The Research on Integrating Heterogeneous Data for Using the Technique of Machine Learning in Flood Prediction

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

MOST109-2221-E019-056

Translated Name/計畫中文名

整合異質性資料以使用機器學習技術於淹水預測之研究

Project Coordinator/計畫主持人

Ya-Hui Chang

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Computer Science and Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=13542120

Year

2020

Start date/計畫起

01-08-2020

Expected Completion/計畫迄

31-07-2021

Bugetid/研究經費

649千元

ResearchField/研究領域

資訊科學--軟體

隨著全球暖化的影響，降雨往往超過預期使洪水帶來可怕的災難，所以淹水預測模式的建立是一個重要的課題。傳統的水利演算乃建構於數學模式之上，需要複雜的運算且費時甚久，無法達到即時預警的目的。近年來，已經陸續有學者試圖應用機器學習的技術於該領域中，但是該方法需要合適的訓練資料集。本計畫的主要研究目的，是整合來自不同來源的資料，產生合理且充分的資料，並利用適當的機器學習技術，建立淹水預測模組，以達到即時預警的目的。針對資料來源部分，我們除了利用水理模式所產生的模擬資料、政府相關單位的開放資料之外，也預計採用新聞網站的資料，以補充修正模擬資料和開放資料的缺失。我們提出利用自然語言處理的技術，從新聞報導中擷取出所需的水深等資料，然後再與其他類別的資料進行整合。另一方面，我們也提出數種神經網路架構，特別設計成適於抓取時序資料和地理資料的特性，以進行水位或淹水與否的預測。最後，我們將利用RMSD、Accuracy、F1 score等公式，透過真實的水位資料，來評估所提出之各種做法是否可行。 Due to global warming, the heavy rain sometimes brings flooding and causes terrible disasters. Therefore, it is important to build a flood forecasting system. Traditional hydraulic modules are built upon mathematical models, which require complicated computations and take a lot of time, and therefore cannot provide instant warning. Recently, researchers have begun to study how to apply the techniques of machine learning to this field, but it requires reasonable sizes of training data. In this project, we focus on integrating different sources of data to produce proper training datasets, so that we can apply the machine learning technique to build the flood forecasting system. We plan to adopt three types of data: numerical data, open data, and news. The technique of natural language processing will be utilized to retrieve the required water depth values from the news, which are further integrated with other types of data. Besides, we also propose several neural network architectures, which are designed to capture the characteristics of temporal and spatial data, to perform the prediction of the water depth and flooding situation. Finally, based on several functions, including RMSD, Accuracy, and F1 score, we will evaluate the performance of our proposed approach.

Keyword(s)

機器學習
異質性資料來源
時空資料
machine learning
heterogeneous data source
spatial-temporal data

DSpace CRIS

The Research on Integrating Heterogeneous Data for Using the Technique of Machine Learning in Flood Prediction

Details

Description