In order to explore the force mechanism of the anchorage zone of the cable-stayed bridge, a local three-dimensional finite element model of the steel anchor box in the anchorage zone of the cable-stayed bridge was established. The adaptive genetic algorithm optimizes the thickness of the steel anchor box plate. The results show that the stress concentration effect at the intersection of the steel anchor box M1 plate and other plates is more obvious, showing a more typical out-of-plane bending state. The objective function with the steel anchor box stress as the constraint condition is established, and the plate M1~M2 , The thickness of M5~M6 has increased to different degrees, and the thickness of M3~M4 plate has a slight decrease. After optimization, the stress and the equivalent peak stress of each plate along the cable force direction of the stayed cable have decreased significantly, and the decrease is about 11% and 10% respectively; the two stress indicators of the plate M1 have increased slightly, and the increase is within 5%. At the same time, the stress concentration phenomenon of the steel anchor box plate is alleviated, and the genetic algorithm based on adaptive variables has achieved better results in the optimization of the steel anchor box, with higher accuracy and strong robustness.