Improving the accuracy of current wind power prediction is very difficult and uneconomic， but power system dispatching and wind power control require accurate power curve. A novel method for classified treatment of wind farms active power using forecasting results is presented， which can reduce the influence of prediction error on the decision. The new method is based on predictive power. Considering the influence of predictive error distribution， a classified mode based on the reliability of predictive power is founded using chance constrained programming method， and predictive power is classified into three types: base load output， suboptimal output and high-frequency output. With the prerequisite of ensuring maximum use of wind power output， the classified treatment can distinguish different reliability components， supply decision basis for power dispatching and wind power control， and reduce the decision risk effectively. The policy iteration and particle swarm algorithm are used to solve the model， and the classified treatment is simulated by 24 h data of actual wind farms. The results verify the feasibility and effectiveness of the method.