Ozone concentration forecast method based on genetic algorithm optimized back propagation neural networks and support vector machine data classification

Multi Artificial Neural Network (ANN) models are used to forecast ozone concentration on single-site for a better forecast accuracy in huge dataset condition. Support Vector Machine (SVM) is used to accurately classify the data into its corresponding categories. Back Propagation neural network (BPNN) was optimized using Genetic Algorithm (GA) to achieve higher forecast stability. SVM classification and GA optimized BPNN (GABPNN) were combined to forecast ozone concentrations in Beijing. The ozone measurements of XiDan sampling site in Beijing were used to test the effectiveness of this method. The modeling dataset used were the records of temperature (T), humidity (H), wind velocity (WV) and UV radiation (UV) from Mar 2009 to Jul 2009. The models were tested using the records of Aug 2009. High accuracy was achieved using this forecast method. Correlation coefficient ( R) of the final models on the test stage ranged from 0.86 to 0.90, with an average of 0.87. The predictions of the final models represented a great forecasting capability that could be applied to the real-life ozone forecast in Beijing. ► We introduced parallel computing theory to solve overfitting problem of Artificial Neural Networks in huge data condition in ozone prediction. ► We used model structure that the O 3 concentration at particular time point can be predicted after entering the predicted value of meteorological conditions obtained from weather report for that time point. ► We used GA to optimize BP Neural Network to increase forecast accuracy. ► We compared the outcome of BPNN, GABPNN and SVM-GABPNN, and concluded SVM-GABPNN is successful.