Abstract: Adverse pressure gradient and flow separation prediction are crucial in industries such as aircraft design, engine design, and mechanical engineering. However, widely used RANS models in the engineering community, such as the SA turbulence model, often perform poorly in predicting adverse pressure gradients. Although the SST turbulence model introduces the Bradshaw assumption, considering a certain degree of adverse pressure gradient, its results are often unsatisfactory under strong adverse pressure gradients. This paper starts from the construction mechanism and control equations of the SST turbulence model, combined with the wall friction decomposition formula, to analyze the reasons for the model's inaccurate prediction of separation under adverse pressure gradients. It points out that the inaccuracy is related to the Bradshaw assumption, specifically manifested in the assumption's forced balance between the production and dissipation terms of turbulent kinetic energy, leading to restrictions on the generation of Reynolds stress and turbulent kinetic energy, and a reduced ability of the flow to resist separation. Finally, by adjusting the structural parameter a_1 based on the ratio of the generation and dissipation terms of turbulent kinetic energy in local regions, an improved SST turbulence model is proposed. The improved model is verified using cases such as Gaussian bumps, blowing/suction plates, and two-dimensional humps. The results show that, for flow separation under adverse pressure gradients, the improved model has better predictive capabilities for separation and reattachment points compared to the original SST model, with varying degrees of improvement in the calculation of Reynolds stress and turbulent kinetic energy.