Smog pollution is becoming a significant problem for people worldwide, becoming an essential threat to the global environment. Many studies on haze already exist, which still need to continue in-depth research to better deal with haze problems. Due to its unique geographical environment, Sichuan has become one of the areas with severe smog pollution. Therefore, the research and prediction of smog pollution in Sichuan has become an urgent need. This paper proposes a deep learning technology based on a Deep Belief-Back Propagation neural network. It makes in-depth prediction research by using the air pollution data of PM2.5, PM10, O₃, CO NO_2, and SO₂ in Sichuan smog to provide a decision-making basis for predicting and preventing smog polluted weather. According to the prediction results of the model, the concentrations of PM2.5 and PM10 in Chengdu were predicted. The analysis shows that the larger the number of hidden layers in the belief network, the higher the prediction accuracy. Under the same network, the prediction accuracy of PM2.5 is significantly higher than that of PM10. Compared with the traditional Back Propagation neural network, the prediction effect of the Deep Belief-Back Propagation neural network is better.

雾霾污染正在成为世界范围内人们面临的重大问题，成为对全球环境的重要威胁。许多关于雾霾的研究已经存在，还需要继续深入研究以更好地应对雾霾问题。由于得天独厚的地理环境，四川已成为雾霾污染严重的地区之一。因此，四川雾霾污染的研究与预测已成为当务之急。本文提出了一种基于 Deep Belief-Back Propagation 神经网络的深度学习技术。_{利用PM2.5、PM10、O 3}、CO NO _2、 SO ₂的空气污染数据进行深度预测研究为四川雾霾天气的预报和防治提供决策依据。根据模型的预测结果，预测了成都市PM2.5和PM10的浓度。分析表明，置信网络中隐藏层数越多，预测精度越高。相同网络下，PM2.5的预测精度明显高于PM10。与传统的反向传播神经网络相比，Deep Belief-Back Propagation 神经网络的预测效果更好。