Rearch On Data Restoration Of Urban Traffic Flow And Short-term Traffic Flow Prediction

Urban traffic flow has many characteristics,such as nonlinear,self-organizing,spatiotemporal,random time-varying and periodic similarity,so the short-term traffic flow prediction technology is complex and difficult.Based on the analysis of urban traffic flow characteristics,this study focuses on the traffic flow preprocessing technology under different missing rate and sampling interval,short-term traffic flow combination prediction technology under different characteristics and sharing mode route recommendation technology.Therefore,it is a very important research topic that how to select the model with good prediction performance according to the characteristics of urban traffic flow to complete the short-term traffic flow prediction.The main innovative research outcomes are as follows:1.According to the periodic similarity,nonlinearity,volatility,spatiotemporal correlation and chaos characteristics of urban traffic flow,the dynamic fluctuation process of traffic flow is analyzed.Using the data quality evaluation standard to analyze the data set used in this research,there are some problems such as incomplete value,short-term missing value and abnormal value.On the basis of existing preprocessing technology,using the advantages of good prediction accuracy,suitable for processing time series and eliminating gradient disappearance of long short-term memory(LSTM)network,the memory gate in its structure is improved.Then the R-LSTM preprocessing repair model is proposed.By analyzing the repair performance under different missing rate and sampling interval,the results show that the data repair performance of R-LSTM model is better than other models.It is more suitable for dealing with the time series with dependence before and after the medium and long term,and the fitting degree of the repair problem data can reach more than97%.2.Based on the spatiotemporal characteristics of traffic flow,this research analyzes the spatiotemporal complexity of traffic flow data collection with different sampling intervals,and studies the temporal and spatial correlation between detectors at adjacent stations,which provides a theoretical basis for the study of short-term traffic flow prediction based on deep learning.In order to get the implicit space-time characteristics of traffic flow time series,C-C algorithm is used to calculate the delay time ?,G-P algorithm is used to calculate the embedded dimension m,and the maximum Lyapunov exponent based on small sample data is used to determine whether the traffic flow time series has chaos.The one-dimensional traffic flow time series is transformed into multi-dimensional traffic flow time series by phase space reconstruction.Combined with the high efficiency of gate recurrent unit(GRU),a short-term traffic flow prediction model based on PSR-GRU is proposed.By analyzing the prediction performance of the model under 5-minute,10-minute and15-minute sampling intervals,it is found that the prediction accuracy of the model is higher than that of other reference models,and it can capture the fluctuation trend before and after the peak time of data.The prediction accuracy of the model is more than 95% under 5-minute sampling interval,97% under 10-minute sampling interval and 94% under 15-minute sampling interval.3.According to the linear and nonlinear characteristics of traffic flow time series,the three-phase diagram is analyzed.Through the study of the day similarity of the traffic flow,the time index of the traffic flow from linear to nonlinear change is determined.The optimal prediction model of each part is found by using the idea of segment prediction.A combined model based on SARIMA-LSTM-XGBOOST is proposed.The linear part of traffic flow can be predicted by using the seasonal autoregynamic integrated moving average(SARIMA)model.The good nonlinear prediction performance of LSTM-XGBOOST based on deep learning is used to predict the nonlinear part.The combined model is tested by traffic flow data sets with different sampling intervals,upstream and downstream detector data sets with strong spatial correlation and taxi speed data sets.The results show that the prediction accuracy of the combined model is about 6% higher than that of other reference models,which can achieve high prediction in complex environment.4.The accurate prediction results can provide the basis for the route recommendation,but the existing taxi sharing mode does not reflect the problem of the driver and the passenger benefit together.In order to improve the taxi carrying rate and reduce the congestion of main roads,the cluster algorithm based on regional division is used to analyze the centralized stations of passengers getting on and off the taxis from the taxis trajectory data,and the station mapping is carried out with Baidu map.Taking the difference between the reduction of passenger travel cost and the increase of driver income as the objective function,combined with clustering method,a new genetic algorithm based on multi-objective improvement is proposed to realize the sharing route recommendation based on the shortest path and road network congestion level.The dissertation compares the process and cost of the passenger sharing by dynamic request.The results show that the method can optimize the driving route of the taxi,improve the efficiency of the taxi and balance the income and expenditure between the driver and the passenger.In this thesis,the problem of high precision repair of missing and abnormal values in data set is mainly solved.The characteristics of traffic flow time series under different sampling intervals are analyzed in detail,and the short-term traffic flow prediction method is proposed for different characteristics.The problem of short-term traffic flow prediction in a complicated and changeable environment is also solved.In addition,based on the characteristics of taxis trajectory data the traffic flow prediction results,the congestion level of the road network is divided.In order to improve the carrying rate and reducing traffic congestion of the main road,the multi-objective sharing mode and sharing pricing strategy are proposed.The research results have important theoretical basis and reference value for the analysis and application of urban traffic flow characteristics.