CN113538910B - Self-adaptive full-chain urban area network signal control optimization method - Google Patents

Self-adaptive full-chain urban area network signal control optimization method Download PDF

Info

Publication number
CN113538910B
CN113538910B CN202110792525.3A CN202110792525A CN113538910B CN 113538910 B CN113538910 B CN 113538910B CN 202110792525 A CN202110792525 A CN 202110792525A CN 113538910 B CN113538910 B CN 113538910B
Authority
CN
China
Prior art keywords
traffic
optimization
traffic flow
network
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110792525.3A
Other languages
Chinese (zh)
Other versions
CN113538910A (en
Inventor
李丹丹
龚云海
肖峰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN202110792525.3A priority Critical patent/CN113538910B/en
Publication of CN113538910A publication Critical patent/CN113538910A/en
Application granted granted Critical
Publication of CN113538910B publication Critical patent/CN113538910B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/01Detecting movement of traffic to be counted or controlled
    • G08G1/0104Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125Traffic data processing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/044Recurrent networks, e.g. Hopfield networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/01Detecting movement of traffic to be counted or controlled
    • G08G1/0104Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125Traffic data processing
    • G08G1/0129Traffic data processing for creating historical data or processing based on historical data
    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/01Detecting movement of traffic to be counted or controlled
    • G08G1/0104Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0137Measuring and analyzing of parameters relative to traffic conditions for specific applications
    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/07Controlling traffic signals
    • G08G1/08Controlling traffic signals according to detected number or speed of vehicles
    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/07Controlling traffic signals
    • G08G1/081Plural intersections under common control
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02TCLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00Road transport of goods or passengers
    • Y02T10/10Internal combustion engine [ICE] based vehicles
    • Y02T10/40Engine management systems

Abstract

The invention belongs to the technical field of ITS intelligent traffic systems, and particularly relates to a self-adaptive full-chain urban area network signal control optimization method. The method comprises the steps of collecting traffic flow parameters by using a machine vision technology, predicting the traffic flow by using the obtained data based on a pre-trained traffic flow prediction algorithm, constructing a microscopic traffic simulation model according to the predicted traffic flow data, an original signal timing scheme and traffic network basic data, constructing a network-level signal optimization model, and actively optimizing the network-level signal optimization model by using a Bayesian optimization algorithm, thereby obtaining an optimal signal timing scheme of a target area network. The method has good integration, and forms an internal and external circulation feedback closed loop, and an internal and external circulation feedback mechanism can realize the interaction of the network signal optimization model and the microscopic traffic simulation model, and can ensure that an optimization result scheme is adaptive to the dynamic change of the external environment, thereby realizing instantaneous dynamic optimization and long-term steady state optimization.

Description

Self-adaptive full-chain urban area network signal control optimization method
Technical Field
The invention belongs to the technical field of an ITS intelligent traffic system, relates to the fields of traffic big data, traffic flow prediction and Bayesian optimization algorithm and computer vision and traffic network optimization technology, in particular to a self-adaptive traffic control technology, and specifically relates to a self-adaptive full-chain urban area network signal control optimization method.
Background
The urban road intersection is an important component of an urban road, is a key node for traffic management and control, and also becomes a key breakthrough for treating urban traffic congestion. Therefore, the research and the treatment on the intersection are continuous all the time, wherein the signal timing of the network traffic intersection is optimized, the large-scale reconstruction and expansion of the space resources of the intersection are not needed, and the method is an effective mode with low cost and quick effect. For example, Chen and the like construct a Simulation-Based large-scale network Signal optimization model in Simulant-Based Travel Time Reliable Signal Control, and optimize and solve by using a trust domain algorithm, so that the average network Travel Time is obviously reduced; liang et al in a statistical method to optimal timing genetic signal and timing plan at a signaled interaction using Connected Vehicle technology, optimize the phase duration and phase sequence of four left turn conflicting approach crossings using an intelligent tree search algorithm and various types of genetic algorithms to minimize the average delay of all vehicles. Gao et al used Meta-heuristic algorithms (such as harmony search and artificial bee colony) in Meta-hesitics for bi-object rban traffic light scheduling schemes to solve the optimization problem of the dual-target urban traffic signal, and the results show that the method is superior to the classical non-dominated sorting genetic algorithm. Kim and Sohn designs a depth map Q network in Area-wide traffic signal controlled based on a deep map Q-network (DGQN) trained in an asynchronous update method to effectively adapt to the space-time dependency of a large-scale network and improve the calculation efficiency, and successfully solves the problem of traffic signal optimization in a large-scale traffic network. Chu et al use a Multi-agent reinforcement learning method in Multi-agent depth requirement for large-scale traffic signal control to solve the problem of adaptive traffic signal control of a complex urban traffic network, thereby effectively reducing the average delay and the average queuing length of intersections.
