Application of Multilayer Perceptron Neural Network for the Forecast of Tropospheric Ozone in Hanoi ☆

This study deals with the application of a multilayer perceptron (MLP) neural network for the forecast of tropospheric (ground or surface) ozone. The model was proposed on the basis that the ozone concentration is a function of meteorological parameters and its precursors. To determine the best fitting architecture of the network, the genetic algorithm (GA) was used. The datasets used for this work are extracted from data of the air quality monitoring station located at Lang, Hanoi, in the period between 2002 and 2004. The forecast period of the MLP model was selected to be eight hours, equivalent to a work shift. The performance of the MLP model proposed is compared with that of multiple linear regression (LR) method. Obtained results show that the forecasting performance of MLP model is better than that of LR model all over studied cases. The statistical indicators of MAE, RMSE and R2 of the proposed MLP model are 4.49 ppb, 5.37 ppb and 0.54, respectively, which are consistent with those of various similar studies in the world. These indicate that artificial neural network (ANNs/MLP) is a good and feasible approach to build the forecasting model for the air quality including ozone concentration