Settlement prediction of a high embankment based on non-linear regression and neural network algorithm

Abstract

This study reports in-situ measured data from single-point settlement meters buried in layers at the road shoulders and the center of a high embankment respectively. Based on the measured data, the nonlinear fitting models and neural network algorithm were proposed to establish the method of settlement prediction, and the mean absolute percentage error (MAPE) and mean square errors (MSE) were used to evaluate the accuracy of the settlement prediction model. The results show that the MAPE value of the exponential curve prediction model is less than 10%, and the MAPE value of the hyperbolic model is between 10% and 20%, and the MAPE value of the logarithmic model is greater than 20%, which shows that the exponential curve model of the three nonlinear fitting models has the highest accuracy. After establishing the back propagation (BP) neural network model, the settlement data of the four monitoring points were learned and trained, and the prediction accuracy of the two BP neural network prediction models and the exponential curve prediction model were compared. The model fitting coefficient of R² of BP neural network were both greater than 0.99, and the MSE and MAPE were less than 1%. In addition, the multi-step rolling BP neural network prediction model has the highest prediction accuracy, followed by the BP neural network prediction model based on influencing factors while the exponential curve prediction model has the worst performance and weak practicability. This research can provide new inspiration for embankment settlement prediction and give technical support to monitor the disaster of high embankment.