A Study Of Urban Air Quality Prediction In Nanning City Base On Combination Of The Genetic Algorithm And Artificial Neural Networks

This thesis gives an analysis on temporal and spatial characteristics of urban air pollutants in Nanning city according to the monitoring density data of SO₂, NO₂, and PM₁₀ from 2001 to 2006. A further analysis based on time delayed cross correlation function was made about the relationship between the pollutants' density and the corresponding surface conventional meteorological data in the year of 2006, the influence of weather conditions on the variation of atmospheric pollutants' density and theirs changing tendencies were also discussed. The input vectors of artificial neural networks were determined by time delayed cross-correlation coefficient analysis, and the model of urban air quality prediction was constructed, which has two hidden layers of neural networks. The values for individual weights and biases were determined by combination of genetic algorithm (GA) and back-propagation algorithm (BP), the training data of artificial neural networks models are 10 days', 15 days' and 20 days' data of history observation, the 24 hour's average density of SO₂, NO₂ and PM₁₀ in urban area were predicted according to above three different training data. Adding today's monitoring density data and today's corresponding observation of meteorological data to the training data and deleting the longest time of history data in order to construct a new training data after the neural networks have forecasted today's air quality. Then using this new training data to retrain the neural networks to determine the values of connected weights and biases again, so that all these internal parameters of urban air quality forecasting model were updated once a day by this way to improve the accuracy of prediction.The results of air pollutants' density prediction from above three forecasting modes were assessed by mean absolute percentage error (MAPE), mean square percentage error (MSPE), average prediction accuracy, API prediction accuracy and air quality grade prediction accuracy, the analysis results show that the prediction errors from different pollutants using different training data are not obvious, all the errors range from 20 percent to 30 percent. The air quality prediction model of artificial neural networks in Nanning city was established utilizing 10 days' training data for forecasting SO₂, 20 days' history data for NO₂ and 15 days' for PM₁₀ respectively to forecast the future of 24 hour's air quality. Then the accuracy of primary pollutant prediction density, API and urban air quality grade forecasting were used to analyze and evaluate the model prediction results, the results from artificial neural networks model show that the accuracy of them are 78.85%, 52.56% and 84.62% respectively. Finally, the graphical user interface (GUI) of air quality prediction of neural networks model was compiled based on MATLAB 7.0 language, the interface of which is simple and convenient to operate.