It is one of the effective methods improving the ambient air quality to establish air quality forecast models to know the possible air pollution process in advance and take corresponding control measures. The secondary modeling of air quality forecast was explored. By determining the definition and calculation method of air quality index (AQI) and primary pollutants, and according to the degree of influence on pollutant concentration, five meteorological conditions, namely temperature, humidity, air pressure, wind speed and wind direction, were classified by a k-means clustering algorithm in line with AQI. According to the measured data and primary forecast data of three monitoring points (A, B and C), a secondary forecast model was developed using BP neural network based on the genetic algorithm. The results show that the weather favorable for pollutant diffusion had the conditions of high wind speed and high air pressure, while the weather unfavorable for pollutant diffusion had the conditions of low wind speed and low air pressure. The secondary forecast model has good prediction accuracy, and can simultaneously predict the concentrations of multiple monitoring points, many days in the future and various pollutants, thus better forecasting the air quality. The built secondary forecast model can improve the accuracy of air quality forecast, and has high economic and ecological value.