|
Abstract:Hourly-based improved prediction of present air quality state (PM10, PM2.5, and NO2) with 15 input independent variables (3 hours\' earlier PM, gas and meteorological data of a Korean city affected by 2 days\' earlier PM and gas data of Beijing city in the Yellow Dust route was performed using Machine Learning (ANN-Tanh) and Multivariate Regression techniques. ANN-tanh model of multilayer perception (MLP) structure with a feed-forward input data and backpropagation training process for error calculation with 15 nodes in a single hidden layer was adopted to calculate present urban air quality. Root mean square error (RMSE) and the coefficient of determination (R2; Pearson R) evaluated the performance ability of the model between the predicted and measured values before, during, and after the Yellow Sand event at the Korean coastal city. Multivariate regression technique was also used to predict the urban air quality with the same input data. The prediction accuracies of the two models were compared before, during, and after the Yellow Dust period and regardless of the event, Pearson R correlation coefficients using the ANN-tanh model (multivariate regression model) were quite high as 0.935 (0.961), 0.943 (0.948) and 0. 947 (0.920) for PM10, 0.942 (0.909), 0.969 (0.977) and 0.938 (0.947) for PM2.5, and 0.925 (0.860), 0.853 (0.875) and 0.886 (0.903) for NO2, respectively. Their prediction ability overall was quite excellent, showing a slightly higher R in the ANN-Tanh. Their hourly distributions of predicted and measured values of 3 outputs were drawn to compare their model performance. |