The signal control optimization research at the present stage has the following problems: firstly, the development requirements of dynamic traffic cannot be met simultaneously; secondly, the existing research method cannot efficiently process the control optimization problem of large-scale network-level traffic signals; thirdly, the existing research mostly focuses on one point or two points of traffic data acquisition, traffic flow prediction and network signal optimization, the integration is poor, and a feedback closed loop cannot be formed.
Disclosure of Invention
In order to solve the problems, the invention provides a set of self-adaptive full-chain urban area network signal control optimization method, which comprises the steps of collecting traffic flow parameters by using a machine vision technology, predicting the traffic flow by using the obtained data based on a pre-trained traffic flow prediction algorithm, constructing a microscopic traffic simulation model according to the predicted traffic flow data, an original signal timing scheme and traffic network basic data, constructing a network-level signal optimization model, and actively optimizing the network-level signal optimization model by using a Bayesian optimization algorithm, thereby obtaining an optimal signal timing scheme of a target area network. The method can integrate traffic data acquisition, traffic flow prediction, microscopic traffic simulation and network signal optimization into an integrated optimization closed loop, realize internal circulation feedback optimization between microscopic traffic simulation and network signal optimization and external circulation feedback optimization between network signal optimization and traffic data acquisition, enable a network signal timing scheme to actively adapt to dynamic changes of traffic demands, and guarantee improvement of intersection traffic efficiency.
The technical scheme of the invention is as follows:
a self-adaptive full-chain urban area network signal control optimization method comprises the following steps:
step 1. traffic data acquisition
(1.1) obtaining traffic flow parameters based on machine vision technology
Suppose that the number of signalized intersections of the urban area network is k, and the number of entrances to the ith signalized intersection is m i Automatically processing the video stream data of all intersections in the target area in batches by using a machine vision technology to obtain traffic flow related parameters of the jth entrance lane of the ith signal control intersection as follows: number of detected target vehicles num _ veh ij Detecting different types of vehicle proportion type _ rate ij Turn ratio turn _ rate ij (ii) a The vehicle is divided into four types, namely a small vehicle, a medium vehicle, a large vehicle and a trailer, namely, the proportion of the four types of vehicles is represented by the proportion type _ rate of different types of vehicles; the turn ratio turn _ rate refers to the proportion of left-turning, straight-going and right-turning vehicles at the entrance lane of a certain intersection; wherein i ∈ [1, k ]],j∈[1,m i ]And i and j are integers. The machine vision technology mainly simulates the visual function of a human by using a computer, extracts target information from an image or a video stream of an objective object, processes and understands the target information, and finally realizes target identification, target detection and target tracking.
(1.2) intersection standard vehicle traffic flow calculation
The number of the detected target vehicles, the proportion of the detected vehicles of different types and the rotation number obtained in the step (1.1)The direction ratio is converted into a standard vehicle traffic flow of component flow direction. Calculating the number of vehicles with different steering directions, determining a conversion coefficient Convert _ coeffient of vehicles of corresponding types according to the planning specification of the international general urban road intersection, and converting the conversion coefficient Convert _ coeffient into the traffic flow of standard vehicles, thereby calculating and obtaining the standard traffic flow of different steering directions of different entrance roads of the intersection; the ith signal controls the standard traffic flow sum _ veh of the jth entrance way of the intersection in different directions ij From the number of detected target vehicles num _ veh ij Detecting different types of vehicle proportion type _ rate ij Turn ratio turn _ rate ij And the conversion factor Convert _ coeffient of the corresponding type of vehicle,
therefore, the standard traffic flow of different steering of all signal control intersections in the target area is obtained.
Step 2, traffic flow prediction
The traffic flow prediction is realized on the basis of the traffic data acquisition in the step 1, and the current and historical standard vehicle flow data are used for carrying out the traffic flow prediction.
