User Travel Hotspot Mining And Traffic Flow Fluctuation Analysis Based On Taxi GPS Data

With rapid development of urbanization and expansion of the urban scale,the number of travel method and the scope of travel are increased sharply.But due to the limited urban space,relatively slowly construction of traffic infrastructure,urban traffic problem has become increasingly serious,which restricts the economic development and the promotion of social operation efficiency,including traffic congestion,uneven distribution of traffic resources,energy waste.The study of the urban resident’s travel pattern and the information of traffic volume fluctuation assists the road user to solve and optimize these traffic problem,such as the high empty loading for the taxi,the unsatisfying demand of residents travel and the less efficient of traffic management.Therefore,based on the analysis of the taxi GPS data,the law of urban residents trip and the fluctuation of traffic flow,we proposed a hybrid predictive model,named Nonnegative Matrix Factorization-Auto Regressive(NMF-AR),which provides the effective and timely information of urban residents,the evolution law of the traffic flow and the fluctuation of the traffic volume for road users.Moreover,this provided information can effectively alleviate the existing traffic problems and improve the efficiency of urban transport.Based on the rapid serious traffic problem,we proposed NMF-AR model,based on the nonnegative matrix factorization and time series model,to estimate the origin-destination(OD)matrix in the future time period,and apply the coarse-graining method and complex network theories to research the fluctuation range from a single area to whole network of urban areas,through by mining and analyzing massive taxi GPS trajectory data to solve traffic problem.The main contributions of this paper are as follows:1.Proposed a hybrid predictive model,named NMF-AR model,which combines the nonnegative matrix factorization algorithm and autoregressive model.Based on the accurate and timely OD matrix,we aimed to solve the OD matrix predictive problem.First,we used the NMF algorithm to extract the base pattern of the resident trip characteristics.Then the AR model was applied to model the nonlinear time series coefficient matrix based on the results of the NMF algorithm.2.Verified the predicting accuracy of the NMF-AR model based on the experiment of the taxi GPS data in Beijing.Firstly,we applied the NMF-AR model to mine and estimate the information of urban residents trip,and introduced some famous short-term traffic flow predicting models to predict the OD flow,including Spatial-Temporal Weighted K-Nearest Neighbor(SWT-KNN),conventional K-Nearest Neighbor(KNN),Back Propagation Neural Network(BP),Na?ve Bayesian(NB),Random Forest(RF)and C4.5.We compared with these predictive model to verify the predictive capability of our model.And we also analyzed the parameters of our model and the sensitiveness for different scale datasets to verify the stability and universality.Moreover,timely predicting OD matrix can provide the timely hot spots and trip counts of urban residents travel and assist in reducing the empty loading rate of a taxi,improving the operation efficiency,and achieve the optimal allocation of resources.3.Analyze the fluctuation in the traffic flow network of urban areas,based on the coarse-graining method and complex network theories.First,we divided the urban area into squares,similar to a chessboard,based on the urban division theory shown in section 2.1.Then,we constructed the traffic flow network of the urban area and done the data preprocess for the traffic GPS data in Beijing.Moreover,we used the coarse-graining method to model the fluctuation in a single area and constructed the network for the fluctuation phenomenon,which was verified by the index of the evaluation in the complex network(e.g.betweenness centrality,inverse participation ratio,point distribution,and so on).At last,due to analyzing the fluctuation of the traffic flow network,we focus on research of the power law of this network,and this power law phenomenon represented the characteristic of network topology in the whole network.Based on the constructed traffic flow network and the taxi GPS data in Beijing and the empirical analysis of the real taxi discharge in urban area network,we studied the relationship between the time-based flow mean and the standard deviation.We can assist the road users and the government about the urban traffic to understand the mechanism of the whole network in the intelligent traffic system and to provide an effective advice for improving the efficiency of urban transport.In summary,based on the mining and analyzing the hot spot of the urban residents travel,we proposed an NMF-AR model to predict and estimate the timely information of urban residents’ trips.Moreover,we also applied the coarse-graining method and the power law distributes to study the traffic flow network.Based on our proposing and analyzing methods,we can timely predict the OD information of urban residents,reduce the empty loading rate of the taxi,improve the operation efficiency,and optimize the allocation of resources.We believe that our research will provide some beneficial reference for solving more similar problems in future.