Abstract: High-precision deformation prediction during deep excavation is critical for ensuring engineering safety. To achieve accurate predictions of excavation-induced diaphragm wall deformation and ground settlement, this study develops an intelligent forward-prediction framework based on optimized back-analysis techniques. The framework establishes a "monitoring-inversion-prediction" rolling closed-loop system to enable dynamic forecasting. Specifically, monitoring data from completed excavation stages are used to identify soil constitutive parameters via optimization algorithms; these optimized parameters are then applied to predict deformations in subsequent unexcavated stages. This process is iterated until project completion. Key technical features include: (1) an inverse analysis method enhanced by differential evolution, where the objective function minimizes the deviation between monitoring data and numerical simulations; (2) stepwise excavation simulation using the finite element software OPTUM G2; and (3) the application of the Hardening Mohr-Coulomb (HMC) model with small-strain stiffness to accurately characterize soft soil behavior under unloading conditions. The method was validated using two typical staged excavation case studies. The results demonstrate strong agreement between predicted and measured data, with accuracy significantly surpassing conventional methods. This approach provides an effective solution for safety control in deep excavations and serves as a valuable decision-making tool for early risk warning