Forecast Of Road Accidents Based On Grey Theory And Neural Networks

The road accidents forecasting is the important content of the researchof traffic safety. The purpose of road accidents forecasting is to analyze thetendency of road accidents under existing road traffic conditions, evaluate thefeasibility and practical effectiveness of traffic safety measures reasonably,control the factors affecting road accidents, and reduce the traffic accidents. Thecharacteristics such as nonlinearity, randomness and uncertainty in traffic systemmake it difficult to forecast road accidents, the behavioral feature of trafficsystem. Based on the analysis of existing macroscopic forecasting methods ofroad accidents, the forecasting models of road accidents inconsideration ofmultiple attribute characteristic of traffic system are constructed in this thesis,which are grey forecasting model, artificial neural network forecasting model,the combined forecasting model based on grey theory and artificial neuralnetwork.The merit of grey model GM(1,1) is that the arithmetic is simple and itcan use few data to construct the model, these bring the facility of constructingmodel but the grey model to forecast the result of the fluctuating systemcursorily and the precision rate decreasing with the time going on. In view of thedeficiency of basic grey model, 3 original grey models for road accidentsforecasting are proposed in this paper, which are the metabolism model based onthe best initialization, unbiased grey forecast model and the grey-Marko modelfor forecasting road accidents, to improve the applicative scope and forecastingprecision. BP neural network has good characteristic of nonlinear, but it won'tworkwell without abundance data. So a combined forecasting model based onthe grey model and BP neural network is presented. The presented newcombined model synthesizes the advantages of GM forecast method, which issimple and needs less original data to discover the rule and model, and BP neuralnetwork which possesses the characteristic of nonlinear fitting and needs all-sided original data at the same time. In view of the all-sided data of the roadaccidents relative gene offered, a grey relational analysis BP neural networkmodel is constructed. At last we use those forecasting models that we putforward in the paper to forecast the road accidents according to the statistics inthe last few years. The result show that the combination model can take fulladvantage of every single model and avoid the disadvantage and the result ofprediction is super than that single model to draw.