Evolving support vector regression based on improved grey wolf optimization for predicting settlement during construction of high-filled roadbed

Abstract

The reconstruction and expansion of highways have become the dominant focus of current infrastructure development. However, this process often results in significant differential settlement between the new and old roadbed, leading to various pavement issues. Machine learning methods undoubtedly provide a new approach for predicting roadbed settlement. However, most of the current research on predicting roadbed settlement focuses on post-construction settlement, and there is relatively little involvement in predicting settlement during roadbed filling. Therefore, it is necessary to accurately predict the settlement during the roadbed construction period. This study establishes an improved settlement prediction model for roadbed construction using a combination of differential evolution (DE), grey wolf optimization (GWO), and support vector regression (SVR). The model incorporates cumulative filling time, cumulative filling height, and daily rainfall as input parameters, while the corresponding cumulative settlement serves as the output. To evaluate the effectiveness of the proposed DE-GWO-SVR model, the training set consists of two weeks' data of sample data, while the latter week's data is used as the test set to predict settlement. Meanwhile, the predictive performance is assessed using various evaluation metrics. The coefficients of determination (R2) for the settlement prediction results were all higher than 0.95. The findings indicate that the settlement predictions made by the model align well with the measured values. Moreover, the conventional SVR and artificial neural network (ANN) models for predicting roadbed settlement have been established. The R2 for the conventional SVR model is 0.7793, while that for the DE-GWO-SVR model is 0.9977. The results demonstrate that the optimized DE-GWO-SVR model outperforms the conventional approaches across all evaluation criteria. This affirms the high prediction accuracy and strong generalization ability of the established DE-GWO-SVR settlement prediction model. Therefore, the DE-GWO-SVR settlement prediction model can realize high-precision prediction of settlement during roadbed construction. This is conducive to the timely adjustment of the filling scheme during the roadbed construction to ensure the safety and stability of the high-filled roadbed.