Research On Variable Speed Limit Control Of Urban Expressway In Mixed Traffic Flow Under Intelligent And Connected Environment

With the continuous growth of travel demand,traffic congestion and safety problems of urban roads are becoming increasingly serious,among which the urban expressway is particularly prominent.It is an important subject of urban traffic management to implement reasonable control means and make the expressway more efficient and safer for road users.Variable speed limit,as a key technology of highway management and control,has been gradually applied to urban expressway mainline control in recent years.By restricting the maximum speed of vehicles,a high-density,low-speed control zone is formed to delay the vehicles from entering the bottleneck area and reduce the speed variance,thus improving traffic efficiency and safety.However,due to the differences in the obedience of human drivers to variable speed limit control,the effect may not be as expected.With the continuous development of intelligent networking technology and automated driving,the road traffic flow will present a mixed operation state of Connected Automated Vehicle(CAV)and Humandriven Vehicle(HV)in the future,which provides a new environment for variable speed limit control.The CAV can interact with the roadside unit for real-time information via wireless communication technology.In this case,the speed limit value will no longer need to be mediated by variable message signs,but will be sent directly to the highly controllable CAV,allowing the HV to indirectly regulate the speed by following the CAV.Taking CAV as the direct control object can overcome the uncertainty of human drivers’ obedience to variable speed limit control and reduce the difference in driving behaviors.So that the loose coupling relationship between vehicles and control facilities can be transformed into a tight coupling relationship and the control utility of variable speed limit can be fully utilized.This paper takes the urban expressway as the study scenario,and regards CAV as the direct control object of the variable speed limit,to study the variable speed limit control problem of mixed traffic flow under an intelligent network environment.According to the analysis of the traffic capacity of mixed traffic flow and the influence mechanism of variable speed limit control on traffic flow,the overall framework of variable speed limit control including two phases of prediction and control is proposed based on the comprehensive consideration of the timeliness of control and the degree of refinement.In the prediction stage,a deep learning algorithm is introduced to construct a hybrid traffic flow prediction model,which extracts the spatial and temporal characteristics of the hybrid traffic flow and realizes the prediction of CAV and HV flow.It provides data input for the real-time control of the variable speed limit.In the control stage,according to the spatial scope of the control,the section-level and lanelevel variable speed limit control models are constructed respectively.By considering the three factors of traffic efficiency,traffic safety,and emission,the overall optimization goal of the model is determined.Then,the deep reinforcement learning algorithm is introduced to solve the optimal speed limit under different traffic states,which is then sent to the CAV to indirectly regulate the HV operation speed and realize the variable speed limit control of the mixed traffic flow.Finally,simulation experiments are designed to verify the effectiveness of the method,and the respective applicable conditions of section-level variable speed limit control and lane-level variable speed limit control are obtained through comprehensive analysis.The specific work is as follows:(1)Mixed traffic flow predictionAccording to the spatial extent,the mixed traffic flow prediction is divided into section-level prediction and lane-level prediction.Since the road section considers all lanes as a whole,the section-level prediction can be regarded as the prediction of one lane.Considering the spatio-temporal change characteristics of heterogeneous traffic flow in the horizontal and vertical spatial scopes,a lane-level prediction model for hybrid traffic flow based on combinatorial deep learning is constructed,which includes three modules: input feature selection,input feature attention calculation,and spatiotemporal information fusion.In the input feature selection module,the Dynamic Time Warping(DTW)algorithm is introduced to quantify the spatio-temporal correlation of the mixed traffic flow,and the spatio-temporal feature matrix is constructed based on the screening of the features that are highly correlated with the prediction target.In the input feature attention calculation module,the 1×1 convolution kernels are used to onedimensionally convolve each input feature,and the attention mechanism is introduced to quantify the difference in the influence of each input feature on the predicted target,so as to construct the spatio-temporal feature matrix with the attention weights.In the spatio-temporal information fusion module,the spatio-temporal feature matrix with the attention weights is inputted to the temporal convolutional neural network(TCN).The temporal and spatial features of the traffic flow in the horizontal and vertical spatial scales are captured,and the complex coupling relationship between the mixed traffic flows is learned,so as to realize the lane-level prediction of the mixed traffic flow.Based on the SUMO platform,a simulation environment for urban expressway mixed traffic flow is built,and three sets of comparison experiments are designed to verify the effectiveness of each module of the prediction model.Besides,the superiority of the proposed model is verified with respect to the comparison prediction model under different prediction steps and CAV penetration rates.