Lane Level Traffic Flow Prediction Based On Spatio-temporal Data Fusion

In recent years,China has embarked on the process of socialist modernization,the Party Central Committee has increased investment in the field of transportation construction,and the distance between urban and rural areas has been shortened to further drive the development of the urban economy,but the current urban road network construction is still in the development stage.With the rise of the car purchase boom,the speed of road traffic infrastructure construction is gradually unable to match the growth rate of motor vehicle ownership,and the problem of urban traffic congestion is becoming increasingly serious.Accurate and real-time traffic information prediction can sense and analyze the evolution trend of traffic data,which is an effective means for traffic management departments to relieve urban traffic pressure.Currently,traditional traffic flow prediction models lack comprehensive consideration of multiple factors affecting traffic flow,such as temporal and spatial characteristics,resulting in low accuracy and poor portability of existing traffic flow prediction models.Meanwhile,with the further development of in-vehicle navigation and driverless technology,the regional traffic flow prediction method with the road network as the research object can no longer match the demand of high-tech for accurate matching of road information,and a more fine-grained data modeling approach is needed,so the lane level traffic flow prediction is a necessary research object.The paper takes the cross-section of urban road traffic system as the research object,constructs the method system of traffic flow prediction based on parameter correlation,horizontal and vertical lane level spatial characteristics,and temporal characteristics,and provides the theoretical basis and basic technical route for analysis,modeling and prediction of lane level urban road traffic information perception.Firstly,the temporal and spatial characteristics of urban traffic flow at the lane level are analyzed in detail,the existing prediction methods and data fusion methods based on the temporal and spatial characteristics are introduced in detail,and the shortcomings of the existing methods are clarified.Secondly,based on the analysis of lane level transverse spatial characteristics,the accuracy of lane level traffic flow prediction is improved by studying the complex coupling relationship of traffic flow parameters among different lanes with joint multi-lane modeling;the integrated multi-lane modeling needs to consider the coordination of features among lanes and lane parameters,and the feature level fusion of different traffic parameters of strongly correlated lanes is performed by principal component analysis to establish the transverse correlation To address the problem of model accuracy degradation caused by insufficient extraction of temporal dependencies in the modeling of spatio-temporal relationships of transverse lanes,the encoder structure of Transformer is introduced to extract features for multiscale temporal dependencies in prediction,and to improve lane level traffic flow prediction by combining long and short time The accuracy of the lane level lateral spatial prediction results is improved by combining long and short time-dependent features.Finally,a lane-level traffic flow prediction scheme is constructed for horizontal and vertical spatial-temporal prediction.Based on the horizontal lane space prediction model,the bi-directional long-and short-term memory network is introduced to model and extract features from the longitudinal space of the road,and the horizontal and vertical spatial-temporal characteristics of the road are combined in a combined modeling approach,which expands the data sensing domain for lane level traffic flow prediction,explores the rich characteristics of traffic flow from multiple perspectives,and achieves a more accurate prediction modeling.Finally,the effectiveness of the proposed method is verified by actual road traffic flow data.This study is provided a systematic technical route for the lane level traffic flow prediction problem based on spatio-temporal characteristics,constructed a lane level traffic flow prediction model with high accuracy and portability,and provided an effective prediction method for the demand of high accuracy matching of roads for lane level intelligent navigation.