Prediction of settlement of soft soil subgrade during operation based on GA-BP neural network

To achieve accurate prediction of the settlement of soft soil subgrade of highways, genetic algorithm (GA) was used to optimize back propagation (BP) neural network, and the influence of three kinds of inputs on the accuracy of prediction results was studied. Time t, settlement amount SM5 before 15 d and average settlement rate vt-15 were selected as the influencing factors. Under the three inputs of t, t-St-15, t-St-15vM5, the first 50% and 80% of the measured settlement data of a highway soft soil subgrade during the operation period were taken as the training set, and the remaining original data were taken as the test set, and the average value was taken as the output value after 10 repetitions of training. The coefficient of determination (R2) was used to discriminate the model fit, and the root mean square error (RMSE) and the mean absolute percentage error (MAPE) were used as the evaluation indexes of the model performance. The results show that the R2 was greater than 0.99 for all three inputs. When the proportion of the training set to the original data is 50%, the t-St-15, A5 input has the smallest prediction error with an RMSE of 1. 31 mm and a MAPE of 4. 71%. When the proportion of the training set to the original data is 80%, the prediction error of the t-St-15-vt-15 input is the smallest with an RMSE of 0. 29 mm and a MAPE of 1.00%. The three inputs of t, t-St-15, and t-St-15-vt-15 can all predict the subgrade settlement. The most accurate prediction results are obtained when 80% of the measured settlement data is taken as the training set for the t-St-15-vt-15 input. © 2023 Southeast University. All rights reserved.