Research On Dynamic Response Control Technology Of Intersection Signal Based On Multivariate Data

Intersection signal control is a key area of road traffic management,and its intelligence degree is directly related to traffic safety,efficiency and travel experience of the public.At present there are still problems such as unreasonable signal timing,accumulation of red light blocking,and difficulty in large-scale implementation of induction/adaptive control,which indicates that there exists much room for improvement in signal intelligent control.In recent years,the development of new detection technologies such as intelligent video and multi-target radar makes it realizable.Meantime more plentiful micro-real-time operational data is accumulated.In addition,the continuous exploration and integration of new technologies such as big data,AI,and Internet of Things in traffic control bring an unprecedented technological innovation foundation and direction to the improvement of signal control intelligence.In view of the previous signal control technology and application bottlenecks,this thesis explores along the direction brought by the new technology,taking the non-supersaturated intersection as the research object.Based on the multiple traffic status data detected in real time by the new sensing means and the matching signal control characteristic parameter data,this thesis proposes a new path for intelligent signal control with the characteristics of "lane-level perception,phase-level control and periodic feedback",in order to actively adapt to the random change characteristics of the traffic flow queuing at the intersection and effectively improve the traffic efficiency and signal control real-time response performance.Following the path of the integration of perception and control,the thesis focuses on the key supporting technology points of each link.The methods are verified through practical application in actual intersections.The main research work and highlights of this thesis include:(1)Integrating the advantages of new technologies,the thesis proposes a new path of the intelligent control of intersection signal control based on multiple traffic perception and control data.Based on new technologies such as traffic perception,big data processing and deep learning,and iterative control,the thesis takes advantages of new sensing means,identifying the traffic state of each import lane in the intersections accurately,including release in green time and queuing in red time.With the support of network interaction perception and control technology,multi-data is integrated and matched.Then the deep learning algorithm develops the methods in extracting the queuing-release evolution characteristics and calibrating phase control feature parameters,realizing phase-level dynamic switching control according to the real-time change of lane-level traffic status.Furthermore,the iterative learning algorithm is used to construct dynamic response control logic and key technologies for periodic-level iterative feedback,and further to achieve the goal of improving the control efficiency with the least green light loss and the maximum release capacity.(2)Based on the integrated acquisition of multiple perception and control data,the extraction method of the evolution characteristics of the import lane-level vehicle flow release queue is studied and established.Based on the matching collection and output of multiple perception and control feature data,a lane-level vehicle traffic status recognition model is established based on the hidden Markov model.The key parameter learning method of the model and the global implicit variable decoding algorithm are given,accurately analyzing the four traffic status characteristics and state transformation rules,including saturation traffic,queue traffic,discrete passage and parking waiting.Based on the actual data comparison and analysis,the effectiveness of the model method is verified.It shows that the proposed model can accurately describe the evolution of the actual traffic state of the intersection entrance lane,and has higher recognition accuracy than the Gaussian mixture model.(3)By proposing the reference green light time,an optimal calibration method for establishing phase-level signal control characteristic parameters is studied.Through the analysis of the evolution characteristics of the traffic flow of the imported lanes,the concept of reference green light time is proposed,and the green light release process is divided into two stages: "minimum green-reference green" saturation flow and "reference green-maximum green" unsaturated flow;Through the analysis of the synergy relationship,the joint optimization model of signal control period and reference green light time based on double-layer planning is constructed.Meanwhile the solving method of the joint optimization model,on basis of particle swarm algorithm,and the calculation method of the fluctuation interval of control characteristic parameters are proposed;Based on the actual intersection operation data and the comparison analysis method of simulation,it is verified that,the proposed model method can effectively reduce the mean delay of each period and the standard deviation of the delay of the multi-day period.Meanwhile the balance between control efficiency promotion and multi-day robustness is improved as well.(4)Based on the online closed-loop optimization control target,the dynamic response control technology of signal with periodic-level iterative feedback is constructed.With the goal of minimizing traffic delay and green light waste,a dynamic response control logic architecture with multi-objective progressive optimization is designed,delimited according to reference green time.The iterative learning algorithm is applied to construct a feedback optimization method for key control parameters,generating the green light time of the present cycle according to the requirements and phases of the previous cycles;In terms of phase switching and dynamic adjustment,the calculation methods of the red light phase queuing switching threshold matrix and the green light phase front distance switching threshold are constructed respectively;After the end of the cycle execution,the closed-loop feedback optimization control is realized on basis of the evaluation of the benefits of period-level signal control;Through simulation with hardware connected to the application environment,the realizability of control logic and the effectiveness of the algorithm model are verified,which lays the foundation for the verification and effect evaluation of actual measurement.(5)Select the actual intersection to verify the application of signal dynamic response control.In order to verify the viable application and practical effect of the new path implementation method and key algorithm model,two intersections are selected to build a hardware support environment that matches the perception and control interaction;After accumulation and then analysis of the perception and control data,the dynamic response control characteristic parameters are generated,supporting the embeddedness of dynamic response control programs.The feasibility of practical application conversion is verified;By comparing and evaluating the control effects of different control schemes,the results show that the dynamic response control can well match the queuing and release requirements of each phase,significantly reduce the green light loss time,the number of stranded vehicles at the end of the green light,and the queue length of the vehicles to be released in green time.The lower the saturation and the greater the difference in flow between weeks,the more significant the control effect.