Research On Travel Time Short-term Prediction Of Urban Road Based On Feature Vector Correlation

Using the temporal and spatial variation of traffic flow to predict travel time can realize intelligent resource allocation and management of urban road traffic system.Traditional travel time prediction methods are largely based on mining the temporal and spatial correlations of traffic flow systems,that is,to study the temporal correlation between the current traffic flow and historical data of a target road segment in a single road segment or the spatial correlation within adjacent road segments.This method can effectively predict the regular components in the traffic flow system.However,the components of the urban road traffic system are complex and easily affected by various traffic events,weather factors,and signal control equipment at road intersections.The impact of these random components on the traffic system is increasing.This makes the prediction accuracy of urban road travel time based solely on correlation less accurate.In view of the limitations of existing methods,this paper proposes to add causal relationship analysis on the basis of correlation analysis to mine the potential correlation between travel time and traffic flow parameters,thereby improving the accuracy of prediction.In this paper,the automatic license plate recognition data is used as the basic data,and the specific research contents are as follows:The first is the data preparation stage,which analyzes the quality of the original data from three aspects: the amount of data captured by the device,the repetition rate,and the recognition rate,selects the data sections with better quality,and processes the noise data to obtain a complete and reliable data set.Secondly,the correlation analysis of the relevant traffic parameters in the traffic flow system is carried out,mainly from the two aspects of correlation and causality.Based on the correlation,the spatiotemporal feature vector is obtained from the time dimension and the space dimension;Based on the causal relationship,a convergent cross-mapping algorithm is proposed to detect the causal relationship between traffic flow parameters and travel time.And a complete set of algorithm process is proposed to obtain the causal relationship direction and influence strength of different traffic flow parameters and travel time,and through the extended convergent cross-mapping,the specific time delay that causes the change of the result is obtained.Then combine the causal feature variables with the spatiotemporal feature variables as the feature vector input for travel time prediction,four prediction algorithms are selected to compare and predict the travel time by time period.It is verified that the XGBoost prediction model based on feature selection is superior to other similar models in terms of interpretability and prediction accuracy.Its prediction error MAPE is between 5.39% and 11.42%;Finally,XGBoost is used to rank the importance of input variable features in different time periods,and the conclusion is drawn: Regardless of whether it is in the off-peak period or the peak period,the travel time of the forecast point lags behind the travel time by one period has the most significant impact on the travel time forecast.