Evaluating the use of neural networks and genetic algorithms for prediction of subgrade resilient modulus

TLDR

The study evaluates artificial neural networks and genetic algorithms for predicting subgrade resilient modulus from soil index properties and examines how prediction accuracy influences mechanistic empirical pavement design guide performance predictions. The authors trained ANN models on soil index data, using genetic algorithms to select input variables, thereby improving prediction accuracy of subgrade resilient modulus. ANN models, particularly those optimized with genetic algorithms, achieved superior subgrade modulus predictions compared to regression models, and the resulting improved modulus estimates led to more accurate pavement performance predictions in the MEPDG analyses.

Abstract

This paper investigates the use of artificial neural networks (ANNs) and genetic algorithms to improve the accuracy of the prediction of subgrade resilient modulus (M r) based on soil index properties. Furthermore, it also examines the effect of the accuracy of the M r estimation on the mechanistic empirical pavement design guide (MEPDG) performance prediction. The results of this paper showed that the ANN models had much better prediction of the M r coefficients of subgrade soils than that of the regression models. In addition, the use of the genetic algorithms in the selection of the input variables of the ANN models enhanced the accuracy of the prediction of those models. The results of the MEPDG analyses indicated that the prediction model used to estimate the subgrade M r input value can have a significant effect on the predicted performance of pavements. Furthermore, those results showed that the use of ANN models yielded much more accurate pavement performance prediction than using regression models; in particular when genetic algorithms were used in developing those models.