過往在進行行人環境風場、室內自然通風評估、風力發電等風環境分析時，常以舒適性或季風為主要考量，蒐集鄰近氣象測站多年之風速風向資料進行統計分析。目前全臺灣的局屬測站共有33 個，涵蓋的範圍並不夠全面，無法確切代表當地的風速及風向資訊。而自動測站目前共有584 個遍佈全台灣，但大部分自動測站因設站時間過短、進行搬遷或儀器故障而缺少資料，進而發生統計資料不足的困境，無法完整地表達當地風環境特性。
近年來類神經網路(Artificial Neural Network；ANN)發展趨近於成熟，可有效解決較複雜的非線性問題，因此本研究將利用類神經網路中具代表性的倒傳遞類神經網路(Back-Propagation Neural Network；BPN)，透過蒐集測站過往氣象資料，調整BPN訓練參數及輸入層參數等，測試測站預測風速結果，並建立BPN模型，對測試區內各個自動測站不足之資料，進行風速及風向之補遺。
本研究發現當輸入層參數考慮風速、風向、溫度、濕度及氣壓，隱藏層轉換函數為線性整流函數(Rectified Linear Unit；ReLU)，並且輸入的測站越多時，可使整體風速預測結果較為準確。當自動測站與測試區域內其他測站風速相關性越低，會使風速預測結果偏差較大。自動測站經過補遺後，可建立各風向發生機率、平均風速、風速標準差及風速累積分布函數，且當補遺資料數增加時，更能突顯當地季風所盛行風向。

Traditionally, when studying wind environment problems such as pedestrian wind environment, indoor natural ventilation assessment and wind power generation, statistical analyses are based on collected monsoon wind speed and direction data from nearby meteorological stations. There are 33 CWB stations in Taiwan as of March 3, 2022, which are not comprehensive enough to represent the local wind speed and direction information. Most of the other 584 automatic stations in Taiwan have been set up for only a short period of time and are frequently relocated or suffer from equipment failure, which leads to the dilemma of insufficient statistical data and cannot completely express the characteristics of the local wind environment.
In recent years, the development of Artificial Neural Network (ANN) has become more mature and can effectively solve more complex non-linear problems. Therefore, this study adopts Back-Propagation Neural Network (BPN) to predict the wind speed and direction data of secleted automatic stations. By collecting past meteorological data from the stations, adjusting the BPN training parameters and input layer parameters, verifying the predicted wind speed and calibrating the BPN model, the missing wind speeds and wind directions of selected automatic stations are supplemented.
It is found that the overall wind speed prediction results are more accurate when the input layer parameters include wind speed, wind direction, temperature, humidity and air pressure, and more stations are considered in the input layer. When the wind speed correlation between the target automatic station and input stations is lower, the wind speed prediction results will be less accurate. After the automatic stations data are supplemented, the mean value, standard deviation and cumulative distribution function of the wind speed along each wind direction can be established. When the number of supplementary data increases, the prevailing wind direction of the local monsoon can be highlighted.

[1] Zuluaga, C. D., et al. “Short-term wind speed prediction based on robust Kalman filtering: An experimental comparison. ” Applied Energy 156: 321-330. , 2015.
[2] Song, J., et al. “A novel combined model based on advanced optimization algorithm for short-term wind speed forecasting. ” Applied Energy 215: 643-658. , 2018.
[3] Khosravi, A., et al. “Energy, exergy and economic analysis of a hybrid renewable energy with hydrogen storage system. ” Energy 148: 1087-1102. , 2018.
[4] Hiester, T. and W. Pennell. Siting technologies for large wind turbine clusters, Battelle Pacific Northwest Labs., Richland, WA (USA). , 1979.
[5] Velázquez, S., et al. “Comparison between ANNs and linear MCP algorithms in the long-term estimation of the cost per kW h produced by a wind turbine at a candidate site: A case study in the Canary Islands. ” Applied Energy 88(11): 3869-3881. ,2010.
[6] Philippopoulos, K. and D. Deligiorgi. “Application of artificial neural networks for the spatial estimation of wind speed in a coastal region with complex topography. ” Renewable energy 38(1): 75-82. ,2012.
[7] Mabel, M. C. and E. Fernandez. “Analysis of wind power generation and prediction using ANN: A case study. ” Renewable energy 33(5): 986-992. ,2008.
