Accurate simulation of separated flows is crucial in aircraft design and optimization, yet remains a significant challenge in Computational Fluid Dynamics (CFD). As higher level of Reynolds-averaged Navier-Stokes (RANS) modeling, Reynolds-stress models (RSMs) tend to get better recirculation predictions than eddy viscosity models (EVMs) in streamwise separated flows. Nevertheless, RSMs suffer from the problems such as irregular streamlines near the reattachment point and a large number of equations, etc. In this paper, we propose a simplified Reynolds-stress model (also called Semi-Differential Reynolds-Stress Model, Semi-RSM), which is based on the kinematic equilibrium hypothesis and symbolic regression training. The simplification is obtained by preserving the three transport equations of Reynolds normal stress and a \(\lambda \) -scale equation originating from SSG/LRR- \(\omega \) RSM, and solving the three components of the shear stress tensor through algebraic relations. The Semi-RSM model was verified firstly by training the redistribution parameter through periodic hill, and then conducted generalizability analysis in typical iced airfoils flows. The numerical results indicate that the Semi-RSM model significantly reduces the computational cost, eliminating the unphysical back bending of the streamline while maintaining computational accuracy. Deeper modifications and validation of the model will continue later.