Application of Composite Grey BP Neural Network Forecasting Model to Motor Vehicle Fatality Risk

An accurate mathematical model for describing traffic accidents is difficult to be constructed due to various factors such as humans, vehicles and environments. To achieve a better estimation of traffic crashes, a novel composite grey BP neural network (CGBNN) model is presented in this paper. First, the original predicted values of traffic accidents are separately obtained by the GM (1,1) model, the Verhulst model and the DGM (2,1) model. Then, a CGBNN model is constructed by fusing the advantages of the grey models and the BNN model to improve the forecasting precision of the original grey models, the reasonable weights of the neural networks are acquired by an iterative training and learning process. The results of the CGBNN model on predicting real-world traffic fatalities show that the forecasting accuracy is much enhanced when the proposed method is applied.