Accurate simulation of typhoon-induced waves is essential for marine hazard forecasting, yet numerical wave models remain limited under extreme wind conditions due to uncertainties in empirically calibrated parameters. In addition, conventional tuning approaches are inefficient for coordinated multi-parameter optimization. This study develops a multi-objective optimization framework for empirical parameter calibration in numerical wave models. Using the WAVEWATCH III model as a testbed, five key parameters influencing offshore and nearshore wave simulations are optimized for typhoon conditions in waters adjacent to Taiwan Island. Latin Hypercube Sampling is used to generate parameter combinations, and batch simulations are evaluated against buoy observations using root mean square error and bias. An adaptive regression model is constructed to map parameter space to error metrics, and the Non-dominated Sorting Genetic Algorithm III (NSGA-III) is applied to identify optimal parameter combinations. Validation with independent typhoon events shows that the optimized configuration effectively improves significant wave height simulations, reducing both RMSE and bias relative to the default scheme. The proposed framework provides an efficient and transferable approach for improving wave model performance under extreme wind conditions.