Research On The Capacity Of Expressway Weaving Area Based On Machine Learning

The weaving area is a very important part of infrastructure in the urban expressway system,and is an important location for vehicles to make traffic flow changes.It is also a common bottleneck point of urban expressway system due to a large number of interweaving behaviors in the weaving area.The traffic capacity is an important parameter for the operation and management of the weaving area,but with the change of different combinations of flow values in the weaving area,there is an obvious non-linear trend of the traffic capacity values in the weaving area.Therefore,it is difficult to obtain high accuracy modeling results by using traditional statistical modeling techniques.As a result,this study has used the microscopic traffic simulation technology to carry out the simulation experimental work under the change of geometric conditions and traffic combination conditions to form the vehicle operation database in the weaving area.Next,machine learning methods are applied to explore the mechanism of influencing each factor of the weaving area capacity and to construct the capacity of the weaving area.Next,machine learning methods are applied to explore the influence mechanisms of each factor on the capacity of the weaving area,and the capacity model of the weaving area is constructed.The specific work and innovations can be concluded as follows:(1)Carry out a multi-mode combination of traffic flow operation data research in the weaving area.This study combines Li DAR,unmanned camera,and on-site manual research to obtain vehicle operation micro-behavior and traffic flow macro-operation parameter data in the weaving area,laying the data foundation for the overall research of the paper.This part focuses on applying LIDAR sensing facilities to collect vehicle micro-operation data and completes the processing of radar data through Python,specifically including: point cloud data pre-processing,trajectory point extraction,and trajectory tracking.After the extraction completed,the vehicle running track data has been formed.(2)Complete the study on the calibration of the parameters of the microscopic simulation model of the weaving area.The paper constructs the simulation model of the weaving area with VISSIM as the simulation platform.Firstly,the sensitivity of the simulation model parameters in the weaving area is analyzed by using single factor analysis,the decision tree method,and the BP-MIV method to clarify the process of the channel change behavior parameters in the simulation platform.And then,the calibration study of the microscopic simulation model parameters is conducted for global and local parameters.For the global parameters,the velocity-acceleration correspondence curves extracted from Li DAR data are used for setting;for the local parameters,the intelligent optimization-seeking calibration is performed based on the velocity and delay data extracted from the dataset.And for more accurate screening of the subsequent calibration results,an iterative optimization method of a genetic algorithm is proposed,which integrates clustering into each iteration.By comparing the simulation results of the calibrated model,it is verified that the accuracy of the simulation model can be increased by the improved genetic algorithm.(3)Microsimulation experiment based on the analysis of the factors affecting the capacity of the weaving area.Applying the calibrated weaving area simulation model,an experimental simulation of the weaving area is designed to generate the weaving area operation data and extract the capacity values under different combinations of conditions using the smoothing time window method.Then,the influence of the length of the interleaving zone,inbound and outbound traffic on the capacity is analyzed.It is verified that the capacity of the weaving area is not equal when the combinations of inbound and outbound flows are different under the same interleaving flow conditions.Moreover,it is verified that the influence of each factor in interleaved area on traffic capacity is not linear,and the SHAP model is also used to analyze the different influencing factors of the weaving area in detail.(4)The neural network is applied to construct the weaving area capacity model.Since the influence of the factors in the weaving area is not linear,the traditional model is difficult to apply.Therefore,the BP neural network is selected to build the capacity model of the weaving area.The specific structure of the model is determined by taking the influencing factors as input variables and the capacity as output variables.The influence of road conditions and traffic conditions on the capacity of the weaving area is modeled.The corresponding model accuracy is achieved through continuous iterative training.And by comparing and analyzing it with the traditional model,it is verified that the prediction accuracy of the model established in this paper is higher.The accuracy of the capacity model established by the neural network is verified.In summary,this study obtains the vehicle operation database of the weaving area based on microsimulation technology,develops the analysis of the traffic operation characteristics of the weaving area,and finally establishes the capacity model through different traffic flow influencing factors,which improves the accuracy of the capacity model of the weaving area.