The Research on RBF Network Structure Optimization and the Application in Transportation Prediction

The grey relational analysis (GRA) and sliced inverse regression (SIR) are applied to reconfigure the input structure of RBF neural network. In GRA, variables which have a higher grey relational grade with the output and lower dependence on other selected variables are reserved as inputs factors. The elimination of information overlapping in these reserved factors is implemented by SIR to extract the principle components as the final network input variables. This two-step method can reduce network dimension, improve generalization capability. The RBF neural network based on GAR-SIR is applied to the transportation freight volume prediction. The prediction evaluation indices verify the availability of the proposed model