A Hybrid Model Based On CEEMDAN And Frequency Division For PM_2.5 Short-term Predictions

Accurate prediction of air quality is of great significance to the prevention of the spread of diseases,and at the same time,it has a certain auxiliary role for decision-makers to formulate future development strategies.Among the various pollutants of air quality,PM_2.5has the greatest impact on the human body,and the general sta-tistical models can no longer satisfy its prediction.In this paper,we will decompose the PM_2.5sequence using the complete ensemble empirical mode decomposition with adaptive noise（CEEMDAN）,and then divide the subsequences into three frequency categories based on the sample entropy threshold.In the empirical process of this article,the performance of a single model is first verified,then the model is verified for different frequency sequences,and finally,a mixed model is established and its performance is studied.In this paper,the high-frequency sequence is fitted with the gated recurrent unit（GRU）,the intermediate-frequency sequence is fitted with the back propaga-tion neural network（BPNN）,and the low-frequency sequence is fitted with the autoregressive model（AR（p））.Compared with the GRU network,this model has improved RMSE and MAPE accuracy and compared with CEEMDAN-GRU（HM1）,the improvement of the two indicators is 12.80%and 12.87%respectively,and the proposed model shows the stable state under the different step-ahead prediction.