With the continuous development of new power systems based on new energy, large-scale and intensive wind power, photovoltaic, and other new energy access to the system has laid a solid foundation for the realization of the “carbon peaking and carbon neutrality” goals, but at the same time, it also leads to the increasing challenges faced by the dispatching operation of new power systems under extreme climates, and the most prominent problem is that the probability of wind power ramp events has increased significantly. Wind power ramp events will not only cause great fluctuations in the frequency of the system but also affect the balance of electric power and energy, threatening the safe and stable operation of the system. Through the statistical analysis of wind power ramp events, a predictive method of wind power ramp events based on a deep auto-regressive (DeepAR) model is proposed. Firstly, combined with the relationship between wind power and wind speed, the impact of wind power ramp events on power grid dispatching operations under extreme climates is analyzed. Secondly, a physical model of wind power ramp events is established to analyze the statistical characteristics of wind power when wind power ramp events occur. Then, the DeepAR model is used to perform the power prediction of wind power ramp events, and the wind power output curve under extreme climates is analyzed. Finally, combined with the measured data of the wind power field, the effectiveness of the proposed method is verified. The verification shows that the proposed method can accurately predict the occurrence probability of wind power ramp events under extreme climates in advance, which is expected to greatly improve the uncertainty faced by the dispatching operation of new power systems in the future.