The Research on Deep Foundation Pit Support Optimization Based on GA-BP Neural Network

This study uses a GA-BP optimized three-layer BP neural network to predict surface settlement and horizontal displacement in deep excavation projects, focusing on the anchor angle and embedment depth of vertical support structures. Results show that the optimal anchor angle range is 15° to 25°, and deeper embedment reduces settlement but has diminishing effects on lateral displacement. The GA-BP network outperforms the BP network, reducing the average relative error by 77.76% (from 1.17% to 0.26%) and closely matching numerical simulations. The study highlights the importance of parameters like learning rate, error tolerance, and GA settings in optimizing performance. The GA-BP model is effective for predicting deformation and can be applied to other nonlinear engineering problems.