[8] Cadenas, E. and W. Rivera. “Short term wind speed forecasting in La Venta, Oaxaca, México, using artificial neural networks. ” Renewable energy 34(1): 274-278. ,2009.
[9] Khosravi, A., et al. “Time-series prediction of wind speed using machine learning algorithms: A case study Osorio wind farm, Brazil. ” Applied Energy 224: 550-566. ,2018.
[10] Salcedo-Sanz, S., et al. “Accurate short-term wind speed prediction by exploiting diversity in input data using banks of artificial neural networks. ” Neurocomputing 72(4-6): 1336-1341. ,2009.
[11] Fadare, D. “The application of artificial neural networks to mapping of wind speed profile for energy application in Nigeria. ” Applied Energy 87(3): 934-942. ,2010.
[12] Li, G. and J. Shi. “On comparing three artificial neural networks for wind speed forecasting. ” Applied Energy 87(7): 2313-2320. ,2010.
[13] Koo, J., et al. “Wind-speed prediction and analysis based on geological and distance variables using an artificial neural network: A case study in South Korea. ” Energy 93: 1296-1302. ,2015.
[14] Filik, Ü. B. and T. Filik. “Wind speed prediction using artificial neural networks based on multiple local measurements in Eskisehir. ” Energy Procedia 107: 264-269. ,2017.
[15] Samadianfard, S., et al. “Wind speed prediction using a hybrid model of the multi-layer perceptron and whale optimization algorithm. ” Energy Reports 6: 1147-1159. ,2020.
[16] Cadenas, E., et al. “Wind speed prediction using a univariate ARIMA model and a multivariate NARX model. ” Energies 9(2): 109. ,2016.
[17] McCulloch, W. S. and W. Pitts. “A logical calculus of the ideas immanent in nervous activity.” The bulletin of mathematical biophysics 5(4): 115-133. ,1943.
[18] Hebb,D. “The organization of behavior; a neuropsychological theory. ” ,1949.
[19] Rosenblatt, F. “The perceptron: a probabilistic model for information storage and organization in the brain. ” Psychological review 65(6): 386. ,1958.
[20] Minsky, M. and S. Papert. “Perceptrons.” ,1969.
[21] Hopfield, J.J. “Neurons with graded response have collective computational properties like those of two-state neurons. ” Proceedings of the national academy of sciences 81(10): 3088-3092. ,1984.
[22] Rumelhart, D. E., et al. “Parallel distributed processing”, IEEE New York. ,1988.
[23] Beale, M. H., et al. “Neural network toolbox user’s guide. ”,2020.
[24] Holland, S. M. “Principal components analysis (PCA). ” Department of Geology, University of Georgia, Athens, GA: 30602-32501. ,2008.
[25] Goodfellow, I., et al. “Deep learning”, MIT press. ,2016.
[26] Hennessey Jr, J. P. “Some aspects of wind power statistics. ” Journal of Applied Meteorology and Climatology 16(2): 119-128. ,1977.
[27] Rehman, S. and T. O. Halawani. “Statistical characteristics of wind in Saudi Arabia. ” Renewable energy 4(8): 949-956. , 1994.
[28] MATLAB. MATLAB documentation R2020a. The Mathworks Inc., 2020.
[29] 朱佳仁，風工程概論，科技圖書，2006年。
[30] 陳若華、吳國昌、陳海曙、蔡明樹，建築配置與自然通風評估模式之研究，內政部建築研究所，2001年。
[31] 郭建源，集合住宅外部環境風場與室內自然通風互制效應分析研究，內政部建築研究所，2017年。
[32] 王安強、林子平、梁文宇、郭建源、曾淑翎、陳育成、侯凱山、藍士棠，跨不同地況區域之風廊建置分析及都市通風環境評估，內政部建築研究所，2018年。
[33] 中央氣象局，測站代號及站況資料查詢，資料來源： https://e-service.cwb.gov.tw/wdps/obs/state.htm，2022年。
[34] 葉怡成，類神經網路模式應用與實作，儒林圖書，2009年。
[35] 中國文化大學大氣科學系，大氣水文研究資料庫，資料來源： https://dbar.pccu.edu.tw/，2018年。
[36] 中央氣象局，氣象資料開放平台，資料來源： https://opendata.cwb.gov.tw/index，2017年。
[37] 中央氣象局，颱風資料庫，資料來源：https://rdc28.cwb.gov.tw/TDB/public/warning_typhoon_list/，2022年。
[38] 內政部營建署，建築物耐風設計規範及解說，2016年。