Forecasting the output power of large wind farms is helpful for planning reserve capacity in a mixed-generation power supply system. After classifying the reserves needed in a power system, this research focuses on the reserve components that are affected by wind–power fluctuations, the regulating, and load-following reserves. Considering the uncertainties in system load forecasts and wind–power forecasts, a general model for calculating the reserve requirements of a power system is proposed. Then, ultra-short-term forecasts of wind–power output 1 h into the future were studied, and a practical method that uses 10-min data to forecast wind output with a rolling average was proposed to predict the amount of power generation in the next hour. Then, using the RTS-96 dataset from the IEEE as an example, the reserve capacity requirement for the power system is determined according to the forecast uncertainty. Accurate predictive models for wind forecasting can reduce the needed reserve capacity of the system while guaranteeing the reliability and safety of the grid.