Study Of Key Problems For Capturing High-Emitters In Road Network Remote Sensing System

Traffic emission is the primary source of air pollution in cities.In order to alleviate the air pollution,the monitoring of on-road mobile pollution source is the prerequisite to take pollution control measures.Currently,the frequently-used method to measure traffic emission is to make emission standards and enforce vehicle emission inspection periodically.In the process of inspection,vehicles are driven in certain operating con-ditions.Then emissions from the exhaust pipe are collected and quantified.Thus,the vehicle emission inspection is of high precision,uniform standard,and good repeata-bility.However,this kind of emission inspection has a long detecting time,and long inspection period,which determines that the emission situation between two inspection is unaccessible.Moreover,the inspection results can not reflect the real emission level when vehicles are running on road.As a complementary means,on-road vehicle emis-sion remote sensing can be applied in the road network,which can measure the concen-trations of pollutants in the vehicle exhaust according to absorption spectrum when they are up and running.On-road remote sensing has been proven to be an effective method to screen high-emitting vehicles and identify clean vehicles.In recent years,on-road remote sensing has been applied in many countries and areas,as a tool to monitor vehi-cle emissions,estimate fleet emissions,and investigate pollution source.However,the practical application encounters several key problems that are pending,especially for high-emitter capturing.In this thesis,we are concentrated on such problems,including the location of remote sensing devices and the identification of high-emitters based on remote sensing measurements.The main work is as follows:1.A location method of remote sensing devices is proposed based on the topology structure of the road network.First,graph-and hypergraph-based models are estab-lished for the road network.Through analyzing the motion rules of on-road vehicles,and considering the turning restrictions,the location problem of remote sensing devices on the road links are transferred into the line graph,and the method to construct the traffic circuit hypergraph and ?-traffic circuit hypergraph is developed.Then,a location algorithm combining depth-first searching and greedy strategy is proposed to capture more on-road vehicles.2.A location method of remote sensing devices is proposed based on the vehicle trajectory.With the application of positioning and navigation system,it is possible to acquire vehicle trajectory data.Thus,we propose a remote sensing device location method based on vehicle trajectory data.First,the trajectory-link association model is developed using the hypergraph.Then,we model the interested location problem based on the defined road characteristic matrix.To solve the problem,we propose a simulate anneal algorithm considering the reachability distance between road links.The experiments show that the proposed algorithm can improve the vehicle coverage rate and increase the average detection times.3.A high-emitter identification method is proposed based on the extreme learning machine.First,a workflow to identify high-emitters is introduced in which the on-road remote sensing measurements are input to a high-emitter identification model to screen in-use vehicles.Considering the imbalance of vehicle emission data,the weighted ex-treme learning machine is employed as the base classifier,and the active sample selec-tion strategy is introduced to update the identification model.The experiments show that the proposed high-emitter identification method can decrease the error rate of high-emitters,thus improving identification performance.4.A high-emitter identification method is proposed based on the one class clas-sification.Considering the discrepancy between the real emission situation of on-road vehicles and their emission inspection results obtained from vehicle inspection stations,we try to identify on-road high-emitters using one class classification model which are trained only according to the high-emitter labels.Further,to deal with the scarcity of high-emitter labels,we develop a high-emitter identification method based on semi-supervised one class support vector machine.The experimental results on a real-world dataset verify the performance improvement by using semi-supervised identification method.In this thesis,we attempt to investigate the on-road remote sensing from the per-spective of practical application,and develop feasible approaches to deal with the key problems encountered for high-emitter capturing in the road network remote sensing system.Through the investigation,this thesis can prompt the further application of on-road remote sensing.