中華民國內政部建築研究所中華民國內政部建築研究所

一、研究緣起
以往在進行風環境分析、室內自然通風評估等作業時，需要蒐集鄰近氣象站10-20年之風速風向資料進行計分析，找出各風向之機率函數，以作為模擬的參考參數。截至110年3月9日統計結果，目前氣象局全臺灣地面氣象站有33個，涵蓋的的範圍並不夠全面，且部分為新設。在選用測站資料時可能會面臨距離測試區過遠，無法真切代表當地的風速風向資訊。近年氣象局廣設自動氣象站，目前有436個，遍佈全台灣。雖可彌補部分地區資料不足的缺憾，但因設站時間未久，且儀器可能因當機而缺乏資料，進而發生統計資料不足的窘境。
此外，目前開發或評估案大多位於都市區域，過往評估者常以距基地最近之測站，統計風速資料作為參數，但建築基地往往距測站有一段距離，且受到地形以及地貌(如建築、植披)之影響，使得建築基地風場特性和氣象站會有所不同。如何合理地將氣象站風速資訊轉換至建築基地之風速，亦是近年風工程界關注之議題。常被提出的解決方案為利用模擬方式(風洞實驗或CFD)，將測站風速與工址風速進行關聯，以考量地貌所造成的影響。
二、研究方法及過程
本計畫主要工作項目為資料收集與理論分析，接下來分為3個分支進行工作，包含氣象資料蒐集與分析、類神經網路預測模式開發、示範區風速地圖建立，說明如後：
一、氣象資料蒐集與分析：
(1)示範區風速資料蒐集；
(2)資料整理與分類；
(3)建立風速機率統計分析工具與風花圖製作。
二、類神經網路預測模式開發
(1)資料整理與分類；
(2)預測模式訓練；
(3)預測模式驗證；
(4)自動測站風速資料預測補充。
三、示範區風速地圖建立：
(1)示範區建築模型資料蒐集；
(2)示範區CFD模擬；
(3)示範區風洞實驗；
(4)CFD模擬驗證；
(5)風速地圖資料與風速機率轉換模式建立；
最後整合工作有工具整合和相關轉換模式方法撰寫、期末報告、成果報告等。執行期間另舉辦專家諮詢座談會，以徵詢學者專家對本案之建議。
三、結論
本研究綜合研究成果，並檢視預期成果效益後，可歸納以下幾點結論：
(1)進行類神經預測時，平均風速可以掌握得很好，而在極值風速預測上略嫌不足，經適當選取地形變化以及正規化後的資料進行訓練，可提升極值風速的預測結果，且對於整體預測風速與風向角的結果依然良好。經預處理後的資料進行分類測試後，對於整體預測的風速與風向角無太大影響。2010年後各自動測站經過補遺後的資料數約可達到八成。
(2)將預測資料整合後利用韋伯分布回歸出其係數，與實測結果之誤差值低於0.1，顯見補遺資料之統計分布結果具有代表性。
(3)經過類神經補遺後，整理研究區內各氣象站近10年資料，統計資料包含各風向平均風速、發生機率、韋伯函數係數等資料。
(4)由各測站四季白天與晚上之溫度年總平均與年標準差發現，各測站之溫度年總平均與年標準差差異不大，應可採用短期資料進行統計即可代表。
(5)應用內政部國土測繪中心「多維度國家空間資訊服務平臺」提供之圖資，建立相應之轉換技術，可快速建立行政區中建築與地形3D模型並進行CFD模擬。
(6)因建築模型以及山區地形模型整合介面處理問題，目前CFD模擬結果未納入東南側山區地形特性，故對於模擬驗證結果造成影響。另採用穩態的realizable k-ε model對於預測結果造成一定程度的誤差。但整體看來，CFD模擬結果可以良好呈現該地區在受到地形地貌影響後的風速剖面特性，可作為後續建立風速地圖之工具。
(7)選定新北市板橋行政區進行數值模擬，建立風速地圖資料。並可藉由建立之風速地圖以及板橋氣象站之關係，在考量地形地貌之影響下轉換得到二測定點之各風向平均風速及機率，此應用技術提供之資訊可作為風環境或室內通風評估之前導資訊。
四、建議
建議一
進行都會區內自動測站風速補遺：立即可行建議
主辦單位：社團法人中華民國風工程學會
協辦單位：內政部建築研究所、國立臺灣科技大學
經測試後，在風速資料經過適當處理後，以前饋式類神經網路進行氣象站風速資料訓練與補遺，所得之風速以及風向與實測值具有高相關性以及可接受之誤差範圍。建議可進一步將本方法做推廣應用，將資料記錄不足之自動測站或局屬測站進行補遺，首先可針對較有應用需求之都會區測站來進行。此外，因氣象站資料為逐時累積，故氣象站訓練資料應逐年更新，以完善補遺資料庫之精確度。
建議二
提供風速地圖查找應用：中長期建議
主辦單位：社團法人中華民國風工程學會
協辦單位：內政部建築研究所、國立臺灣科技大學
建議未來研究應開發自動地圖查找轉換技術，結合GIS圖資搜尋技術，提供更完整且友善之查找環境。另都會區建築或地貌會因開發、都市更新或其他因素改變，故CFD建立之風速地圖應逐年更新，或可依使用對象需求即時提供相關服務。

Motivations
In either the wind environment analysis or indoor natural ventilation assessment, collection and analysis of 10-20 years’ wind speed data of nearby weather stations are necessary for evaluating the wind speed probability distribution in each direction. As of 110/3/9, the Central Weather Bureau has 33 ground-based weather stations; their coverage is not so comprehensive and some of them are new. The wind speed information of the test area may not be well represented by the data of the station since it may be far away from the test area. In recent years, the Central Weather Bureau has set up 436 automatic weather stations all over Taiwan, but the amount of data is still insufficient due to shorter collection time and instrumental malfunction.
