| abstract |
The invention discloses a PM2.5 hour-by-hour prediction method based on a neural network. The method includes acquiring aerosol optical thickness AOD data, ECMWF meteorological data, ground-based data and auxiliary data, preliminarily processing the data, constructing a first training sample, and training the AOD. Fill the model, predict and complete the AOD data of aerosol optical thickness, construct the second training sample, train the learning model to obtain the PM2.5 prediction model, and predict the PM2.5 concentration in eight steps. The invention realizes the expansion of samples based on the deep neural network complementing the aerosol optical thickness AOD data, and corrects the atmospheric boundary layer height BLH to the atmospheric haze layer height HLH for prediction of PM2. .5 The problem of insufficient coverage has achieved high-precision hour-by-hour prediction of PM2.5 concentration. |