Abstract:In order to improve the timeliness and accuracy of PM2.5 concentration prediction, this paper integrates observation factors such as atmospheric pollutants, GNSS PWV and wind speed, and uses the methods of FFT and LSTM neural network to build the PM2.5 concentration prediction model to predict PM2.5 concentration in the next 24 hours. Firstly, fast Fourier transform is applied to the observation elements such as air pollutants, GNSS PWV and wind speed, and the common change period of various elements is extracted to obtain the optimal common period of 216 hours. Then, various elements of the optimal common period length are selected as the model input, and the PM2.5 concentration of 24-hour sequence are taken as the model output. The RBF neural network of PM2.5 single elements and the LSTM neural network integrating atmospheric pollutants, wind speed and GNSS PWV are respectively used to construct the PM2.5 concentration prediction model. Finally, the measured PM2.5 concentration sequence is used to test the external reliability of the two models,RMSE and IA are used as evaluation indexes to evaluate the model accuracy. The results show that the RMSE and IA tested by the PM2.5 concentration prediction model based on FFT-LSTM are 16.22 μg /m3 and 84.36%, respectively. The prediction accuracy of the model could effectively predict the PM2.5 concentration in the next 24 hours. The model can be used as a reference of air quality prediction for air pollution prevention department.
WANG Yong,WANG Hongyi,LIU Yanping et al. Study on the Prediction of PM2.5 Concentration of Hebei Province in Winter by Combining GNSS PWV, Wind Speed[J]. jgg, 2020, 40(11): 1145-1152.