Abstract: With the continue increase in the penetration rate of wind power, traditional day-ahead dispatch models struggle to meet the requirements for guaranteeing both consumption and power supply of the power system. Effectively characterizing the uncertainty of wind power through wind power scenario sets is becoming increasingly important. Aiming at the difficulty of existing statistical methods in balancing both the volatility and temporal characteristics of wind power, we propose a method for generating day-ahead wind power output scenarios that considers both statistical and temporal characteristics. First, a prediction interval partitioning method is constructed based on Bayesian optimization. Kernel density estimation is used to obtain the probability distribution of prediction errors across different predicted output levels, while the temporal characteristics of prediction errors are modeled using Markov chains. Secondly, considering the historical volatility characteristics of wind power, the state transition matrix of prediction errors is dynamically updated. A rolling sampling model is designed by integrating the prediction error distribution, which updates the sampling probability interval at each step and generates scenarios by sampling from the prediction error distribution. Finally, the proposed method is tested with actual wind power data. The simulation result demonstrates that the method is capable of generating accurate and time-correlated wind power output scenarios.