(2.1) design traffic flow prediction algorithm
The traffic flow prediction algorithm is used to learn to obtain a function f that can predict the traffic flow V 'of the next time T' from the traffic flow observation data V of the current time T obtained from the conventional road sensors (video, radar, etc.) as an input of the microscopic traffic simulation model of step 3. Wherein the function f is learned as follows:
Figure GDA0003741638970000041
the traffic flow prediction algorithm is structurally divided into three parts: the first part is data preprocessing, mainly correcting error data and restoring partial missing data of a road section, and the processed data is directly used as the input of the second part; the second part is coupled with a hierarchical graph convolution module, a graph signal set is given, a self-learning adjacent matrix of each graph convolution layer utilizes a gated circulation unit to set space dynamic information, and then a full connection layer is connected to map a low-dimensional feature vector space to a high-dimensional vector space, wherein the dimension of the high-dimensional vector space is lower than that of an input space; and a third part, namely a long-time and short-time memory neural network module, which utilizes the higher-dimensional vector space obtained by the second part to carry out mining and aggregation on the affinity and periodic depth information of the traffic demand based on the long-time and short-time memory neural network, simultaneously realizes the deep fusion of the space dynamic information and the time correlation information, and then is connected with a full-connection layer to map the feature vector space obtained by the third part back to a target space, wherein the target space result is the predicted traffic flow. A specific traffic flow prediction algorithm structural framework is shown in fig. 1. In addition, a loss function is defined by using a mean square error method, and an error between a predicted value and a true value is described. Wherein, the affinity means that the traffic conditions in the recent time period are more relevant than the traffic conditions in the old time period; periodic is a pattern of periodic changes in traffic conditions over a period of time. The coupled hierarchical graph convolution network is a graph convolution architecture which has different adjacency matrixes at different layers, and all the adjacency matrixes can be self-learned in the training process; the framework adopts a layered coupling mechanism and can relate an upper-layer adjacent matrix with a lower-layer adjacent matrix; the convolution architecture is an end-to-end network, and output of a feature space is realized by integrating a hidden space state and a gate control cycle unit. The gate control circulation unit uses a gate control mechanism to control information such as input, memory and the like so as to make prediction at the current time step.
(2.2) traffic flow prediction algorithm training and testing
Training the traffic flow prediction algorithm in the step (2.1) to obtain a set of applicable hyper-parameter sets, which are as follows: dividing a traffic flow data set into a training set and a testing set by adopting historical traffic flow data in a complex traffic scene (peak, working day, non-working day and severe weather); carrying out traffic flow prediction algorithm training by using training set data to obtain an algorithm model parameter set when a storage loss function is minimum; and testing the trained traffic flow prediction algorithm by using a test set, and evaluating the real-time performance and the accuracy of the design algorithm by using the average data processing speed, the average absolute error and the root mean square error as test evaluation indexes. The evaluation is mainly to check the effect of the parameters obtained by training, and if the evaluation index is good, the set of hyper-parameter set is adopted; if the evaluation index is not good, retraining is needed until a set of good hyper-parameter sets is found.
Through algorithm training and testing, a set of good hyper-parameter sets, namely a complete traffic flow prediction algorithm, can be obtained, and then current and historical standard traffic flow can be input to directly obtain future standard traffic flow.
Step 3. micro traffic simulation model construction
Collecting and organizing traffic basic data required by microscopic traffic simulation, wherein the traffic basic data comprises network road basic data, traffic signal timing initial scheme data and road network traffic flow data. And (3) performing road section traffic flow distribution by using the standard vehicle flow data of the branch flow direction of each signalized control intersection obtained by predicting in the step (2). The collected and sorted network road basic data are imported into microscopic traffic simulation software (such as SUMO and VISSIM), and the acquired road section traffic flow distribution data are used in the simulation software to input road section traffic flow and the sorted traffic signal timing initial scheme data are used to set initial signal timing. And operating the established microscopic traffic simulation model to obtain traffic evaluation indexes (delay, queue, average travel time and the like).
Step 4, network signal optimization
And (4) aiming at the micro traffic simulation model constructed in the step (3), constructing a network signal control multi-objective optimization model corresponding to the micro traffic simulation model.