(2)Mixed traffic flow section-level variable speed limit control methodThe Markov Decision Process(MDP)is used as the framework to model the mixed traffic flow section-level variable speed limit control problem.All the CAVs on the control roadway are considered to be a whole as a single agent.Then,the CAV flow,HV flow,average vehicle speed,and average time occupancy in different areas are collected by road sections as the variables to represent the traffic state,and the state space is constructed according to the current traffic state and the predicted traffic state.Considering the maximum speed limit of the urban expressway and the computation complexity,the speed limit value is discretized in 5km/h intervals to construct the action space.In addition,the overall reward function is established by considering the three indicators of average vehicle speed,time-exposed collision risk,and emissions.Moreover,the Dueling Double Deep Q Network with Prioritized Experience Replay(PER-D3QN)algorithm,is introduced to solve the optimal speed limit of the road section to obtain the maximum reward within the range of the action space for different traffic states,thereby sending to the CAV within the road section’s control area to realize variable speed limit control for mixed traffic flow.(3)Mixed traffic flow lane-level variable speed limit control methodThe section-level variable speed limit control method is further extended to refine the control scope from section to lane.The Decentralized Partially Observable Markov Decision Process(Dec-POMDP)is used as the framework to model the lane-level variable speed limit control problem for mixed traffic flow.For each lane,all CAVs in the control area are considered to be a whole as a single agent.Then,the CAV flow,HV flow,average vehicle speed,and average time occupancy are collected in different areas to represent the traffic state.Considering the differences in traffic conditions and the mutual influence of each lane,the partial observation state space of each agent and the global state space are constructed according to the current traffic state and the predicted traffic state,respectively.In addition,the united action space is constructed to realize the cooperative control of multi-agents through the united speed limit of each lane.The reward function is the same as that of the road segment-level variable speed limit control method,and the Q-mixing network(QMIX)algorithm is introduced to solve the optimal combination of speed limits for each lane to achieve the maximum reward in the range of the united action space under different traffic states,thereby sending to the CAVs within the control range of the lanes and realizing the lane-level variable speed limit control for the mixed traffic flow.(4)Comprehensive simulation experiment on variable speed limit control for mixed traffic flowA comprehensive simulation experiment is designed by combining the two phases of prediction and control to verify the effectiveness of the variable speed limit control method for the urban expressway under the intelligent mixed traffic flow environment,and the experimental effect is evaluated.According to the refinement degree of control,two groups of experiments are designed,respectively,for section-level variable speed limit control and lane-level variable speed limit control.For each group of experiments,three compared experiments are designed to explore the influence of different reward functions on the control effect.Besides,the effectiveness of the control algorithm and its superiority to the compared control algorithms are validated.Moreover,a sensitivity analysis is carried out on the control effect under the conditions of different CAV penetration rates to prove its applicability.In addition,a comprehensive comparative analysis of the effects of section-level variable speed limit control and lane-level variable speed limit control is carried out to explore the applicability conditions of the two control methods.The results show that the proposed variable speed limit control method for mixed traffic flow can significantly reduce the spatial and temporal range of congestion and improve the comprehensive traffic operation level.Furthermore,under different CAV penetration conditions,the applicability of section-level variable speed limit control is different from that of lane-level variable speed limit control.This paper focuses on the urban expressway under the environment of intelligent mixed traffic flow,to improve the comprehensive operation level of mixed traffic flow as the starting point,gives full play to the controllability and communicability advantages of CAV,and carries out the research on the variable speed limit control method.It provides a theoretical foundation and technical support for vehicles to drive efficiently,safely,and greenly on the urban expressway,which is of great value for perfecting the theoretical system of the intelligent transportation system and promoting the application of the intelligent and connected network and autonomous driving.