Application of GA-BP Neural Network Optimized by Grey Verhulst Model around Settlement Prediction of Foundation Pit

Abstract

Due to the limitation in the prediction of the foundation pit settlement, this paper proposed a new methodology which takes advantage of the grey Verhulst model and a genetic algorithm. In the previous study, excavation times are often the only factor to predict the settlement, which is mainly because the correspondence between real-time excavation depth and the excavation time is hard to determine. To solve this issue, the supporting times are precisely recorded and the excavation depth rate can be obtained through the excavation time length and excavation depth between two adjacent supports. After the correspondence between real-time excavation depth and the excavation time is obtained, the internal friction angle, cohesion, bulk density, Poisson’s ratio, void ratio, water level changes, permeability coefficient, number of supports, and excavation depth, which can influence the settlement, are taken to be considered in this study. For the application of the methodology, the settlement monitoring point of D4, which is near the bridge pier of the highway, is studied in this paper. The predicted values of the BP neural network, GA-BP neural network, BP neural network optimized by the grey Verhulst model, and GA-BP neural network optimized by the grey Verhulst model are detailed compared with the measured values. And the evaluation indexes of RMSE, MAE, MSE, MAPE, and $R^2$ are calculated for these models. The results show that the grey Verhulst model can greatly improve the consistency between predicted values and measured values, while the accuracy and resolution is still low. The genetic algorithm (GA) can greatly improve the accuracy of the predicted values, while the GA-BP neural network shows low reflection to the fluctuation of measured values. The GA-BP neural network optimized by the grey Verhulst model, which has taken the advantages of GA and the grey Verhulst model, has extremely high accuracy and well consistency with the measured values.