(4.1) constructing a network signal control multi-objective optimization model
And 3, in order to realize the improvement of traffic efficiency and the full utilization of traffic resources in the network, the traffic evaluation index in the step 3 is taken as an optimization target. The green signal ratio of each phase of the intersection is used as a decision variable, the variable limit condition reference signal controls a universal standard, and the signal period time, the phase structure and the phase display sequence of each intersection are preset. Aiming at a target area, a network traffic signal control optimization model is established by comprehensively utilizing a traffic flow theory, a traffic signal control principle, a traffic network analysis technology, an optimization theory and system engineering, and the network traffic signal optimization model is specifically as follows:
Figure GDA0003741638970000061
Figure GDA0003741638970000062
x low ≤x≤x up
wherein x is the green signal ratio of the intersection; p is an intersection signal period, a phase structure and a phase display sequence which are preset; g denotes the optimization objective function, g 1 ,…g z Representing all traffic evaluation indexes needing to be optimized, wherein z is the number of optimization targets; m is an intersection number and is an integer; n is a phase number; x is the number of low And x up Representing the upper and lower bounds of the split ratio, respectively.
(4.2) network signal control multi-objective optimization model solution
And (3) processing the network signal control optimization model constructed in the step (4.1) by adopting a simulation-based optimization algorithm, wherein the steps are as follows:
(4.2.1) random sampling using Latin hypercube sampling method to select initial set X ═ X 1 ,x 2 ,…,x d D is the number of samples;
(4.2.2) carrying out nonlinear regression fitting on the initial set by adopting a proxy model to optimize an objective function g (x);
(4.2.3) mean and variance obtained with the proxy model (i.e.
Figure GDA0003741638970000063
) Searching a next sampling point x 'through the optimizer, and substituting the next sampling point into the simulator to obtain an objective function value g (x');
(4.2.4) if the target is a single-target optimization function (namely, the parameter z in the step (4.1) is 1), directly updating the initial set to be X', and then updating the proxy model h; if the multi-objective optimization function is adopted (namely the parameter z in the step (4.1) is more than or equal to 2), updating the pareto front, updating the initial set to be X', and then updating the proxy model h;
(4.2.5) repeating the steps (4.2.1) - (4.2.4) in sequence until the termination requirement is met to obtain the optimal solution. If the network signal optimization function is a single-target optimization function, obtaining an optimal network signal optimization scheme; if the optimization function is a multi-objective optimization function, the pareto optimal leading edge is obtained, and a series of optimal network signal optimization schemes are obtained.
The system comprises an optimizer, a proxy model, a simulator, a micro simulation software, a terminal condition, iteration times and iteration time, wherein the optimizer adopts Bayesian optimization, the proxy model adopts a Gaussian process regression model, the simulator adopts urban traffic micro simulation software, and the terminal condition is set by a user according to actual requirements. Bayesian optimization is a high-efficiency global optimization machine learning algorithm, and the next search point is determined by using the information of the previously explored point and is used for solving the black box optimization problem. The pareto optimal leading edge is a pareto optimal solution set formed by all pareto optimal solutions, and the solutions form the pareto optimal leading edge or the pareto leading edge surface of the research problem through the mapping of an objective function; for a problem with two targets, its pareto optimal front is usually a stripline, while for multiple targets, its pareto optimal front is usually a hypersurface. The framework of the network traffic signal control multi-objective optimization model is shown in the following figure 2.
Step 5, internal and external circulation double feedback full chain optimization
And (4) after the network signal optimization scheme obtained in the step (4) is implemented, the traffic flow state in the network is changed, the traffic data acquisition is realized again through the step (1), and the step (2), the step (3) and the step (4) are repeated in sequence. In the process of repeated circulation, data are dynamically updated and accumulated, traffic flow prediction in the step 2 is more accurate, further, the optimization result in the step 4 is more accurate, instantaneous dynamic optimization or long-term steady state optimization is finally realized, and the optimal network signal timing scheme is obtained. The part integrates traffic data acquisition, traffic flow prediction, a microscopic traffic simulation model and a network traffic signal control model, and designs a full-chain urban area network signal optimization scheme by adopting an internal and external circulation double-feedback optimization mechanism. The internal circulation feedback optimization refers to feedback optimization inside a system between the microscopic traffic simulation module and the network signal optimization module, and the external circulation feedback optimization refers to feedback optimization between the traffic data acquisition module and the network signal optimization module, so that instantaneous dynamic optimization and long-term steady state optimization are realized, and finally, an optimal signal timing scheme of the urban traffic network based on external environment feedback is obtained.
