Research On Key Techniques Of Urban Short-term Traffic Flow Forecasting In Cloud Computing Enviroment

As one of the most common used,the most general,the most complex and challenging contents in human life,transportation has been changing with the continuous changing of the times and the continuous developing of society,from ancient times to the present.Intelligent Traffic System(ITS)is considered to be an important breakthrough to solve the urban traffic congestion,and then build an integrated transportation system,realize the intelligent transportation infrastructure and promote the harmonious and sustainable development of transportation.An important part of the intelligent transportation system is the traffic control and induction system,and one of the key problems in the realization of traffic control and induction is the real-time accurate short-term traffic flow forecasting,that is,how to effectively use real-time traffic data to scroll forward Traffic in a short period of time.This paper studies the key technologies and algorithms of urban short-term traffic flow forecasting under cloud environment.Mainly to solve the accuracy and forecasting speed of short-term traffic flow forecasting under urban complex road network environment.Based on the temporal and spatial similarity of urban traffic flow,a k nearest neighbor nonparametric regression model based on the spatiotemporal similarity of traffic flow(STS-KNN)is proposed,which improves the state vector and distance metric function used in the search for the neighborhood of the nonparametric regression.The model can describe the temporal and spatial characteristics of the traffic flow more accurately.Therefore,the accuracy of prediction is improved.Then,in order to solve the performance bottleneck problem of STS-KNN model in searching historical traffic flow data,a distributed spatial index,M-Quadtree(M:Modified)index based on improved quadtree coding method is proposed to make multi-dimensional floating vehicle trajectory data can be efficiently stored and queried in one-dimensional cloud storage environment.Finally,in order to solve the problem of processing speed for big traffic data,we design a common Traffic Data Parallel Mining Framework based on MapReduce.Based on this framework,the parallelization of STS-KNN model is realized,and a parallel prediction algorithm MSTS-KNN based on MapReduce is proposed.Details of the study are as follows:(1)A A K-nearest neighbor nonparametric regression model based on the spatial and temporal similarity of traffic flow,STS-KNN model is proposed.With the help of the basic theory of graph theory,the concepts of "traffic flow spatiotemporal similarity"and "traffic similarity" are proposed.A new state vector of KNN based on the temporal and spatial similarity of traffic flow,named STSV is designed.A suitable distance function based STSV is defined.Based on these,a nonparametric regression model based on traffic flow spatiotemporal similarity,named STS-KNN is proposed.The model has two advantages:(1)Compared with other parameterized prediction methods,the model has no fixed parameters,which is driven by historical data,and does not depend on the overall distribution form,which is suitable for complex network environment in different regions of the city.(2)Compared with the existing nonparametric regression prediction method,this model not only considers the influence of the past and the surrounding road segments on the traffic state of the target road segment from the Spatial-temporal dimensions,but also uses the "traffic flow similarity degree" to measure the influence of the surrounding road segments,and as an important factor in the introduction of the forecast model,and achieved good predictions.(2)A spatial index M-Quadtree based on improved quadtree coding method is proposed in cloud storage environment.Data-driven short-term traffic flow prediction requires frequent search of data from the massive historical traffic data that matches the current state,and the performance of the search will seriously affect the time efficiency of the prediction algorithm.Therefore,this paper designs a spatial indexing method based on improved quadtree coding method in cloud storage environment.According to the characteristics of cloud storage environment,the existing spatial data partitioning method is improved.First,a spatial data partitioning method based on improved quadtree coding method is designed.Based on this method,a distributed spatial index,named M-Quadtree index for cloud storage "key-value" data structure is proposed,and the algorithm of constructing,querying and updating the index is developed.Finally,we illustrated how to apply the method to short-term traffic flow forecasting.(3)A method of short-term traffic flow parallel prediction,named MSTS-KNN is proposed.In order to solve the problem of search performance bottlenecks in short-term traffic flow prediction algorithms in the face of big traffic data in cities,a general framework of traffic data parallel mining based on MapReduce,named MF-TDPM is designed,based on the M-Quadtree distributed spatial index.and the STS-KNN model is implemented under the MF-TDPM.A multiple sections short-term traffic flow parallel forecasting algorithm,named MSTS-KNN algorithm,based on the MF-TDPM is developed.The experimental results show that the proposed algorithm can improve the performance of traffic flow forecasting in cities and make the short-term traffic flow forecasting to meet the real-time demand of urban traffic managers and travelers.Moreover,the MF-TDPM framework and the MSTS-KNN algorithm also can provide a reference for other similar algorithms in the cloud environment.