Abstract: As the proportion of new energy integration continues to increase, the highly random fluctuations of wind power pose challenges to traditional unit commitment methods, leading to frequent unit start-ups, shutdowns, and complex and variable operating modes. Particularly under "severe scenarios" with a significant deviation between the actual output and the predicted value of wind power, finding a feasible unit commitment scheme may be impossible, drastically increasing the system operation risk. Therefore, incorporating the probabilistic characteristics of wind power forecast error into unit commitment has become an urgent issue. This paper proposes a two-stage robust unit commitment model and its optimization method that accounts for the operational risk associated with the tail of the wind power output probability distribution. Firstly, through numerical weather prediction data and spectral clustering methods, the conditional probability distribution of the forecast error of a single wind farm's output is accurately characterized, and the regional wind power output probability distribution to obtain the output confidence interval is constructed. Then, the tail probabilistic characteristics of wind power forecast error are integrated into the unit commitment model to optimize system risk margins, thereby reducing operational risk under "severe scenarios" and enhancing system safety and economic efficiency. Finally, tests with actual data from a certain region In Jiangsu, the system load shedding is only 14% or even lower than that of traditional methods, and the system risk margin increases by 30% to 50%, verifying the effectiveness and practicality of the model.