The invention has the beneficial effects that:
the traffic data acquisition, the traffic flow prediction, the microscopic traffic simulation and the network signal optimization are integrated into a whole to form an internal and external feedback closed loop, so that a regional network signal timing scheme with a better effect can be obtained. At present, related research in China is mainly focused on one point or two points, and a feedback mechanism is difficult to form, so that the practical application of the feedback mechanism is limited. The invention has good integration, and forms an internal and external circulation feedback closed loop, the internal and external circulation feedback mechanism can realize the interaction of the network signal optimization model and the microscopic traffic simulation model, and can ensure that the optimization result scheme is adapted to the dynamic change of the external environment, thereby realizing the instantaneous dynamic optimization and the long-term steady state optimization. In addition, the optimization result and the optimization efficiency can be further improved by adopting a Bayesian optimization method.
Drawings
FIG. 1 is a structural framework diagram of a traffic flow prediction algorithm;
FIG. 2 is a framework diagram of a network signal control multi-objective optimization model;
FIG. 3 is a schematic view of an investigation region in an embodiment of the invention;
FIG. 4 is a graph of total delay time of a first experimental region as a function of iteration number;
FIG. 5 is a graph of total delay time of a second experimental region as a function of iteration number;
FIG. 6 is a graph of total delay time of a third experimental region as a function of iteration number;
fig. 7 is a graph of the minimum delay time value of three experiments as a function of the number of iterations.
Detailed Description
The following further describes the specific embodiments of the present invention with reference to the drawings and technical solutions.
In the embodiment, the best network signal timing scheme can be finally obtained by verifying the case by taking the urban area of the peony river city as the case and optimizing the case, and the scheme can greatly reduce the total delay time of the area. The method comprises the following specific steps:
1. basic data acquisition
The road network of the central urban area of the peony river city is selected as a research area (as shown in fig. 3), and the research area is 1600 meters long and 531 meters wide, and comprises 122 nodes and 232 edges. And the total number of the intersections is 53, and 20 main trunk road signalized intersections and 8 secondary trunk road signalized intersections are selected from the intersections.
The flow of each signal control intersection entrance road is obtained through the flow video of the peony river research area, and the traffic flow of the main road and the secondary road intersection is measured in the early peak period of 7:00-8: 00. And obtaining the standard traffic flow in the future short time by using a trained traffic flow prediction algorithm.
The signal timing scheme running in the early peak time period of 7:00-8:00 in a research area is obtained by coordinating with the traffic bureau in the city of the peony river, and the specific signal timing scheme is as follows:
table 1 network original signal timing scheme
Figure GDA0003741638970000091
Figure GDA0003741638970000101
2. Microscopic traffic simulation building
And obtaining basic data of a road network of a research area through Openstreetmap, and constructing a network simulation model by using urban micro traffic simulation software SUMO in combination with an original signal timing scheme and the predicted standard traffic flow of each intersection diversion direction. The simulation runtime length set in this embodiment is 2050 seconds, where the first 800 seconds is the unstable period of the simulation model, and the last 1250 seconds is the stable period of the model (i.e., the effective simulation time). And operating the simulation model, and recording and counting results of the effective simulation time period to obtain the total delay time of the region.
3. Network signal optimization model establishment
In the embodiment, the total regional vehicle delay is used as an objective function, the signal cycle time, the phase structure and the phase display sequence of each intersection are preset, and the green signal ratio is used as a decision variable. There are a total of 28 signalized intersections and 90 decision variables. And determining the limiting conditions of the obtained decision variables by referring to the Chinese signal control universal standard. The embodiment adopts a single-objective optimization function, mainly for simplifying the model to facilitate the verification of the method, and of course, the method provided by the patent is also applicable to the multi-objective optimization problem of network signal control. The network signal optimization model formula used in this embodiment is as follows:
Figure GDA0003741638970000102
Figure GDA0003741638970000111
x low ≤x≤x up
wherein x is the green signal ratio of the intersection; p is an intersection signal period, a phase structure and a phase display sequence which are preset; delay represents the total Delay time of the area; m is the number of the intersection, and m belongs to [1,28 ]]And m is an integer; n is a phase number; x is the number of low And x up Representing the upper and lower bounds of the split ratio, respectively.
