應用長短期記憶遞歸神經網路建立精準的離岸風能預測

Application of long short-term memory recurrent neural networks for accurate offshore wind energy prediction

張嘉媛(Chia-Yuan Chang)

指導教授 ： 張建成

共同指導教授 ： 郭志禹(Chih-Yu Kuo)

國立臺灣大學/工學院/應用力學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201902359

摘要

近年來，台灣提倡能再生能源，其中離岸風能為重點發展項目之一。台灣海峽為良好的風場，因此吸引國內外風能公司來台灣西部海岸進行建設及投資，在政府推行離岸風能的政策之下，2015年三家國內離岸風電示範業者分別於苗栗及彰化外海設置三座離岸觀測塔提供風能監測服務。
風場與能源預測對於風能產業以及公部門的電力調度非常重要，觀測塔的即時監測數據可以提升附近風場預測的精準度。在本研究中，我們證明此系統可以透過結合中尺度天氣預報（WRF, the Weather research and forecast model）的模擬資料、氣象觀測塔的觀測資料，以及機器學習模型來實現。觀測塔的資料來源為永傳能源公司2015年9月於福海風場設立的氣象觀測塔所提供的數據。利用長短期記憶遞歸神經網路（LSTM, Long short-term memory recurrent neural networks）模型以觀測塔數據為目標，修正WRF模擬資料來建立人工智慧（AI, Artificial Intelligence）模型，此模型可以用來提供精準的風能預測。
利用皮爾森相關係數（Pearson correlation coefficient），以及均方根誤差（RMSE, the root-mean-square error）做預測準確度評估，用來評估WRF和機器學習的預測結果。預測結果為三天，評估結果顯示，在東北季風時期（10月到隔年4月）的預測精準度可以大幅的提升，因此，在冬季時期可以提供良好的風能預測。未來，機器學習模型可以進一步擴展，提供離岸風能產業準確的風能預測。

關鍵字

離岸風能 ； 氣象觀測塔 ； 中尺度天氣預報 ； 長短期記憶遞歸神經網路 ； 福海風場 ； 東北季風 ； 皮爾森相關係數 ； 均方根誤差

並列摘要

Taiwan has been promoting renewable energy vigorously in recent years, of which offshore wind energy is one of the most important energy resources. The Taiwan Strait is an outstanding wind field, so it attracts wind energy operators both from domestic and abroad to invest, construct and operate wind farms. Under the government's policy of the offshore wind energy, three domestic offshore wind power corporations were appointed and each of them constructed one offshore meteorological masts. They are separately in the offshore areas of Miaoli and Changhua in 2015, serving weather condition monitoring purposes.
Wind field and energy forecast is extremely essential for wind energy industry and power dispatching in public sectors. The real-time data of the wind mast may be incorporated to improve the forecast accuracy for the wind farms in the vicinity area. In this study, we demonstrate that such system can be achieved by combining numerical weather forecast model data (WRF model, the Weather research and forecast model), the offshore wind mast observation data, and machine learning facilities. The project is incorporation with Taiwan Generations Corporation so that the wind mast data are taken from the wind mast of the company. The mast was built in the Fuhai wind farm in September 2015. The LSTM (Long short-term memory recurrent neural networks) machine learning model is applied to train an Artificial Intelligence (AI) model from the WRF forecast to the observation. The AI model is then used to provide accurate wind forecast.
The forecast accuracy is assessed by using the Pearson correlation coefficient and RMSE (the root-mean-square error). They are applied to evaluate the forecasts of both WRF and machine learning model. The forecasts are three-day forecasts. The assessments show that the forecast accuracy can be greatly improved in the northeast monsoon period (October to April). It is expected that excellent wind energy prediction can be achieved in the winter monsoon periods. The machine learning model in future can be further extended to provide accurate wind energy production forecasts prediction for the offshore wind energy industry.

並列關鍵字

Offshore wind energy ； meteorological mast ； WRF model ； long short-term memory recurrent neural networks ； Fuhai wind farm ； Northeast monsoon ； Pearson correlation coefficient ； the root-mean-square error

參考文獻

[1]S. S. Soman, H. Zareipour, O. Malik, and P. Mandal, "A review of wind power and wind speed forecasting methods with different time horizons," in North American Power Symposium 2010, 2010: IEEE, pp. 1-8.

[2]A. Kusiak, Z. Zhang, and A. Verma, "Prediction, operations, and condition monitoring in wind energy," Energy, vol. 60, pp. 1-12, 2013.

[3]M. Lei, L. Shiyan, J. Chuanwen, L. Hongling, and Z. Yan, "A review on the forecasting of wind speed and generated power," Renewable and Sustainable Energy Reviews, vol. 13, no. 4, pp. 915-920, 2009.

[4]L. Li, M. Wang, F. Zhu, and C. Wang, "Wind power forecasting based on time series and neural network," in Proceedings. The 2009 International Symposium on Computer Science and Computational Technology (ISCSCI 2009), 2009: Academy Publisher, p. 293.

[5]J. P. d. S. Catalão, H. M. I. Pousinho, and V. M. F. Mendes, "An artificial neural network approach for short-term wind power forecasting in Portugal," in 2009 Fifteenth International Conference on Intelligent System Applications to Power Systems, 2009: IEEE, pp. 1-5.