Langmuir circulations and turbulence (LT) are crucial in the upper ocean mixed layer, significantly affecting the air-sea exchange of momentum, heat, and mass. The development of an appropriate LT parameterization scheme is vital for ocean modeling. This study employed the Large-eddy Simulation (LES) and the Physics-informed Neural Network (PINN) to optimize the KC04 Langmuir turbulence scheme by dynamically adjusting E6 as a key parameter determined by winds and waves. The LES simulations under different wind wave states indicated the PINN-inferred values for E6. Modelling results from GOTM in OCSPapa station demonstrated that the optimized scheme outperformed the original KC04 scheme in simulating the vertical eddy diffusivity and temperature, with an ∼6.24% annual reduction in the root mean square error (RMSE) for the temperature and an ∼8.23% reduction in the RMSE during autumn. Furthermore, the optimized scheme resulted in a thicker mixed layer, reaching 4.9 m. This enhanced LT parameterization scheme exhibited the improved robustness for variable spatiotemporal resolutions, significantly improving the modeling accuracy.