Research On Traffic Flow Condition Recognition Algorithms Based On Data Mining

Traffic flow condition recognition is one of the important issues of Intelligent TransportationSystems (ITS), especially for its Advanced Traffic Management Systems (ATMS) and AdvancedTraveler Information Systems (ATIS) research. The objective of traffic flow condition recognition is torecognize the condition of traffic flow timely and accurately, take effective traffic incident management,traffic control and traffic flow guidance measures to clean out traffic incidents timely. In this way, theloss of casualties will be lowered and the second incidents will be avoided. In addition, the controlmeasures of transportation demand can be taken effectively and the route recommendations can beavailable to travelers so as to lessen traffic pressure, economize energy and reduce pollution.According to the demands of dynamic traffic management, traffic incident management and urbantraffic flow guidance, the theory of traffic flow condition recognition, including traffic flow short-termforecasting and real-time detection, is studied deeply in this paper. An architecture of traffic conditionrecognition system is put forward based on the basic thought of dynamic traffic management and thearchitecture of traffic incident management system. Two traffic flow condition recognition algorithms,namely the traffic flow short-term forecasting algorithm based on non-parameter regression splinefitting and the traffic incident detection algorithm based on wavelet multiresolution analysis, arepresented according to the characteristic of traffic flow. The parameters of algorithms are calibrated andtwo off-line tests are made according to the data under the condition of incident and non-incidentobtained from Transportation Research Group (TRG) of University of Southampton of UK and thesimulation. The results of tests proved that both algorithms perform better in traffic flow conditionrecognition.