A Hybrid GMDH Neural Network-Genetic Algorithm Approach for Predicting Shear Wave Velocity Profiles

Shear wave velocity (Vs) is a fundamental parameter in geotechnical earthquake engineering and seismic site characterization; however, direct in-situ measurement remains expensive and spatially limited. This study proposes an interpretable hybrid group method of data handling-genetic algorithm (GMDH-GA) framework for predicting shear wave velocity profiles in heterogeneous alluvial deposits. A comprehensive geotechnical-geophysical database comprising ۱۹۹۱ records from ۲۲۶ boreholes in Mashhad, Iran, was compiled using downhole measurements and standard penetration test (SPT) data. Depth (Z), SPT blow count (N), and upper-layer shear wave velocity (Vsu) were selected as predictor variables. Separate soil-specific models were developed for clay, silt, sand, gravel, and combined soil datasets using a ۷۰/۳۰ training-testing strategy. The proposed models generated explicit polynomial equations while maintaining high predictive capability. Statistical evaluation demonstrated strong agreement between measured and predicted Vs values, with testing-stage coefficients of determination reaching ۰.۹۸-۰.۹۹ and low prediction errors across all soil groups. Comparative analyses showed that the proposed GMDH-GA models substantially outperform widely used empirical correlations in terms of accuracy and generalization capability. Sensitivity analysis indicated that Vsu is the dominant predictor, followed by SPT resistance and depth. The results demonstrate that the proposed framework provides a reliable and interpretable alternative for estimating shear wave velocity profiles in regions where direct geophysical measurements are unavailable.

کلیدواژه ها:

Genetic Algorithm ، Group method of data handling ، Mashhad City ، Shear wave velocity ، Standard penetration test