Research On Preprocessing Methods Of Loop Detector Data

In recent years, in order to alleviate congestion which is more and more serious, the managers used the advanced theoretic or technical method to manage and control city road as well as highway. Intelligent Transportation Systems (ITS) play an important role in this way. Traffic can't be managed effectively using ITS without traffic data. There is no meaning for accurate methods without traffic data. Pretreatment of traffic data before it is applied is a key question and will be solved urgently. Therefore, the collecting principle of traffic data and mechanism of abnormal data in loop detector were discussed, and the pretreatment technology of traffic data which can be applied into practice was set up.As the reason of the disturbing environment and interrupted communicating network, the real-time traffic data is half-baked and has some lost and abnormal data. So it can be the import of traffic models. On the base of the characteristic in traffic data, this paper provided four processes method of data screening which contains basic screening, threshold value screening, screening based on traffic flow theory and screening based on quality-control theory. This four processes method which considered the reason of abnormal data and interval of sampling can eliminate abnormal traffic data effectively.Given data reconstruction methods can be used to replace the screened abnormal traffic data for applying real-time. Four data reconstruction methods were provided such as method based on time-serial, method based on history data, method based on space position, and method based on the correlation of time and space. Different methods can be chose in different conditions and different interval of sampling.Utilizing provided data screening and reconstruction methods, the process of traffic data pretreatment were found to confirm real-using of traffic data. In order to validate the effect of methods above, loop detector datum in city expressway ofBeijing and Shandong road of Qingdao were employed. Using real datum, false data can be screened fast and reconstructed accurately. The error between real data and reconstructed data were acceptable in engineering projects.Speed is the most basal and important parameter to describe traffic condition, and also can be used in traffic incident detection and forecast of travel-time. But it can't get speed from single loop detector. The expense of upgrading single loop detector to dual loop detector is so great and also need to interrupt traffic flow. After analyzing the shortage of g-factor method, a linear regression model which considered the relation between occupancy and speed was provided. In order to reduce error of linear model, nonlinear Fuzzy Neural Network (FNN) model was provided. These models improved the precision of estimating speed.Furthermore, the passage time over the detector for every vehicle can be get from single loop detector. Different vehicle types have different passage time. Using real datum, the area of every vehicle's speed was got, and the maximum likelihood method (MLM) was established. This method can reduce the error more deeply than former methods. After sensitivity analysis, MLM is not sensitive to parameter, and can ensure estimating speed steadily.The research content of this paper mainly contains three parts: data screening, data reconstruction and speed estimation from single loop detector. These methods can ensure the accuracy and validity of traffic data. The research does not only improve the theory of traffic data's pretreatment, but also can be used in engineering projects and have the economy value.