Pattern Recognition And Forecasting Of Traffic Volume Based On Multi-dimensional Feature Clustering

The air pollution can affect the development of national livelihood and economic.Frequent pollution occurs more in areas such as Beijing,Beijing-Tianjin-Hebei,and the Yangtze River Delta city clusters that are densely populatied and have rapid development of urbanization and industrialization.Therefore,the meteorological and environmental department works hard on establishing a dynamic air quality measurement model for realization of real-time prediction of atmospheric pollutant concentrations and early warning of severely polluted weather in recent years.Emissions dischargerd from industrial and vehicles in the atmosphere account for more than 80% of the total pollutants.Among them,the industrial pollution can now achieve real-time monitoring at the source basically.But it is difficult to get emissions from every vehicle.Therefore,the annual emissions inventory of motor vehicles is mainly estimated based on the vehicle ownership and used for assessment nowadays.However,the static total estimation cannot reflect the differences in emissions between individual links of the road network,and it is difficult to estimate the time-varying emissions of the road network on a specific date,of which the spatiotemporal flexibility is poor.Therefore,compared with the static pollutant emission inventory,if the dynamic total vehicle emissions can be take as a dependent variable and entered into the air quality model,it will be able to vigorously promote the construction of the model,which is a key problem that needs to be solved urgently.The dynamic estimation of road network emissions is mainly based on the timevarying vehicle kilometers traveled(VKT)and static emission factors(EFs).The traffic volume of road link is the basis for VKT calculation.However,due to the high cost and low coverage of traffic volume monitoring,it is still difficult to obtain large-scale and real-time traffic volume data through detection devices.Considering that the traffic volume of hourly granularity can meet the accuracy requirements of dynamic emission measurement,and the pattern of daily traffic volume on road links has the characteristics of periodicity and similarity,this study attempts to mine the multi-source traffic data and its advantages,then build a pattern library that can describe road traffic volume patterns,and use massive floating car data to realize the link-based traffic pattern recognition and retrieval,and finally propose a dynamic traffic volume forecasting method based on link traffic volume pattern for emission measurement application,which can serve for the research and application that need dynamic traffic data.The main research of the thesis include the following parts:(1)According to the existing research experience on traffic volume prediction and traffic pattern clustering,combined with the purpose of dynamic emission estimation and requirement of air quality modeling in this study,the time granularity of traffic volume prediction and the research technical route are determined.(2)Analysze the characteristics,advantages and disadvantages of the RTMS data,highway traffic observation station data,floating car data and daily average traffic data that can be collected in Beijing,and propose a dynamic traffic forecasting scheme based on multi-source data fusion.(3)First process the RTMS data of 4 types of urban road and the data from observation stations of 5 types of highway in Beijing in 2018.Subsequently,compare different schemes of clustering characteristic parameter and choose the best.Thirdly,use the improved K-means clustering method to construct a library of traffic volume patterns under different road grades and the date types.Then analyze the clustering results of the traffic volume under different date types and road levels combining with the reality.(4)Based on the Underwood traffic model,a concept of the normalized fundamental diagram is proposed and the calibration of relevant parameters is realized,so that the conversion between floating car speed and the characteristic parameter of traffic volume can be achieved.After that,the shortest weighted distance principle was used for retrieval and identification of road traffic volume patterns.(5)A method for dynamic traffic volume prediction is proposed based on link traffic volume pattern,and then it is compared with the existing speed inversion method using fundamental diagram.The prediction results of sample links show that the average root mean square error and mean relative error of the mentioned method in this study are reduced by 43.0% and 82.6% than the speed inversion method respectively,which effectively solve the problems of inaccurate prediction of the speed inversion method in high-speed interval and the poor universality of the fundamental diagram.(6)Finally,based on the speed of floating car,the dynamic traffic volume prediction method proposed above is applied to the dynamic vehicle emissions estimation of road and the prediction of congestion spreading speed.It shows that the dynamic estimation of vehicle emission under different area ranges and date requirements can be achieved using the proposed method,and the prediction accuracy of the congestion spreading speed reaches 85.12%,which can be applied to dynamic traffic guidance and control.