In addition, most of the current development or evaluation cases are located in urban areas and engineers often adopt the wind statistics from the nearest station. However, it is unsuitable since the construction site is often not close to the station and the impact of the terrain and landscape is ignored. How to reasonably convert wind speed information from meteorological stations to wind speeds at construction sites has also been a topic of concern to the wind engineering community. The often proposed solution is to use simulations (wind tunnel tests or CFDs) to correlate station wind speeds with site wind speeds.
Research Methods and Progress
The main work items of this project are data collection and theoretical analysis, followed by three branches of work, including meteorological data collection and analysis, the development of neural network prediction model and the establishment of the wind speed map of a demonstration area.
(1)meteorological data collection and analysis:
(i)demonstration area wind speed data collection;
(ii)data collation and classification;
(iii)establishment of statistical analysis tool and wind rose map production.
(2)the development of neural network prediction model:
(i)data collation and classification;
(ii)mode training;
(iii)model validation;
(iv)supplement of automatic station wind speed data
(3)establishment of the demonstration area wind speed map:
(i)building model data collection;
(ii)CFD simulation;
(iii)wind tunnel test
(iv)verification of CFD simulation;
(v)conversion of wind speed map data and wind speed probability
Finally, the tools are integrated and the final report is prepared. Additional expert consultation seminars are held to seek the advice.
Conclusions
The following conclusions can be summarized:
(1)When making neural network model prediction, the mean wind speed can be predicted very well while the extreme wind speed prediction is slightly inadequate. The prediction results of extreme wind speed can be enhanced by appropriate selection of terrain changes and normalization of data. After adding the model predictions, the available data for automatic stations after 2010 can reach about 80%. 
(2)After the integration of the prediction data, the errors of the resulting Weibull distribution parameters compared with those from the field measured data are less than 0.1, which shows that the statistical results of the supplementary data are representative. 
(3)After adding the predictions from neural network model for the past ten years, the mean wind speed and Weibull distribution parameters for each direction are obtained.
(4)It is found that, for any station, the fluctuation of the mean and standard deviation of annual temperature is very small such that short-term data will be sufficient.
(5)Applying the resources provided by the "Multi-Dimensional National Spatial Information Service Platform" of the Land Mapping Center of the Ministry of the Interior, this study establishes the corresponding conversion technology to quickly set up 3D models of buildings and terrains for CFD simulations.
(6)Due to the difficulty of integrating building model and mountain topographical model, the mountain terrain characteristics on the southeast side are not included in the current CFD simulation. Steady-state realizable k-ε model also causes some error in the prediction results. On the whole, however, the CFD simulation results can well show the wind speed profile characteristics of the area after being affected by the topography, and can be used as a tool for the subsequent establishment of wind speed maps.
(7)The wind speed map data is established for Banqiao District of New Taipei City based on numerical simulations. In addition, the mean wind speeds and probability distributions of two selected locations can be obtained by combining the wind speed map data and their relationship with the Banqiao weather station.
Recommendation
(1)Immediately feasible proposal: to carry out automatic station wind speed supplement in the metropolitan area 
Organizer: Architecture and Building Research Institute, Ministry of the Interior; National Taiwan University of Science and Technology
This study has shown that the supplementary wind speeds, predicted by the proposed neural network model with properly processed recorded data, are acceptable. It is suggested that the proposed method can be further applied to stations in the metropolitan area. In addition, the meteorological station data should be updated regularly to improve the accuracy of the database.
(2)Medium and long-term recommendations: to provide wind speed map applications
Organizer: Architecture and Building Research Institute, Ministry of the Interior; National Taiwan University of Science and Technology; Wind Engineering Society
It is suggested that an automatic map search system should be developed to provide a more complete and friendly search environment. The wind speed map established by CFD should be updated regularly to fit the needs of users.