In this work, the experiment was carried out through the pilot test platform for vertical mill, and the change rule of the classification performance of the classifier under different system air volumes and classifier speeds were compared and analyzed. The cumulative distribution function of the normal distribution was used to correlate the operating parameters of the classifier with the collection probability of particles of different particle sizes. The test data of the system air volume and the speed change condition of the powder separator were used to fit the mean and variance in the model. A mathematical model can be established to quantitatively analyze the relationship between the system air volume, the speed of the separator and the particle classification efficiency. The fitted standard deviation of the verified regression was RMSE=0.0046, the deviation between the predicted value and the true value was small, and the coefficient of determination R-square=0.9863, which was close to 1, the model had higher credibility. Under the conditions of known system air volume and separator speed, the model prediction curve of the classification efficiency of the separator basically coincided with the test curve, and the model prediction effect was good.