4. Inner and outer loop double feedback optimization solution
The optimal signal timing scheme is obtained through the total vehicle delay in the internal and external circulation double-feedback full-chain optimization area, and the experimental results are shown in fig. 4, 5, 6 and 7. In order to verify the applicability of the method and ensure the reliability of the experiment, the embodiment performs 3 times of repeated experiments, and finally can find the optimal solution which is close to the optimal solution. Taking a third experiment as an example, the optimal signal timing scheme is shown in table 2 below by optimizing the total regional vehicle delay from the first 2988976 seconds to 2575134 seconds.
TABLE 2 optimal signal control timing scheme
Figure GDA0003741638970000112
Figure GDA0003741638970000121

Claims (1)

1. A self-adaptive full-chain urban area network signal control optimization method is characterized by comprising the following steps:
step 1. traffic data acquisition
(1.1) obtaining traffic flow parameters based on machine vision technology
Suppose that the number of signalized intersections of the urban area network is k, and the number of entrances to the ith signalized intersection is m i Automatically processing the video stream data of all intersections in the target area in batches by using a machine vision technology to obtain traffic flow related parameters of the jth entrance lane of the ith signal control intersection as follows: number of detected target vehicles num _ veh ij Detecting different types of vehicle proportion type _ rate ij Turn ratio turn _ rate ij (ii) a The vehicles are divided into four types, namely small vehicles, medium vehicles, large vehicles and trailer vehicles, namely the proportion type _ rate of the vehicles of different types indicates the proportion of the four types of vehicles; the turn ratio turn _ rate refers to the proportion of left-turning, straight-going, and right-turning vehicles at the approach lane at an intersection; wherein i ∈ [1, k ]],j∈[1,mi]I and j are integers; the machine vision technology mainly simulates the visual function of a human by using a computer, extracts target information from an image or a video stream of an objective object, processes and understands the target information, and finally realizes target identification, target detection and target tracking;
(1.2) intersection standard vehicle traffic flow calculation
The number of the detected target vehicles, the proportion of the detected vehicles of different types and the steering acquired in the step (1.1)The standard vehicle traffic flow of which the component flow direction needs to be converted; calculating the number of vehicles with different steering directions, determining a conversion coefficient Convert _ coeffient of the vehicles of corresponding types according to the planning specification of the international general urban road intersection, and converting the conversion coefficient Convert _ coeffient into the traffic flow of standard vehicles, thereby calculating the standard vehicle flow of different steering directions of different entrance roads of the intersection; the ith signal controls the standard traffic flow sum _ vehi of the jth entrance way of the intersection in different directions j From the number of detected target vehicles num _ veh ij Detecting different types of vehicle proportion type _ ratei j Turn ratio turn _ ratei j Calculating the conversion coefficient Convert _ coeffient of the corresponding type of vehicles so as to obtain the standard traffic flow of different steering of all signal control intersection entrances in the target area;
step 2, traffic flow prediction
The traffic flow prediction is realized on the basis of the traffic data acquisition in the step 1, and the current and historical standard vehicle flow data are used for the traffic flow prediction;
(2.1) design traffic flow prediction algorithm
The traffic flow prediction algorithm is used for learning to obtain a function f, and the function f can predict the traffic flow V 'of the next time T' from the traffic flow observation data V of the current time T obtained by the sensors on the previous roads to be used as the input of the microscopic traffic simulation model in the step 3; where the function f is learned as follows:
Figure FDA0003741638960000021
the traffic flow prediction algorithm is structurally divided into three parts: the first part is data preprocessing, mainly correcting error data and restoring partial missing data of a road section, and the processed data is directly used as the input of the second part; the second part is coupled with a hierarchical graph convolution module, a graph signal set is given, a self-learning adjacent matrix of each graph convolution layer utilizes a gated circulation unit to set space dynamic information, and then a full connection layer is connected to map a low-dimensional feature vector space to a high-dimensional vector space; a third part, namely a long-time and short-time memory neural network module, which utilizes the higher-dimensional vector space obtained by the second part to carry out mining and aggregation on the affinity and periodic depth information of the traffic demand based on the long-time and short-time memory neural network, simultaneously realizes the deep fusion of space dynamic information and time correlation information, and then is connected with a full-connection layer to map the feature vector space obtained by the third part back to a target space, wherein the target space result is the predicted traffic flow; in addition, a loss function is