Study on PM2.5 DNN-LSTM hybrid neural network prediction model considering climate factors

Abstract

Aiming at the problem of the single data source in PM2.5 prediction, a PM2.5 DNN-LSTM hybrid neural network prediction model that takes into account climate factors is proposed. First, the DNNnetwork is used to abstract the characteristics of climate and seasonal factors and climate factors as an additional part of the prediction process. Input and analyze in collaboration with LSTM network. Experiments with pollution data and weather data (sampling interval of one hour) are collected from monitoring sites in Beijing from 2010 to 2014, and comparing the DNN-LSTM model with other prediction models, the results show that this model is compared to LSTM. The RMSE of the model is reduced by 10.71%, which is 5.52% lower than the RMSE of the multi-source data fusion LSTM model. Research shows that the multi-source data fusion DNN-LSTM model proposed in this paper has better predictive ability. Compared with the LSTM model, the RMSE of this model is reduced by 10.71%, compared with the multi-source data fusion LSTM model, the RMSE is reduced by 5.52%, compared with the LSTM model, the MAE is reduced by 21.55%, and compared with the multi-source data fusion LSTM model, the RMSE is reduced by 12.94%.