Multi-feature PM2.5 Prediction with ARIMA-LSTM

One of the most harmful environmental risk to human health is air pollution. Effective prediction of pollutants in the air is conducive to air improvement and pollution control. Air quality data has the characteristics of instability and nonlinear, to more accurate prediction of pollutants. The ARIMA-LSTM model is applied to forecast PM2.5 content in a continuous-time in us paper. The data provided is time-series data, so we use LSTM for processing. The traditional prediction method has many limitations. It can not take all aspects into account in the multi-feature aspect, and the accuracy is not so accurate. We use the ARIMA model to deal with linear and nonlinear problems in the LSTM model. LSTM is an optimal model for predicting time series, but the data used have multiple characteristic values, which has no significant effect on seasonal prediction and trend prediction. The result shows our prediction model performs better than other models in our paper.