The tire flexible ring model can accurately express tire deformation, but the stiffness parameters of the model cannot be directly measured, so identifying the stiffness parameters of the model becomes the key in the modeling process. Based on the kinematic equation of the tire flexible ring model, the relationship between the natural frequency and stiffness parameters of agricultural tires was analyzed, and a method for identifying the stiffness parameters of the flexible ring model was proposed based on particle swarm optimization (PSO) algorithm. Based on the kinematics equation of the flexible ring model tire, the relationship between the natural frequency and the stiffness parameters of the agricultural tire was analyzed, and a method for identifying the stiffness parameter of agricultural tire flexible ring model based on PSO algorithm was proposed. A tire testing platform was built, the natural frequency was obtained through tire modal testing, and PSO algorithm was used to identify the stiffness parameters of the flexible ring model. Using the average error between the experimental and predicted values of the natural frequency as the evaluation index, the identification results of PSO algorithm were compared with traditional methods and genetic algorithm(GA). The results showed that PSO algorithm had the highest accuracy, with an average absolute error of 1.67Hz and an average relative error of 1.66%. Compared with GA, the average relative error was decreased by 16.16% and the computation time was decreased by 93.19%. The correctness and accuracy of the stiffness parameter identification method was proved based on PSO algorithm. The grounding angle of agricultural tires was obtained through the contact patch test, and the vertical force on the tires was estimated based on the identified stiffness parameters. The experimental and predicted values of vertical force were compared, and the results showed that the parameter identification results obtained by the particle swarm algorithm had the highest accuracy. The average relative error of vertical load estimation was 1.97%, which was reduced by 12.05% compared with the genetic algorithm.