Deep excavation deformation prediction method based on BP Neural Network with integrated attention mechanism
ZHANG Mingju
QIN Shengwang
LI Pengfei
GE Chenhe
YANG Meng
XIE Zhitian
Abstract：To address the issues of poor generalization and susceptibility to local optima when using a single Back Propagation(BP)neural network for predicting excavation-induced deformations,this study employs Genetic Algorithms(GA)and Particle Swarm Optimization(PSO)for optimization and integrates an Attention mechanism to construct hybrid GA-Attention-BP and PSO-Attention-BP neu-ral network models.The Nanjing Twin Towers excavation project is used as a case study,with PLAXIS 2D simulating the deformation characteristics of the retaining structure and ground surface un-der 680 different conditions.Additionally,20 sets of field monitoring data from foundation pits in the Nanjing area are included in the dataset.The prediction results of different neural networks are com-pared with actual monitoring data under evaluation metrics including Mean Squared Error(MSE),Mean Absolute Error(MAE),and the coefficient of determination(R²).The results demonstrate that the GA-Attention-BP and PSO-Attention-BP models achieve MSE values of 3.47 and 3.22,MAE values of 1.59 and 1.47,and R² values of 0.93 and 0.96,respectively,indicating significant perfor-mance improvements over the standard BP and Attention-BP neural networks.Furthermore,the attention-based weight allocation results indicate that excavation depth and diaphragm wall width have the most substantial influence on retaining structure deformation,with weight coefficients reaching 1.33 and 1.17,respectively.
Keywords：deep excavation engineeringnumerical simulationattention mechanismback propagationgenetic algorithmparticle swarm optimization
Publication Date：2025-04-30
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：10( 95-104 )