Date Driven Method For Urban Traffic Environment Problems Analysis

Urban traffic environment generally contains factors as traffic emissions,transportation safety,and traffic policy.Meanwhile,along with the highly rapid development of mobile communication technology,sensor technique,the data involved with urban traffic environment are becoming more integral and precise,and in the meantime,which giving birth to new mode of transportation such as free-floating car sharing(FFT).Considering the background depicted above,this paper took generalized traffic environment as the object of study,the data-driven method was used in solving problems such as emissions,safety and FFT travel modeling,then took step further adopting method as from point to surface combined with macro-micro data coupling to study on traffic policy impact analysis considering FFT.First of all,this paper describes the way of getting multi-source data includes traffic emissions,traffic safety and travel survey.Then,a fusion analysis method of multi-source data sets including missing data completion,outlier detection and processing is given on the basis of traffic environment analysis.The missing data completion mainly includes average value,maximum expectation and data enhancement method,while the outlier detection and processing contains three times standard deviation and box diagram test.On the basis of data fusion technique,the paper puts forward a set of analysis method of traffic safety environment research based on deep learning.Firstly,the original data samples are fused and processed.Secondly,different network training methods and structures are compared and analyzed to determine the optimal network structure.Then,the improved meta heuristic optimization algorithm is used to improve the search efficiency and accuracy of the learning network solution.Finally,the key factors affecting the research site are accurately identified according to factor analysis and importance ranking algorithm.Based on the analysis of the freeway accident data set in Seattle,Washington,USA,it is found that the factors with relative importance more than 50% are vehicle age,driver age,accident type,month,accident location type,road function and light conditions,among which vehicle age and driver age are the two most important factors.Next,this paper builds a data-driven traffic pollutant emission analysis model,including the emission analysis model represented by nitrogen oxides(NOx)and carbon monoxide,and the traffic noise level analysis model based on dynamic traffic parameters.The vehicle exhaust emission model divides the data variables which can be collected into different subsets.By judging the accuracy of the network,it can quickly identify the key variables that contribute most in the study,which improves the model efficiency.In the meantime,the traffic noise model combines the static noise emission calculation process with the dynamic environmental parameter which improves the model accuracy.The data of 20 locations of an expressway in Seattle,Washington,USA,with a total of 200 days are selected to carry out the empirical vehicle exhaust emission analysis,three stake points with two observation points within an hour of an expressway in Seattle,Washington are selected as the test objects.Then,in order to predict the travel demand and travel distance of a new type of transportation mode,the deep learning time-series model(LSTM-RNN)is proposed.Firstly,the data processing and fusion methods are used to make a preliminary visual analysis using the original data,and the overall demand characteristics of the travel mode are analyzed from the two dimensions of time and space;secondly,the optimal efficiency of the algorithm is determined by the combination test of different network considering the hyper-parameters,finally,the comparative analysis results of different time series models are used as model accuracy testing process.This paper also makes an empirical study on 78 days order data of car2 go company in Seattle.Finally,using system dynamics as the research method,this paper takes the comprehensive traffic environment as core research object,the research break the limitations of general macro numerical analysis in traffic policy analysis,and takes into account of micro data in the research results of pollutant emission,traffic safety environment and FFT travel demand and build a comprehensive policy evaluation platform.This paper carries out an empirical analysis of traffic policy evaluation taking Seattle as an example.