, pp. 253–256, 2010.
[14]CHEN Xianchang.“Research on Algorithm and Application of Deep Learning Based on Convolutional Neural Network,”Thesis, Zhejiang Gongshang University, 2013.
[15]XU Shanshan.“Research and Application of the convolution neural network,”Thesis, Nanjing Forestry University, 2013.
[16]Erdem E, Shi J., “ARMA based approaches for forecasting the tuple of wind speed and direction,” Appl Energy, pp.1405–1414, 2011.
[17]Kavasseri R G, Seetharasman K., “Day-ahead wind speed forecasting using ARIMA models,” Renewable Energy, pp.1388-1393, 2009.
[18]Hochreiter S, Schmudhuber J., “Long short-term memory,”Neural computation, pp.1735–1780, 1997.
[19]Tian Y, Pan L., “Predicting Short-Term Traffic Flow by Long Short-Term Memory Recurrent Neural Network,” IEEE International Conference on Smart City, pp.153–158, 2015.
[20]C. Wan, Z. Xu, P. Pinson, Z. Y. Dong and K. P. Wong, "Probabilistic Forecasting of Wind Power Generation Using Extreme Learning Machine," in IEEE Trans on Power Syst, vol. 29, no. 3, pp. 1033-1044, 2014.
[21]J. Dowell and P. Pinson, "Very-Short-Term Probabilistic Wind Power Forecasts by Sparse Vector Autoregression," IEEE Trans. on Smart Grid, vol. 7, no. 2, pp. 763-770, 2016.
[22]P. Pinson and G. N. Kariniotakis, "Wind power forecasting using fuzzy neural networks enhanced with on-line prediction risk assessment." IEEE Power Tech Conference Proceedings, Bologna, Italy. 2003 vol. 2, pp. 8-16.
[23]P. Pinson and G. Kariniotakis, "Conditional Prediction Intervals of Wind Power Generation," in IEEE Trans on Power Sys, vol. 25, no. 4, pp. 1845-1856, 2010.
[24]P. Pinson, H.A. Nielsen, J. K. Møller, H. Madsen and G. N. Kariniotakis, "Non-parametric probabilistic forecasts of wind power: required properties and evaluation." Wind Energy, vol. 10, no.6 pp: 497-516, 2007.
[25]Y. Zhang, J. Wang, and X.Wang, "Review on probabilistic forecasting of wind power generation." Renewable and Sustainable Energy Reviews, vol. 32, pp. 255-270, 2014.
[26]A. P. A. da Silva and L. S. Moulin, "Confidence intervals for neural network based short-term load forecasting," in IEEE Trans on Power Sys., vol. 15, no. 4, pp. 1191-1196, Nov 2000.
[27]“https://portal.stpi.narl.org.tw/index/article/10399”
[28“https://www.trademag.org.tw/Upload/tam_tam/737643/%E6%88%91%E5%9C%8B%E9%A2%A8%E5%8A%9B%E7%99%BC%E9%9B%BB%E7%94%A2%E6%A5%AD%E7%99%BC%E5%B1%95%E7%8F%BE%E6%B3%81%E8%88%87%E6%9C%AA%E4%BE%86%E5%B1%95%E6%9C%9BWind%20Energy%20in%20Taiwan.pdf ”
[29]“https://pro.re.org.tw/powerhistory.aspx”
[30]“https://www.taipower.com.tw/tc/page.aspx?mid=204