Study On The Arrival Time Estimation Of A Floating Car

The movement of the floating car on the urban road network is regular.Through the technology of traffic flow prediction,the time when the vehicle arrives at the destination can be predicted on the premise of knowing the departure and arrival location of the floating car in advance.The use of historical data and real-time traffic data to estimate the arrival time of a floating vehicle is one of the important research areas for smart transportation.This paper fully exploits the characteristics of the massive traffic trajectory data,analyzes the problems in the trajectory data pre-processing,model feature evaluation,and predictive training model process of arrival time prediction,and proposes a feature selection based on association mining and neural network-based Predictive models for time estimation solutions.The main research contents and achievements of this article are as follows:(1)Complete the trajectory data preprocessing by using the stroke extraction method of geofence screening and density clusteringThe trajectory of a vehicle is an always continuous coordinate data,which may contain multiple trips with a lot of redundancy and interference information in the middle.How to identify the stroke pattern from continuous motion is a common problem in trajectory data preprocessing.Based on common knowledge,we know that the traffic trajectories in different regions often show different patterns due to social economy,traffic rules,and urban planning.It is difficult to have a unified model to fit the trajectory of the full-range region and achieve better results.effect.The common practice is to use geofences to divide cities into functional areas,and to individually model and adjust parameters for trajectories of different functional areas in order to achieve good results.This article uses data processing based on the expression of the driving state and text vector similarity,and uses the spatial density clustering method to recall the mis-segmented trip,which improves the accuracy and recall rate of trip extraction.After the trip extraction is completed,it is completed.The mapping of trajectory data to traffic behavior,some information sparse features can be aggregated based on the travel dimension,making the data easier to understand and easy to train.(1)Complete the trajectory data preprocessing by using the stroke extraction method of geofence screening and density clusteringThe trajectory of a vehicle is an always continuous coordinate data,which may contain multiple trips with a lot of redundancy and interference information in the middle.How to identify the stroke pattern from continuous motion is a common problem in trajectory data preprocessing.Based on common knowledge,we know that the traffic trajectories in different regions often show different patterns due to social economy,traffic rules,and urban planning.It is difficult to have a unified model to fit the trajectory of the full-range region and achieve better results.effect.The common practice is to use geofences to divide cities into functional areas,and to individually model and adjust parameters for trajectories of different functional areas in order to achieve good results.This article uses data processing based on the expression of the driving state and text vector similarity,and uses the spatial density clustering method to recall the mis-segmented trip,which improves the accuracy and recall rate of trip extraction.After the trip extraction is completed,it is completed.The mapping of trajectory data to traffic behavior,some information sparse features can be aggregated based on the travel dimension,making the data easier to understand and easy to train.(2)Applying association mining to complete the evaluation of the model selection.Actual traffic is a complex process that is affected by many variables.In addition to the vehicle's performance,road congestion status,mileage,and road class,it will also be affected by unexpected events such as weather conditions,traffic accidents,and construction closures.How to find the feature set with the greatest impact on the arrival time from the complex feature set,and the feature set with high information density is the key to improve the prediction accuracy.The traditional feature evaluation methods include filter algorithm and embedded algorithm,but the former is difficult to quantitatively describe the correlation between features and features and the target.In this paper,the algorithm of association mining is used,and based on rule annotation and discretization,the improved filtering method is successfully applied to realize the analysis of the correlation between features and features and targets in a short time.(3)The arrival time estimation model of the extreme learning neural network algorithm is studied.The core issue of this paper is to estimate the arrival time of a passenger by knowing the starting point,destination point,departure time and other conditions of a passenger.There are many mature algorithm models in scientific research and industry that can be used for time-of-arrival estimation,including linear model-based regression,nonlinear model-based regression,traffic simulation-based prediction regression,and so on.At present,the algorithm based on tree model(regression tree,random forest,gradient-elevation tree)is the most widely used.In recent years,with the development of deep learning theory,neural network models have attracted more and more attention.Traditional neural networks are often used for classification learning.Some scholars also apply neural networks to classification through model modification(activation function and output mode modification).However,the neural network itself is more complex in structure and the cost of adjustment is higher.In this paper,using the idea of extreme learning,the extreme learning machine neural network algorithm is applied to the prediction process of arrival time.In order to measure the accuracy and reliability of the model horizontally,two comparatively extensive model regression trees and support vector machines were used to compare experiments.By comparing the three prediction methods,the prediction accuracy,reliability,and computational cost are compared.It is found that the neural network model can achieve higher accuracy after simple adjustment of parameters on the premise of a large number of samples.