Urban Road Network Partitioning And Speed Prediction Based On Spatiotemporal Similarity

In recent years,with the rapid development of economic,the number of cars continuously increases in big cities.It causes many severe problems such as traffic congestion and pollution.For reducing traffic congestion,effective monitoring and analysis for traffic state,assessment management,and state prediction are required for road networks.Proper partitioning for heterogeneous large-scale road networks and accurate prediction of traffic states by using the spatiotemporal traffic data in the networks,not only are important research topics in the field of Intelligent Transportation Systems,but also provide decision support for traffic management.In this paper,noticing the characteristics of macroscopic traffic flow that changes severely over time and is heterogeneous in spatial distribution,the spatiotemporal similarity of the speed time series is analyzed,the road network partitioning algorithm based on spatiotemporal similarity is studied,and the short-term traffic speed of the target link is predicted,by algorithm design and experiments using the car speed data collected on the links of a large-scale road network in the northeastern second ring of Beijing.The works in this paper are of practical significance for traffic flow management and control of large-scale road network.The main contents of this paper are as follows.First,for measurement problem of distance and trend for the traffic flow time series,a time series similarity measure called PE distance,that combines Pearson correlation coefficient and Euclidean distance,is introduced.After pretreatment of the speed data collected in a real road network,the temporal similarity and spatial similarity of the speed time series are analyzed based on the PE distance.The periodicity of the speed time series is verified in the time domain.And,it is also verified that the speed time series similarity between the link and its adjacent links decreases when the spatial distance between them increases.Second,considering spatiotemporal similarity of traffic speed time series,a road network partitioning algorithm based on spatiotemporal similarity is proposed using the technique of Normalized Cut(NCut),to which the spatial connectivity constraint is introduced.The objectives of road network partitioning are homogeneity within subregion and differences between adjacent subregions.Based on multiple evaluation indexes,the result of road network partitioning is optimized,the effectiveness and advantage of the proposed algorithm for road network partitioning are verified by experiments and comparisons.Finally,by using spatiotemporal similarity of traffic flow and road network partitioning result,short-term traffic speed prediction problem is studied.NCut algorithm for road network partitioning is incorporated into Long Short-Term Memory Neural Network(LSTM NN),and a short-term traffic speed prediction model called NCut-LSTM is proposed in this paper by constructing a neural network’s input data set composed of the links spatially correlative with the target link.The orthogonal genetic algorithm is used to optimize the hyperparameters of the model.In comparison with two speed prediction models based on different spatially correlative link sets in experiments,the advantages of the proposed NCut-LSTM short-term traffic speed prediction model are shown.