Development of a Convolutional Neural Networks Model for Estimating Extreme Rainfall during a Structure under Construction

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

MOST109-2622-M019-001-CC3

Translated Name/計畫中文名

卷積神經網路模式於營建施工時期極端降雨量推估之研發

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=13443282

Year

2020

Start date/計畫起

01-06-2020

Expected Completion/計畫迄

31-05-2021

Bugetid/研究經費

545千元

ResearchField/研究領域

大氣科學

本計畫將研發一套營建施工時期極端降雨推估技術和評估是否停工之方法。此方法將可使營建業者於施工期間遭遇極端降水事件（如颱風）時，事先估計工程施工地點之降雨資訊，再根據降雨資訊進一步評估因極端氣候因素而可停工日數。本案預期可協助營建廠商估算免計工期並向主管機關申請延長施工之日數，達到工程如期竣工且免於罰款之目標。本計畫以極端降水之颱風事件為系統建置對象。本案降雨量推估方法採用人工智慧深度學習法，模式輸入屬性為中央氣象局雷達回波影像。深度學習法之卷積神經網路(Convolutional neural networks)是一種可有效擷取影像特徵再加以取樣分析，近年被用來進行影像辨識的熱門方法。根據營建施工相關法規，常見的氣候因素「大雨」為停工之可能條件。中央氣象局所定義的大雨為「二十四小時累積雨量達五十毫米以上，且其中至少有一小時雨量達十五毫米以上之降雨現象」。因此，本計畫初步設計為在施工地點及其鄰近區域內推估即時雨量和未來24小時累積雨量推估值（若契約規範有不同的停工氣候條件，亦可修正降雨標準）。本計畫將研發「區域極端雨量及營建停工評估系統」以滿足營建業者之需求。此系統概念化可分為兩個模式（共包含五個模組），一為極端氣候降雨量預估模式（模組一、二、三），另一為營建停工評估模式（模組四、五）。首先，極端氣候降雨量預估模式包含：模組一「定點時雨量推估模型（SRE1）」可用來推估施工地點之小時降雨量；模組二「鄰近區域平均時雨量推估模型（ARE1）」為將施工地點鄰近之地面氣象測站（含氣象局測站和自動測站）時雨量平均化後再推估小時雨量；模組三「未來24小時累積雨量區間推估模型（ICPE24）」為用來推估未來24小時累積雨量。另外，營建停工評估模式包含：模組四「雨量歷線圖示化模型」為用來繪製時雨量歷線圖和24小時累積雨量歷線圖；模組五「停工評估模型」為計算推估誤差量和是否停工之決策建議。本計畫系統開發完成後預期可給予施工單位更多的氣候資訊，以便施工單位據以預判是否因極端氣候因素而須停工並估算其可申請延長的施工日數，以利營建廠商重新規劃施工排程。本計畫將以實際案例進行分析和探討，使學術理論能進一步活用於實務面上，開拓解決營建工程界之新技能。The construction industry is statistically one of the most hazardous industries. Numerous structures fail during construction because calculations are based on the requirements of the complete structure, and sufficient provisions are not made to enable structures to withstand loads before completion. Extreme weather such as typhoons can generate violent rains and cause serious engineering delay on incomplete structures. Hence, this plan on typhoon rainfall is of great significance in the field of construction industry and operation management. For construction industry, timely and reliable information on current and future rain information are thus vital for enabling forecasters to make accurate and timely forecasts and to be operated under construction appropriately. Because construction often delayed works during their construction stage under typhoons, a useful scheme for rain forecasts during typhoon periods is highly desired. This study will develop the convolutional neural networks (CNNs) model for estimating extreme rainfall during a structure under construction. The CNNs are currently among the most commonly applied to analyzing visual imagery. One of the biggest advantages of CNNs over other image classification algorithms is that, they are relatively independent from pre-processing, or prior knowledge provided by humans. This means that the network learns the filters that in traditional algorithms were hand-engineered.The purpose of this study is to develop a “Regional Extreme Precipitation and Construction Suspension Estimation System” for construction industry when a structure is under construction stage. There are two major models are involved in the system, one is the regional extreme precipitation estimation model (comprising modules 1, 2, and 3), the other is the construction suspension estimation model (modules 4 and 5). For module 1, the specific-point hourly rainfall estimation model (shortly as SRE1) is employed for estimating the hourly rainfall estimation at the construction spot. For module 2, the adjacent-region hourly rainfall estimation model (ARE1) is used for estimating the hourly rainfall estimation nearby the construction spot. Module 3 (ICPE24) is used for estimating the intervalized cumulative precipitation within 24 hours. Module 4 is the designed to plot the hyetograph using the results from modules 1, 2, and 3. Then, module 5 is then determine whether the construction suspension or not according to these plots from module 4. By finishing the integrated system, we can provide the suggestion whether suitable to work or not and assess the days of construction suspension for the builders.

Keyword(s)

工程
施工
降雨推估
模式
深度學習
颱風
Engineering
Construction
Rainfall estimation
Modeling
Deep learning
Typhoon

DSpace CRIS

Development of a Convolutional Neural Networks Model for Estimating Extreme Rainfall during a Structure under Construction

Details

Description