Estimation of scour depth at bridges with complex pier foundations using support vector regression integrated with feature selection

This study aims at establishing machine learning models based on the support vector regression (SVR) for estimating local scour around complex piers under steady clear-water condition. A data set consisting of scour depth measurement cases has been collected to construct the prediction models. The data set includes eight influencing factors that consider aspects of pier geometry, flow property, and river bed material. Moreover, to enhance the performance of the SVR model, filter and wrapper feature selection strategies are used. The research finding is that all feature selection approaches can help to improve the prediction accuracy compared with the SVR model that uses all available features. Notably, the feature selection method based on the variable neighborhood search (VNS) algorithm achieves the best performance (MAPE = 21.65%, R2 = 0.85). Accordingly, the prediction model produced by SVR and VNS can be useful for assisting decision makers in the task of structural health monitoring as well as the design phase of bridges.

Language

  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01678002
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Jul 11 2018 8:36AM