Research On Proactive Distributed Signal Control Based On Stochastic Traffic Demand Prediction

The traffic control technology which uses signal control means to regulate traffic flow becomes an important way to alleviate traffic congestion and ensure the smooth operation of urban traffic.However,the traditional signal control system is usually based on the estimation or detection of traffic flow parameters to establish a signal timing scheme that adapts to different levels of traffic demand,that is,the timing control which adapts to the traffic demand in the large period,the induction control in the hour segment,and the adaptive signal control which adapts seamlessly to the traffic demand.With the improvement of traffic control requirements,adaptive signal control based on traffic prediction has become a developing trend,that is,proactive signal control.However,at present,the design of proactive control algorithm is usually based on a given distribution of traffic demand,which can not describe the stochastic characteristics of traffic parameters and reduce the accuracy of signal timing optimization.In addition,when the signal structure is optimized,it depends on the centralized control structure established by the complex model,and the computational complexity increases step by step in the large-scale signal control optimization,so it can not cope with the real-time changing traffic demand.Moreover,no matter how the traffic information is obtained,there must be a certain gap between the prediction and the actual data,and most of the existing signal control systems do not consider the effect of this error on the signal control effect.To overcome the limitation of the above-mentioned signal control system,in this paper,based on the premise of obtaining traffic parameters in real time by means of traffic detection,and starting from the stochastic traffic demand prediction.The input parameters of traffic signal timing optimization(queue length prediction)and the evaluation index of traffic signal timing optimization(delay prediction)are obtained in advance.Combined with the characteristics of max pressure(MP)distributed signal control and the advantages of feedback optimization of model predictive control(MPC),an proactive distributed signal optimization control system based on group-based MP signal strategy and model predictive control is established.The contributions of the paper mainly include the following aspects:1.To calculate the queue length by lane,the traffic flow ratio of each lane should be obtained in advance.Using the recursive circularity of Kalman filter theory,when we predict the traffic flow ratio of the lane,we call the traffic flow ratio data of all lanes in the first three moments.It overcomes the defect of using only the proportion of traffic flow in the first three moments of lane i in the previous method,and makes full use of historical data,which improves the prediction precision effectively.The method uses R language programming to obtain the initial value of state vector estimation by least square method,which is convenient for the fast convergence of filtering process,makes up for the shortage of given initial parameters in previous methods,and makes up for the reliability of prediction results.The result of the model is based on the field survey data of the north entrance of the South Qilin Road–Wenchang Street intersection in Qilin District,Qujing City,Yunnan Province,and compared with the traditional prediction method(traditional Kalman filter prediction method,single exponential smoothing method,quadratic exponential smoothing method and third-order moving average method).It can be seen from the prediction error of the proportion of traffic flow of the method in this paper.The prediction error of vehicle flow is not more than 3 vehicles,which is better than the other four prediction methods,and shows that the proposed model has high accuracy.2.Queue length is one of the important traffic evaluation indexes for traffic signal control at signalized intersections.Most previous researches focus on estimating queue length and can not predict queue length effectively.Firstly,we use the Robertson platoon discrete model to predict the arrival of vehicles at intervals of 5 seconds.Secondly,based on the Lightthill-Whitham-Richards(LWR)traffic wave theory,the queue length is predicted in real time.This method fully describes the arrival characteristics of cars and buses,and makes up for the shortcomings of the previous research on the queue length in a single vehicle environment.In addition,in order to predict the queue length of multi-lane at the same time,the above-mentioned method is integrated with the traffic flow ratio predicted by Kalman filter,and the lane-based queue length is predicted by real-time.The model is verified by the field survey data in Qujing City.The results show that the average error of queue length prediction in the model is less than 3 vehicles and has a good prediction accuracy.3.Based on the prediction of the lane-based queue length of homogeneous traffic flow,taking heterogeneous traffic flow environment as the research object,we first estimate the arrival of cars and buses according to the modified mixed platoon dispersion model(MPDM),which is based on the change of queue length.In addition,the heterogeneous traffic flow model(Qian model)is used to analyze the parameters of heterogeneous traffic flow,and the queue length is estimated in real time based on mixed traffic wave analysis.This method fully describes the relationship between the arrival of different turning traffic flows and IQA,and makes up for the deficiency of the hypothesis of uniform arrival in previous studies.The experimental results show that the proposed model has better accuracy and robustness than the single traffic flow LWR model(using equivalent car to capture the heterogeneity of traffic flow).The accuracy of the proposed model(MAE=2.05vehs)is higher than that of the single LWR flow model(MAE=3.05vehs)to capture heterogeneity.In addition,when verifying the queue length of homogeneous traffic flow and heterogeneous traffic flow,the field survey data of Qujing City and Kunming City are used respectively,and the prediction accuracy is satisfactory.It reflects that the model can adapt to the traffic flow environment of different cities and has strong reliability and robustness.4.Based on the real-time lane-based prediction of queue length,the real-time lane-based delay prediction method based on IQA is established,which effectively overcomes the difficulty of data acquisition when using IQA to calculate the delay.In the model,we use the LWR traffic wave theory to analyze the change of IQA in different traffic waves in detail,and describe the change of IQA when the remaining queue is unsaturated and saturated.The delay on the non-coordinated lane and the coordinated lane are calculated respectively.The validity of the model is verified by the field data of Qujing City and compared with Webster delay model and HCM 2010 IQA delay model.The results show that the model has good precision and strong robustness.Among them,the MAPE,predicted by this model is 63.02% and 75.09% lower than that of Webster delay model and HCM 2010 IQA delay model,respectively.More importantly,this model overcomes the limitation that the delay changes in the Webster model and the HCM 2010 IQA model increase with the increase of saturation,and calculates the vehicle delay according to the real-time arrival dynamics of the vehicle,which fully describes the stochastic characteristics of the vehicle arrival.5.Based on the real-time lane-based queue length and delay prediction model established in this paper,combined with the improved MP signal control strategy and the group-based MP adaptive signal optimization method,it maximizes the output of throughput rate and minimizes control delay.To further optimize the effect of signal control,based on the advantages of this paper in real-time prediction of queue length trajectory,the MPC carries on the rolling optimization to the control strategy based on the group-based MP signal optimization method.The results of model verification show that the delay of intersection 1 and intersection 2 after MPC embedding decreases by 13.47% and 15.35%,respectively.On the basis of single-intersection optimization,based on the delay model of coordinated lane and non-coordinated lane established in this paper,the offset is optimized,and the influence on control output before and after the MP split weight correction is compared.The queue overflow times of lane 4 and lane 10 of bottleneck section before and after optimization were reduced from 6 times and 9 times to 0 times,respectively.It is shown that the MP green signal ratio based on the length ratio of section queues can effectively prevent the occurrence of queue overflow.The hybrid MP split weight optimization of the ratio of queue length on segments can effectively prevent the occurrence of queue overflow and further alleviate the traffic congestion.