Study on Combined Weather Numerical Model with Machine Learning on Big Data Computing Platform for Real-Time Typhoon Wind Velocity Prediction

瀏覽統計 Email 通知 RSS Feed

簡歷

基本資料

Project title

Code/計畫編號

MOST106-2111-M019-001

Translated Name/計畫中文名

結合氣候數值模式與機器學習法於大數據運算平台即時預測颱風風速之研究

Project Coordinator/計畫主持人

Chih-Chiang Wei

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Marine Environmental Informatics

Website

https://www.grb.gov.tw/search/planDetail?id=12226756

Year

2017

Start date/計畫起

01-08-2017

Expected Completion/計畫迄

31-07-2018

Bugetid/研究經費

760千元

ResearchField/研究領域

大氣科學

"本計畫擬以機器學習法結合數值天氣預測模式並建構大數據平行分散式運算平台研發一套颱風時期即時預測地表或離岸風速風向預測模式整合系統。本計畫所提出的數值天氣預測模式（擬採用 Weather Research and Forecasting, WRF）結合機器學習法為一種混合數值模式與統計模式的概念，可用以結合統計和物理模式兩者之間的特性，便於解決短時距（如1 小時或更短）即時預測地表或離岸風速風向問題。一般而言，數值模式優點為所有參數都是由大氣動力學、數值方法計算得出的，其預測結果可由物理關係解釋，因此數值模式可用以求解高度非線性的氣象問題。在數值模式中，同化資料和客觀分析可提供動力模式合理的初始條件。然而，數值模式模擬過程常須耗費大量計算時間（如數小時至數日，視運算處理器效率和演算尺度而定），在模擬短時距預測時，可能無法提供即時性；相對的，統計模式在計算效率則較數值模式為高，其因是統計模式乃由過去長期觀測資料建立預測變數與其他屬性間之統計關係，利用合適的統計方法可快速地建立預測模式。然而，對數值模式而言，統計模式之物理基礎較為薄弱，如兩參數統計關係很強但其物理上可能無因果關係可解釋，或是物理上顯而易見的兩參數關係亦可能未反應在統計分析上，因此統計模式有其侷限性。因此若能綜合兩者之優點或可提供一預測時的高時效性和準確性，因此衍生了本計畫之構想。由於建置整合運算平台繁雜故擬以兩年為期完成所提計畫。本計畫擬以海洋大學所在之臺灣東北部（或其他地區）為研究區域。本計畫所發展的系統可用於在颱風期間地表或離岸任意位置和高度之風速風向預測功能，以便防災上之使用，預測長度為1 至12 小時（或更長，將視模式誤差評估後而定）。本計畫發展模式過程中，首先利用WRF 數值模式模擬臺灣東北部區域的環流分布及各網格點上風速風向。WRF 數值模式為新一代中尺度數值天氣預測系統，其發展研究團隊包括美國國家大氣研究中心（National Center for Atmospheric Research, NCAR）、美國海洋暨大氣總署(National Oceanic and Atmospheric Administration, NOAA)等。WRF 可作為數值預測以及大氣研究之用途，可用於真實天氣個案模擬或者是應用其理想化模組作為基本物理過程探討的理論根據。模組化主要過程，首先，本計畫利用WRF 數值模式模擬多場颱風之環流分布，以便得到各網格點上風速風向數值解，這些網格點亦包含了地面氣象站、海面浮標站和設計的任意點（簡化上，本計畫擬將各測站和任意點所在位置之數值解以最鄰近的網格點代替或以內插方式處理，以進行後續建模工作）。其次，本計畫以機器學習法之常用演算法（如回歸支援向量機、決策樹、隨機森林等）建構風速風向統計預測模式。為了得到任意點之預測值，首先將利用有觀測資料測站（地面氣象站和海面浮標站）建立統計模式，以便評估統計解與觀測值的誤差，同時並比較WRF 數值解與統計解之優劣；經由驗證統計模式後，本計畫即可進行任意點之建模工作（即WRF 模式結合機器學習法），建模過程中因任意點並無風速風向觀測數據，因此風速風向值將以高精度解析下之數值解（視為真值）替代；計畫主持人於2015 年架設氣象自動觀測站於海洋大學校區臨海建物頂樓上，因此可將其規劃為任意點之一，作為驗證機器學習法結合數值模式之用。最後，本計畫採大數據最新技術Apache Hadoop Spark 2.0，以便進行運算平台和線上預測分析，Hadoop Spark 將可達成1)同時接收來自各地遠端伺服器(Data Nodes)的資料來源，2)在主機端(Master Node)建構叢集運算框架和線上預測模式以便即時預測任意點風速風向。本計畫預期將完成上述整合系統後，同時綜合討論各模式（即數值模式、統計模式與兩者結合模式）預測能力的優缺點及適用性，以便提供地表或離岸風速風向即時資訊之參考。""A two-year project is proposed here. A useful scheme for wind speed and direction forecasts during typhoons is highly desirable for Taiwan, because the powerful winds accompanying these severe typhoons drastically affect, for examples, the under-construction structures and wind turbines of the wind farm. A weather numerical model combined with machine learning algorithms is proposed for forecasting the wind speed and direction during typhoons. The developed combination model can be used to determine the wind speed and direction on a short time step (1 h or less) of a forecasting operation. We will employ the mostly used support vector machines for regressions, decision trees, and random forests to build the machine learning-based prediction model. The WRF (Weather Research and Forecasting) numerical model will be employed as the numerical simulation based model. The experimental area of Northeast Taiwan will be used for simulation. Regarding a complex typhoon system, the collected data comprise the typhoon tracks, FNL (Final) Operational Global Analysis Data for WRF model, the typhoon characteristics, the buoy atmospheric properties, and ground weather data. This study includes 40 typhoon events and more over the past 10 years (2007–present). Four criteria, including the relative mean absolute error, coefficient of variation of the root mean squared error, correlation coefficient, and efficiency coefficient, will be used to highlight the model cases capable of identifying the optimal performance level. This study will compare the WRF numerical model, the machine learning models, and the WRF combined with machine learning model, and observations. We also propose a Hadoop Spark Big Data platform applied to the real-time wind speed and direction predictions. There are various tools which can be used in Big Data management from data acquisition to data analysis. Hadoop brings the ability to cheaply process large amounts of data, regardless of its structure. Hadoop is made up of two core projects: Hadoop Distributed File System (HDFS) and MapReduce. Hadoop splits files into large blocks and distributes them across nodes in a cluster. To process data, Hadoop transfers packaged code for nodes to process in parallel based on the data that needs to be processed. In addition, Apache Spark provides programmers with an application programming interface centered on a data structure called the resilient distributed dataset, a read only multiset of data items distributed over a cluster of machines, that is maintained in a fault-tolerant way. It was developed in response to limitations in the MapReduce cluster computing paradigm, which forces a particular linear dataflow structure on distributed programs: MapReduce programs read input data from disk, map a function across the data, reduce the results of the map, and store reduction results on disk. Finally, several typhoons will be used as simulation typhoons for the real-time forecasts by using the Hadoop Spark Big Data platform. This study will demonstrate the feasibility of the proposed methodology and Big Data platform, and discuss the various model cases on the basis of increasing the accuracy and efficiency of the wind speed and direction predictions."

Keyword(s)

颱風
風速
數值模式
機器學習
巨量資料
Typhoon
Wind speed
Numerical model
Machine learning
Big Data

DSpace CRIS

Study on Combined Weather Numerical Model with Machine Learning on Big Data Computing Platform for Real-Time Typhoon Wind Velocity Prediction

基本資料

Description