Abstract: Common genetic algorithm had some disadvantages, such as premature convergence, greater randomness and slower convergence speed. A new improved adaptive genetic algorithm was proposed based on the Sigmoid function, and it effectively improved the convergence speed of the algorithm and avoided falling into local optimal solution. The feasibility and effectiveness of the method were proved by an example of a cantilever truss. This new genetic algorithm could provide theoretical support for its popularization and application in the structural optimization design of large civil engineering.