Study On The Feature Extraction Of Traffic Travel Time Probability Density And Its Application Of Aggregation Query

The parameters such as mean,variance,and percentile,etc.are important indicators for evaluating the stability of traffic travel.However,it is not easy to get these indicators through aggregation query in the original data at the interactive speed.Because of the massive characteristics of traffic data,there is a large amount of dirty data that is not conducive to the calculation of traffic travel time.It is time-consuming and laborious to separate them.Besides,there are mixed invalid observations that can't reflect the normal traffic conditions and they can't be simply separated by thresholds because of the variability of travel time at different times of the day.When necessary,travel time evaluation also involves factors that consider the types of vehicles(passenger cars,taxis,and buses)individually or collectively.The existence of these problems leads to a very low utilization degree of traffic travel time information services.In view of above problems,probability density feature extraction model for traffic travel time was first proposed.After separation of dirty data,the model utilizes the long-tailed distributions of invalid observations to cluster travel time observations into components and distinguish between valid data and invalid data in the form of density component according the finite lognormal distribution mixture model.It also implements a quantitative description of the shape of the probability density of irregular traffic travel time.Then,according to the density component parameters and the data amount of the clustering group,the feature vectors of the traffic travel time are formed in different time intervals.The feature vectors are the output of the model of travel time probability density feature extraction and stored persistently.On this basis,a calculation method for traffic travel time aggregation query indicators based on feature vectors is proposed.According to aggregated queries across different time intervals,the method can generate feature random data according to the proportion of data in different time intervals to represent the probability density distribution characteristics of corresponding original data.The methods above can calculate the approximate values of the mean,variance,and quantile,etc.at an interactive speed based on feature vectors without relying on the original data of traffic travel time.Besides,the method analysis obtains the characteristic vectors of the observations of different vehicle types and different time intervals,and feature vectors contain both valid data's and invalid data's feature values,so that multiple types of combination of these indicators can be calcuated.The results have a positive effect on improving the application value of traffic travel time observations.