<p>Friction is a key nonlinear factor in pneumatic servo systems, and its accurate modeling and parameter identification directly affect control performance. The conventional LuGre model fails to describe smooth static-to-dynamic transitions, especially during low-speed startup. It also involves strong parameter coupling, which often causes traditional optimization algorithms to converge to local optima. To address these issues, an improved LuGre model and an intelligent identification method are proposed. A transition function is introduced to capture continuous friction behavior, and Lyapunov theory is used to prove model stability. Experiments show that the improved model increases friction peak prediction accuracy by 43–45%. For parameter identification, a hybrid evolutionary algorithm is developed by combining chaotic mapping and Gaussian convolution. The chaotic mapping enhances population diversity, while the Gaussian convolution improves local search capability. This dynamic combination balances global exploration and local exploitation. In experimental validation, the proposed algorithm maintains relative parameter identification errors below 2%, showing better convergence speed and accuracy than conventional methods.</p>