The prediction of AQI in the future by using machine learning algorithm is helpful to analyze the trend of air quality change in the future from a macro perspective. When a single machine learning model is traditionally used to predict air quality, it is difficult to obtain good prediction results under different AQI fluctuation trends. In order to effectively solve this problem, the prediction method is improved. When using random forest model and long and short-term memory model based on convolution neural network and attention mechanism to predict the AQI data in Chengdu, a difference fusion analysis model is designed according to the characteristics of different prediction accuracy under different AQI fluctuation trends. The experimental results show that the MSE of the proposed difference fusion analysis model is 5.8% lower than that of the random forest model, and 6.3% lower than that of the long-term and short-term memory model based on convolutional neural network and attention mechanism.