With the rapid development of urbanization and industrialization, the problem of air pollution has become increas-ingly prominent, and air quality prediction is particularly important. Some representative studies currently monitor and forecast air quality in real time. For example, ZHOU Guangqiang et al. Used numerical prediction to analyze air quality in eastern China. However, experimental results show that this method is difficult to predict and is very important. SANKAR et al. Used multiple linear regression to predict air quality, but the experimental results showed that the linear model had low prediction accuracy and slow efficiency;PREZ et al. Used statistical methods to predict air quality, and the experimental results proved the prediction accuracy of the statistical method relatively low; WANG et al. Used an improved BP neural network to establish a prediction model for the air quality index, and their experiments verified that the BP neural network has a slow convergence rate and is prone to fall into the local optimal solution problem; YANG et al. Air quality concentration effect, a PM2.5 concentration prediction model based on random forests was established, and the empirical process proved that the meshing program weakened the quality and efficiency of subsequent air quality analysis; these methods are difficult to model from a time perspective, and the prediction accuracy is low is a more important issue. Because low prediction accuracy may lead to large errors in air quality prediction results. 河北科技大学学报 2020年 第1期 张冬雯,等:基于长短期记忆神经网络模型的空气质量预测 In this paper, a neural network model based on long -term memory (LSTM) is proposed to solve the problem of low prediction accuracy in air quality research.MAPE, RMSE, R, IA and MAE were used to test the predictive performance of LSTM neural network and the comparison model.Since Delhi and Houston are cities with high levels of air pollution, the experimental data sets used in this paper were from the air quality data of Punjabi Bagh monitoring station in Delhi from 2014 to 2016 and the air quality data of Harris County monitoring station in Houston from 2010 to 2016.By comparing LSTM neural network with multiple linear regression and regression model (SVR), the experimental results show that LSTM neural network is suitable for time series prediction with multiple variables or multiple inputs LSTM neural network has the advantages of high prediction accuracy, high speed and strong robustness.