Study on the application of deep learning model for estimation of activity duration in railway construction project

Abstract

Currently, as there are no prescribed regulations, construction activity duration for civil engineering project including railways is calculated by subjective judgment based on previous construction records or the experience of the construction manager. Since the method of estimating the activity duration can vary depending on the capabilities of the construction manager, it is difficult to appropriately calculate the construction duration. In this study, the authors propose a methodology for predicting the net construction duration for each activity by applying a deep learning algorithm that discovers patterns in large amounts of construction history already executed, self-learns, and provides comprehensive judgment and prediction for the calculation of net construction duration. For this, a deep learning application model that can be used for construction duration estimation was developed, and practical applicability was analyzed by comparing predicted and actual duration. Through this, it is considered possible to derive an objective construction duration more accurately than possible using existing estimation methods, and it is thought that utilization of deep learning in the construction management process will also increase. ? 2020 The Korean Society for Railway. All rights reserved.