Research On The Traffic Flow Data Recovery And State Classification Methods Of Expressway Based On Microwave Data

Expressway state classification is the precondition for traffic management department to formulate appropriate measures.And quality traffic flow data is the basis of traffic flow state classification.With the development of intelligent transportation,the types and quantity of traffic flow data are also increasing from expressway.How to make full use of various traffic flow data become the key to provide reliable information service for formulating expressway traffic management and control measures.Therefore,by analyzing the traffic flow spatiotemporal characteristics,effective and reliable traffic flow data recovery and state classification are established based on deep learning and fuzzy clustering theory.Then,proposing a expressway traffic flow data recovery and state classification model collaborative framework.Firstly,by comprehensively analyzing the research status at home and abroad about traffic flow data recovery and state classification,the limitations of the research are summarized: the existing recovery and state classification methods are difficult to fully mine the relationship between traffic flow variables of different expressway lanes,as well as satisfy the needs of delicacy traffic management and control.In order to overcome the current research limitations,expressway multi-lane traffic flow are taken as research target and the technical route of this paper is proposed.Secondly,the collection methods and the preprocessing process of expressway traffic flow data are introduced.As for the pre-processed multi-lane traffic flow data,verificating its temporal stationary and randomness as well as spatio similarity and difference from two aspects of temporal and spatio characteristics,which provides the theoretical support for the researches on expressway traffic data recovery and state classification.Thirdly,to deal with the multi-lane traffic data missing problem,according to the distribution of missing data on the time axis,the styles of the data missing are divided into three types including centralized missing,random missing,mixed missing.the multi-lane traffic flow volume missing data recovery method based on deep convolutional generative adversarial network(DC-GAIN)was proposed.The DC-GAIN uses deep convolutional network to capture spatiotemporal characteristics of multi-lane traffic flow and introduce a hint mechanism.By the outer ring traffic flow data form Beijing East Third Ring Road,comparing and analyzing the effect of the DC-GAIN to recovery the traffic flow volume data with different missing rates under three missing types.Fourthly,in order to fully mine the temporal characteristics of various traffic flow variables and accurately classify the expressway traffic flow state,a multivariate fuzzy cmeans clustering based on penalty terms(MFCM-PT)was established based on multivariable fuzzy clustering theory.The MFCM-PT makes high cohesion within the clusters and low coupling between the clusters by introducing membership term.In addition,the MFCM-PT balances the influence weight of each variable on the clustering results to avoid a single variable extremely influence on the state classification,and reduces the sensitivity of the model to initialized clustering center by using weight term.Then,XieBeni index is used as evaluation index.And the results of example show the MFCM-PT illustrates better performance than traditional fuzzy clustering method s in terms of accuracy and stability.Finally,to further improve the applicability of established recovery and state classification model and and avoid the limitation of applying single model.Hence,a collaborative framework of expressway traffic flow data recovery and state classification model was proposed combined with the DC-GAIN and the MFCM-PT.The experimental results show that the proposed collaborative framework is effective and accurate for different multi-lane traffic flow data.There are 50 figures,22 tables,and 90 references in this paper.