The Traffic Forecasts Based On GM-BP Model

Urban traffic congestion, traffic accident frequency has become a serious social problem. The basic idea of the traffic accident forecast is basis on analyzing and processing the accident statistics, and the development of the cause of the accident. At the same time, it can make logical speculated that judgment prior to happened or not clear accident.Firstly, this paper introduces the basic theory of used knowledge of the forecasting traffic accident: forecasting theory, the gray system theory, the neural network. And secondly, according to the advantages and disadvantages of the several mentioned about road traffic accidents prediction methods, the paper chooses grey forecasting model and neural network. In other words, it is called GM-BP(grey forecasting model-neural network)model, and then realize the prediction of traffic accidents. The disadvantages of BP algorithm are slow learning speed. And the objective function exist the local minimum of points. This paper proposes an improved fast algorithm, namely combining the adding momentum item and adaptive learning rate to improve the method of the learning rate; Meanwhile, in view of the deficiency of basic grey model in regular changes faster data of series prediction accuracy, the unbiased model grey model for road accident forecasting is proposed in this paper. At last this paper uses the fast algorithm by using the modified and improved model forecast one important index of road traffic safety factor, which is direct economic loss caused road traffic accident. And verify the effectiveness of this model in the prediction of road traffic accident. The result shows that the combination model can take full advantage of every single model and improve the practicability in road traffic accident forecast model.