The Traffic Jam Prediction And Active Metering Method At Urban Expressway Weaving Area

The interweaving of urban expressway main line traffic flow and on-ramp traffic flow will lead to traffic congestion in the interweaving area of main line.Usually,traditional methodologies use the control of the entrance ramp flow to reduce the proportion of interlacing behavior to alleviate congestion.However,there is a defect in controlling hysteresis.To this end,an active control method for integrated congestion prediction and multiple on-ramp coordination is proposed.Weaving area is the medium for the traffic interaction between the urban road and the expressway.The interweaving of traffic flow between main line and on-ramp causes the descent of road capacity,which brings about traffic congestions happened easily.In order to alleviate traffic jam,the on-ramp metering methodology is adopted to reduce the proportion of interweaving.Traditional methods have the defect of control hysteresis and cannot make full use of the section capacity.To solve the problem,an active metering method for weaving area is proposed,which integrates congestion prediction and multi-ramp coordination.Aiming at accurately determining the traffic flow state in the weaving area and laying foundation for traffic congestion prediction method in weaving area based on deep learning.Considering the characteristics of the projection pursuit model,the traffic state coefficient is defined.By analyzing the numerical relationship of the evaluation of clustering effect and introducing the regularization coefficient,the expression of evaluation coefficient of clustering effect is established.The improved genetic algorithm and K-means clustering algorithm are applied to calculate the clustering center and the optimal projection direction.Using the optimal projection direction,the traffic flow parameters are transformed into traffic state coefficients,and the neighboring clustering centers are determined to obtain the corresponding traffic flow states.The new traffic state discrimination methodology overcomes the dependence of traditional methods on expert experience and solves the problem of information entropy weight method.The comparison experiment shows that the new method can accurately distinguish the traffic flow state at weaving area.The discriminant accuracy of new methodology is 96.63%,which is 5.58% and 7.01% higher than the discriminant accuracy of the neural network and decision tree algorithm,respectively.According to the method of state discrimination and the principle of statistical analysis,the capacity and critical occupancy of each state are obtained.Based on the new state discrimination method,the Convolutional Neural Networks model is applied,cross-entropy is used as cost function,Adam adaptive algorithm is selected to optimize the model,and Dropout regularization method is used to establish a traffic congestion prediction model at urban expressway weaving area based on deep learning.The experimental results show that the prediction accuracy of the model based on deep learning is 96%.Compared with the traditional neural network,the prediction accuracy is improved,the iteration convergence speed is higher,and the model can accurately predict the traffic flow state at urban expressway weaving area.This paper studies the causes of congestion at urban expressway weaving area,analyses the characteristics of classical weaving area control methods,integrates the traffic congestion prediction model at urban expressway weaving area based on deep learning,sets control heuristic conditions according to the predicted states at weaving area,and takes the capacity and critical occupancy of predicted states as control parameters.Active local and cooperative control strategies are proposed respectively,and the active on-ramp metering methodology at urban expressway weaving area is proposed.According to the actual investigation results,the traffic simulation model of Nanjing Kazimen Expressway is established by using VISSIM simulation software.Four kinds of simulation experiments are designed respectively: uncontrolled,ALINEA,BOTTLENECK and active control.The simulation results show that the active control method can make full use of capacity of the weaving area,improve the operation efficiency and reduce the probability of congestion at weaving area.According to the comprehensive evaluation,compared with ALINEA and BOTTLENECK,the average delay of active control methodology is reduced by 25.64% and 2.93%,the overall delay is reduced by 18.94% and 1.28%,and the average speed is increased by 15.11% and 3.11%,respectively.