Use of Neural Net Models to Forecast Atmospheric Pollution

The forecast of the CO and NO₂ concentration levels has been obtained by the 3-layer Perceptron Neural Network with Error Backpropagation learning rule.This study shows the 3-layer Perceptron performances in relationship with the choice of the activation functions parameters (depending on the statistical values of the input/output variables).A first simulation set using data at 1 hour before has been utilized to forecast CO levels. A second simulation set with data at 12–24 hours before has been used to forecast NO₂ levels.The Neural Net's performance appears to be very good both for the parameters activation function optimisation and the variables choice.