The Research Of Expressway Traffic Incident Detection Algorithm Based On The Prediction Deviation

With the rapid development of transportation and the increasing number of urban motor vehicles, the construction of urban expressway has also been rapid developed in many civil metropolitan cities. The traffic incident occurred in urban expressway will not only threaten the safety of people and vehicles, more importantly, it will produce more serious bottlenecks of traffic congestion which dramatically reduce the capacity and efficiency of urban expressway system, and it will give rise to urban traffic problems eventually. Intelligent Transportation System is widely recognized as one of effective method in solving urban traffic problems. As the core component of Intelligent Traffic Monitoring System, the research of Automatic Incident Detection is of great significance in improving the overall management of urban expressway standard and giving full play to the superiority of urban expressway.Based on the characteristics of urban expressway traffic flow and the basic principles of traffic incident detection, this thesis studies the prediction-deviation-based urban expressway traffic incident detection algorithm with the combination of dynamic prediction and incident detection of traffic flow.Through analysis, I found that the prediction of traffic flow parameters is the key factor which influences the detection rate and false alarm rate of the algorithm. For the low accuracy and stability of the present algorithm which uses the single forecasting model, this thesis establishes a multi-model fusion forecasting method which uses statistical analysis method, artificial neural network, data fusion technology. And when the weight is calculated, the determination length of stride smallest absolute value law is proposed. On the basis of the multi-model fusion forecasting method, a traffic incident detection algorithm which is based on the multi-model fusion prediction deviation is proposed in this thesis. And in the process of applying artificial neural network to prediction, in the view of problem that the present single point and single-step prediction model which is established only considers the influence of time factor, a new double point and single-step prediction model which considers the influence of traffic flow in the space from upstream to downstream is proposed in this thesis. In addition, through the study of traditional AID system which is based on three-stage alarm system, this thesis found that the system can't give full play to the performance of the three-stage alarm, and put forward some improvement measures to this problem.In this thesis, simulation results show that the prediction accuracy of multi-model fusion prediction method is higher than the single model prediction method, and the traffic incident detection algorithm based on the multi-model fusion prediction deviation has higher detection rate and lower false alarm rate, thus has better comprehensive detection performance, and it provides a method for designing high-performance detection algorithm.