Real-time Extreme Rainfall Evaluation System for the Construction Industry Using Deep Convolutional Neural Networks

Abstract

For the construction industry, timely and reliable information on current and future rain information is vital for enabling forecasters to make accurate and timely forecasts and to allow appropriate construction operations. Because construction works are often delayed during typhoons, a useful scheme for rain forecasts during typhoon periods is highly desirable. This study developed a regional extreme precipitation and construction suspension estimation system (REPCSES) for the construction industry to use when a structure is in the construction stage. The REPCSES has two major functions: a regional extreme precipitation estimation model (comprising Modules 1 and 2) and the construction suspension estimation model (Modules 3 and 4). Module 1 is a regional 1-h-ahead rainfall estimation model, which is used for estimating the hourly rainfall near the construction location. Module 2 is used for estimating the cumulative rainfall within 24 h. Module 3 is designed to plot a hyetograph using the results from Modules 1 and 2. Then, Module 4 determines whether the construction should be suspended according to the plots from Module 3. In addition, this study developed a deep convolutional neural network model for estimating extreme rainfall during a structure under construction, and the experimental area was Nantou County, Taiwan. The collected typhoons (i.e., Soulik, Trami, Kong-Rey, Matmo, Dujuan, and Nesat) affecting the study area occurred from 2013 to 2017. The results indicated that the integrated system could provide accurate estimations of whether work could proceed as well as the number of days that construction should be suspended for.