Prediction and analysis of deformation of deep foundation pit herringbone retaining support(HRS) based on machine learning

With the development of utilization of underground space, a large number of large-scale and deep foundation pit projects have emerged, and there is an urgent need to develop new supporting facilities to meet the needs of foundation pit engineering. The herringbone supporting facility is one of the economical and effective new supporting facilities for deep foundation pits. Based on 100 sets of three-dimensional finite difference(FLAC3D) simulation results and canonical correlation analysis, the correlation between input parameters and output parameters is analyzed. Deformation prediction analysis models of four types of machine learning algorithms (the BP neural network, the particle swarm optimized support vector machine, the decision regression tree and the integrated algorithm (random Forest)) for herringbone support facilities are established. The study found that based on the results of canonical correlation analysis, the influencing factors on the horizontal displacement of the pile top in descending order are cohesion, internal friction angle, soil density, shear modulus, and the angle between the inclined pile and the side wall of the foundation pit, bulk modulus and length of inclined pile; The influencing factors on the settlement around the foundation pit in descending order are internal friction angle, soil density, length of the inclined pile, cohesion, shear modulus, bulk modulus, inclined pile, the angle between the inclined pile and the side wall of the foundation pit. The HRS-ANN model can accurately predict the deformation of the herringbone support facilities and the settlement around the foundation pit and has higher prediction accuracy and stability than other machine learning algorithms. Furthermore, the prediction of horizontal displacement of pile top by HRS-ANN model is better than that of settlement around foundation pit. The prediction and analysis of the deformation of deep foundation pit herringbone support based on neural network can provide theoretical analysis methods for the promotion and application of support facilities and engineering design. © 2020, Science Press. All right reserved.