Research And Application Of Traffic Congestion Identification And Prediction Algorithm Based On Data Mining

There is a close connection between urban traffic development and human life.So how to allocate the transport resource reasonably and how to plan the urban traffic has become an urgent problem to be solved.Among the traffic problems,traffic congestion has the most impact,the longest duration and highest frequency.The traffic congestion can be classified as frequent congestion and incident congestion.The frequent congestion always happens at rush hours,due to the traffic demanded beyond the normal capacity of roads,with predictability.The occurrence of incident congestion is not fixed,it is due to the sudden traffic accident,with contingency.Above all,the two types of congestion have different conditions and time when they occur.If we can identify the characteristics of these types correctly,it can assist the traffic police to take the right measures timely,improve the utilization efficiency of transport resources,and reduce the economic losses and other external costs.In one word,it is of great significance to explore the regularity of frequent congestion and incident congestion.Meanwhile,the prediction of traffic state also crucial for the public travel and traffic police who have to plan strategies.This paper discusses 2 problems: how to identify the incident congestion,frequent congestion and how to predict traffic accidents.With the data mining technology,this paper explores the application of such technology in the field of transportation.For the identification of frequent congestion and incident congestion,firstly the support vector classifier is trained by the cross validation method.Secondly the traffic flow is input to the classifier to identify the traffic state.Finally,the method of statistics is used to distinguish between frequent congestion and incident congestion.After experimental verification,the accuracy and speed satisfied the actual requirements.In this paper,the recurrent neural network(RNN)is used to study the problem of traffic states prediction.Deep learning is widely used in many fields,such as image recognition and natural language processing.However,the application in the field of transportation is very few.In order to explore this new field,this paper adopts the method of contrast experiment testing on different roads so as to get the best RNN model and solution.The experimental results show that we cannot achieve the expected result by using deep learning model blindly.We need take some steps to increase the accuracy.For example,through the related roads' traffic flow to predict the one next to them,we can achieve the desired results.