Traffic Feature Mining And Overall Mode Prediction In Urban City With Large Scale Data

The urban traffic condition can affect the citizens' travel experience,which is also one of the important developmental indices for urban city.The intelligent transportation system is the focus of related researches and departments.However,under the big data backgrounds,the massive traffic data provides us abundant information about road network topologies and time-spatial information,in the mean time brings challenges about huge amount data computing.To realize an accurate and reliable overall traffic mode definition and prediction can lay the foundation for the researches of intelligent traffic system.On the basis of this research backgrounds,this paper's study object is the large-scale urban traffic network.First of all,we design a feature mining method,in which the traffic states are devided into several different modes,we propose a mode prediction method based on variable order markov method.Furthermore,we try to figure out the correlation information between different roads or regions,and propose a high efficient algorithm suitable for large-scale road network.Finally,we imrove the forecast method in order to solve the coupled network mode prediction.The research contents are introduced as below:1)Define the overall traffic mode and the mining method.Use the traffic partition and data refusion to devide the similar roads into one region,thus transform the road data into the region data.To obatain the overall traffic mode according to the region's data,and then discuss the different results using different clustering method or distances indices.2)Propose a prediction method about the overall traffic mode series.First we use different ensemble learning methods to predict the future traffic mode.Then we put forward our forecast framework similar to the emsemble learning,whose base leaners are designed by the variable order markov model and probability suffix tree.We calculate the transition probability matrix between different modes,and get the final voting results from a weighted ensemble way.We get a good experimental result using the real traffic data in Shanghai.3)Study the association rules apply in the urban traffic.Propose a multivariate association rules algorithm according to the concept of hash map.This algorithm is applied to the large-scale urban traffic network.We also introduced the improved confidential index.We compare this method with the traditional association mining algorithm regarding to the time and space complexity.Next,we do the experiments about the road and region analysis with synchronous and asynchronous association.4)Design the multi-level traffic network overall modes multiple step forecast.Due to the more complex and coupling features,the multi-level traffic network has a difficulty to guarantee a precise result.First,we devide the level of the whole network.Secondly,making use of the association beteween the coupled network and the predicted network,we create a new traffic feature.Last,we design the iteratively prediction with multi-step pruning.The algorithm is verified by the real traffic data.