Concrete dam deformation prediction model for health monitoring based on extreme learning machine

• Published in: Structural control and health monitoring Vol. 24; no. 10; pp. e1997 - n/a
• Main Authors: Kang, Fei, Liu, Jia, Li, Junjie, Li, Shouju
• Format: Journal Article
• Language: English
• Published: Pavia John Wiley & Sons, Inc 01.10.2017
• Subjects: Artificial neural networks; Back propagation; Back propagation networks; Concrete dams; dam health monitoring; Dams; Detention dams; extreme learning machine; Feedforward control; Gravity dams; Learning algorithms; Mathematical models; Monitoring; multiple linear regression; Neural networks; Nodes; Prediction models; Regression analysis; Seepage; Space shuttle; Statistical models; stepwise regression
• DOI: 10.1002/stc.1997
• ISSN: 1545-2255

Summary Structural health monitoring via quantities that can reflect behaviors of concrete dams, like horizontal and vertical displacements, rotations, stresses and strains, seepage, and so forth, is an important method to evaluate operational states of concrete dams correctly and predict the future structural behaviors accurately. Traditionally, statistical model is widely applied in practical engineering for structural health monitoring. In this paper, an extreme learning machine (ELM)‐based health monitoring model is proposed for displacement prediction of gravity dams. ELM is one type of feedforward neural networks with a single layer of hidden nodes, where the weights connecting inputs to hidden nodes are randomly assigned. The model can produce good generalization performance and learns faster than networks trained using the back propagation algorithm. The advantages such as easy operating, high prediction accuracy, and fast training speed of the ELM health monitoring model are verified by monitoring data of a real concrete dam. Results are also compared with that of the back propagation neural networks, multiple linear regression, and stepwise regression models for dam health monitoring.