|
Abstract
|
Culverts are essential hydraulic structures for managing water flow and runoff, yet poor designs often lead to
erosion and structural instability due to inaccurate scour depth predictions. Conventional empirical methods for
estimating scour depth at culvert outlets suffer from limited generalizability, high sensitivity to parameters, and
computational inefficiencies, hindering robust culvert design. To overcome these limitations, this study introduces a novel hybrid approach that integrates the Adaptive Neuro-Fuzzy Inference System (ANFIS) with
advanced metaheuristic optimization algorithms, including Colliding Bodies Optimization (CBO), Enhanced CBO
(ECBO), and Vibrating Particles System (VPS). Leveraging a comprehensive dataset of 249 experimental cases,
the ANFIS-VPS hybrid model achieves superior predictive performance, with a root mean square error (RMSE) of
0.49, a mean absolute percentage error (MAPE) of 55.6 %, and a coefficient of determination (R²) of 0.96. This
outperforms ANFIS-CBO, ANFIS-ECBO, and conventional empirical models. Notably, ANFIS-VPS reduces RMSE
by 46 % compared to standard ANFIS by harnessing VPS’s efficient global search capabilities. Sensitivity analysis
reveals that the particle densimetric Froude number (Frd) exerts the greatest influence, with its exclusion causing
a 246 % RMSE increase. Overall, the proposed framework provides a reliable, accurate tool for scour depth
prediction, enhancing the resilience and durability of hydraulic infrastructure in transportation networks.
|