Using Neural Networks for Short-Term Prediction of Air Pollution Levels

The present paper focuses on the prediction of hourly levels up to 8 hours ahead for five pollutants (SO2, CO, NO2, NO and 03) and six locations in the area of Bilbao (Spain. To that end, 216 models based on neural networks (NN) have been built. The database used to fit the NN's has been historical records of the traffic, meteorological and air pollution networks existing in the area corresponding to year 2000. Then, the models have been tested on data from the same networks but corresponding to year 2001. At a first stage, for each of the 216 cases, 100 models based on different types of neural networks have been built using data corresponding to year 2000. The final identification of the best model has been made under the criteria of simultaneously having at a 95% confidence level the best values of R-2, d1, FA2 and RMSE when applied to data of year 2001. The number of hourly cases in which due to gaps in data predictions have been possible range from 11% to 38% depending on the sensor. Depending on the pollutant, location and number of hours ahead the prediction is made, different types of models have been selected. The use of these models based on NN's can provide Bilbao's air pollution network originally designed for diagnosis purposes, with short-term, real time forecasting capabilities. The performance of these models at the different sensors in the area range from a maximum value of R-2=0.88 for the prediction of NO2 1 hour ahead, to a minimum value of R-2=0.15 for the prediction of ozone 8 hours ahead. These boundaries and the limitation in the number of cases that predictions are possible represent the maximum forecasting capability that Bilbao's network can provide in real-life operating conditions.