Application Study Of RBF Neural Network On Urban Air Quality Evaluation

Urban air quality evaluation is pattern recognition problem essentially, but there are many air quality evaluation indexes and there exist more complex nonlinearly relation between them and air levels, conventional air quality evaluation methods more or less effected by subjective factor, the evaluation result is hard to ensure accurate reliability. Based on the complementarity between rough set and the neural network, air quality evaluation method was proposed which based on rough set and RBF neural network,and used the improved heuristic attribute reduction algorithm, which based on entropy of information to get rid of redundant or unimportant indicators, through it received the best property reduction the best air quality evaluation factors;Then reductive regulation passed for training samples and the part of reductive regulation passed for the center of radial basis function,and establishde RBF neural network model used an RBF neural network with centers,variances and weights directly determined based on matrix pseudo– inverse, which not only optimize the network's structure ,but also avoid iterative process, improve the training speed.The advantages of this method is verified through the comparison with other air quality evaluation methods.