PREDICTION OF PM2.5 CONCENTRATION IN XI'AN BASED ON CEEMDAN-SE-BiLSTM MODEL
Atmospheric particulate matter concentration is closely related to environmental pollution,and accurate prediction of PM2.5 concentration is crucial for ecological environmental protection.Based on PM2 5 concentration and meteorological data from January 1,2020,to December 31,2021,in Xi'an,the PM2.5 concentration series was decomposed into multiple eigenmodal components by complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)for the non-stationary and non-linear characteristics,and the sample entropy(SE)was used as an indicator to perform k-means clustering to reduce data noise.Then the reconstructed components were inputted into a bi-directional long short-term memory model(BiLSTM model),supplemented with the enhanced information of meteorological data and temporal data after unique thermal coding processing to output the prediction results of each component,finally superimposed to obtain the final PM2 5 concentration prediction results.The results showed that the CEEMDAN-SE-BiLSTM model had better prediction performance at four future moments(T+3,T+6,T+12,and T+24)compared with the XGBoost model,long short-term memory neural network(LSTM)model,BiLSTM model,and other combined models.The CEEMDAN-SE-BiLSTM model had better prediction performance,in terms of root mean square error(RMSE),mean absolute error(MAE),and mean absolute percentage error(MAPE)were all decreased and at the moment of T+3,the determination coefficient(R2)was 0.993.The prediction accuracy was greatly improved.In addition,five cities(Zhengzhou,Chengdu,Beijing,Shanghai,and Guangzhou)were randomly selected nationwide to verify the model's generalization,and the results showed that the prediction errors in the five cities were all small.The CEEMDAN-SE-BiLSTM model can be extended to other regions and cities and is capable of accurate short-term prediction.
BiLSTMCEEMDANPM2.5 predictionsample entropytime series
万方数据
