Estimation of local scour at bridge piers using genetic programming to enhance predictive accuracy

Valladares-Sandoval E.D.; Herrera-Riega L.F.; López-Silva M. · Tecnologia y Ciencias del Agua (2026) · DOI: 10.24850/j-tyca-2026-03-01

Resumen

Local scour is a hydraulic phenomenon resulting from the interaction between the riverbed and a structure that is altering the natural flow and eroding the bed material where piers are located. Therefore, determining the correct depth at which the piers are to be set is crucial for the structures' longevity. However, there is no general formula for its calculation today. This study utilized Gene Expression Programming (GEP), which allows for the derivation of equations to predict local scour around bridge piers. For this purpose, 919 laboratory data and 746 field data were gathered. Based on this data set, predictive models for local scour were established in three case scenarios: laboratory, field, and combined. Model selection was done via evaluation measures as well as validation using external research data. The laboratory scenario with GEP demonstrated better robustness, efficiency, and result stability, giving a coefficient of determination (R2) of 0.899, root mean square error (RMSE) of 0.054, and mean absolute error (MAE) of 0.027. At the validation stage, however, field scenario model provided successful results, with the average error ranging from 6.0 to 14.0 % compared to analytical equations. It is therefore concluded that genetic programming based artificial intelligence improved the prediction of local scour near bridge piers compared to analytical models.

Instituciones: Universidad Ricardo Palma; Universidad Católica Sedes Sapientiae