Abstract

Rapid seismic damage evaluation of subway stations is critical for the efficient decision on the repair methods to damaged subway stations caused by earthquakes and rapid recovery of subway networks without much delay. However, the current methods to evaluate the damage state of a subway station after earthquakes are mainly field investigation by manual or computer vision, which is dangerous and time-consuming. Given this, a novel methodology that adopts machine learning techniques as the classification model to rapidly and accurately evaluate the post-earthquake damage state of subway stations is proposed. Four machine learning techniques including artificial neural networks (ANNs), support vector machine (SVM), random forest (RF), and logistic regression (LR) are adopted. The interrelated intensity measures of ground motions (IMs) and their uncorrelated principal components (PCs) are, respectively, taken as the input to find the most suitable classification model as well as to investigate how the correlation among IMs affects the performance of these models. The results show that the LR taking IMs as inputs provides the best performance as it has the highest accuracy (87.7%) as well as stable performance. Additionally, taking PCs as input can improve the performance of RF, while for ANN, SVM, and LR, taking PCs as input will reduce their prediction performance. The research conclusions can provide a reference for the selection of the machine learning technique and its inputs when establishing a rapid assessment model for the post-earthquake damage state of subway stations.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call