Study On Vehicle Rear-end Collision Prediction Model In Urban Road Intersection

Urban Road Intersection is a road node which with a high rate of accidents. Rear-end accident accounts for the largest proportion of all the accidents. In recent years, researchers have attempted to solve the problem from the theory and technological levels. The establishment of rear-end predictive mathematical model is the basic starting point for the research of rear-end accident. Usually the establishment of safety distance model is the foundation of the study. Probability theory is used to descript the possibility of adjacent cars’ rear-end collision. Based on the probability of rear-end, the risk of rear-end accidents can be studied quantitatively.Firstly, the mechanism of the rear-end accidents in urban road intersections and related factors were analyzed in this thesis. With the method of the combination of qualitative and quantitative analysis, the incentives(such as driver factors, environmental factors and traffic factors) were deeply discussed. We found that the driver is the most important factor. However, environmental factors and road traffic factors also play a great impact. Secondly, with the establishment of safety distance model, reasonable values of the safety distance between vehicles were calculated in different situations. Through the discuss of the basic principles of the theory of Support Vector Machine(SVM), then SVM is applied to the analysis describes the degree of association between road safety and its impact factors of steps, then discusses the genetic algorithm(GA) to optimize the parameters of the principles and methods. Thirdly, the factors affecting urban road intersections vehicle rear-end were analyzed to in this thesis. Five indexes(driving distance, the vehicle speed, the speed difference between the front and rear, the car braking time, braking deceleration speed) which have important relation with rear-end collision were confirmed as input variables of the rear-end collision probability prediction models. Meanwhile, rear-end collision probability was confirmed as output variables. SVM probability prediction model is established by MATLAB, and then MATLAB can scale the impact on rear-end probability of various factors by multivariate linear fitting method. The results are as follows: driving distance, the car speed, the speed difference between the front and rear, the car system Activity time, braking deceleration. Finally, the speed difference is used as the measurable parameters to quantify the severity of the accident. Combined with the probability of rear-end results using fuzzy c-means(FCM) clustering algorithm to establish the risk of rear-end with relaxation matrix.In this thesis, the experimental data is collected by Video Drive Recorder(VDR). Twenty sets of sample data as the training set SVM model, and with another 30 groups of test data sets as model validation. Through the analysis of experimental data and using ORIGIN analyses actual and predicted values, good results can be obtained.