defined by adopting a mean square error method, and the error between a predicted value and a true value is described; wherein, the affinity means that the traffic conditions in the recent time period are more relevant than the traffic conditions in the old time period; the periodicity refers to a periodic change mode of the traffic condition within a certain time interval; the coupled hierarchical graph convolution network is a graph convolution framework which has different adjacency matrixes at different layers, and all the adjacency matrixes can be self-learned in the training process; the graph convolution architecture adopts a layered coupling mechanism and can associate an upper-layer adjacent matrix with a lower-layer adjacent matrix; the graph convolution architecture is an end-to-end network, and output of a feature space is realized by integrating a hidden space state with a gate control cycle unit; the gate control circulation unit uses a gate control mechanism to control input and memorize information and makes prediction at the current time step;
(2.2) traffic flow prediction algorithm training and testing
Training the traffic flow prediction algorithm in the step (2.1) to obtain a set of applicable hyper-parameter sets, which are as follows: dividing a traffic flow data set into a training set and a testing set by adopting historical traffic flow data in a complex traffic scene; carrying out traffic flow prediction algorithm training by using training set data to obtain an algorithm model parameter set when a storage loss function is minimum; testing the trained traffic flow prediction algorithm by using a test set, and evaluating the real-time performance and the accuracy of a design algorithm by using the average data processing speed, the average absolute error and the root-mean-square error as test evaluation indexes; the evaluation mainly comprises the steps of checking the effect of parameters obtained by training, and adopting the set of hyper-parameter set if the evaluation index is good; if the evaluation index is not good, retraining is needed until a set of good hyper-parameter set is found;
through algorithm training and testing, a set of good hyper-parameter sets, namely a complete traffic flow prediction algorithm, can be obtained, and then current and historical standard traffic flow can be input to directly obtain future standard traffic flow;
step 3. micro traffic simulation model construction
Collecting and organizing traffic basic data required by microscopic traffic simulation, wherein the traffic basic data comprises network road basic data, traffic signal timing initial scheme data and road network traffic flow data; performing road section traffic flow distribution by using the standard vehicle flow data of each signal control intersection diversion direction obtained by predicting in the step 2; importing the collected and sorted network road basic data into microscopic traffic simulation software, and using the obtained road section traffic flow distribution data to input road section traffic flow and using the sorted traffic signal timing initial scheme data to set initial signal timing in the simulation software; operating the established microscopic traffic simulation model to obtain a traffic evaluation index;
step 4, network signal optimization
Aiming at the micro traffic simulation model constructed in the step 3, constructing a network signal control multi-objective optimization model corresponding to the micro traffic simulation model;
(4.1) constructing a network signal control multi-objective optimization model
In order to realize the improvement of traffic efficiency and the full utilization of traffic resources in the network, the traffic evaluation index in the step 3 is taken as an optimization target; using the green signal ratio of each phase of the intersection as a decision variable, controlling a universal standard by a variable limiting condition reference signal, and presetting the signal cycle time, the phase structure and the phase display sequence of each intersection; aiming at a target area, a network traffic signal control optimization model is established, wherein the network traffic signal optimization model is specifically as follows:
Figure FDA0003741638960000041
Figure FDA0003741638960000042
x low ≤x≤x up
wherein x is the green signal ratio of the intersection; p is an intersection signal period, a phase structure and a phase display sequence which are preset; g denotes the optimization objective function, g 1 ,…g z Representing all traffic evaluation indexes needing to be optimized, wherein z is the number of optimization targets; m is an intersection number and is an integer; n is a phase number; x is the number of low And x up An upper and a lower bound representing the split ratio, respectively;
(4.2) network signal control multi-objective optimization model solution
(4.2.1) random sampling using Latin hypercube sampling method to select initial set X ═ X 1 ,x 2 ,…,x d D is the number of samples;
(4.2.2) carrying out nonlinear regression fitting on the initial set by adopting a proxy model to optimize an objective function g (x);
(4.2.3) mean h (x) and variance obtained with a surrogate model
Figure FDA0003741638960000043
Searching a next sampling point x 'through the optimizer, and substituting the next sampling point into the simulator to obtain an objective function value g (x');
(4.2.4) if the target is a single-target optimization function, namely the parameter z in the step (4.1) is 1, directly updating the initial set to be X', and then updating the proxy model h; if the function is a multi-objective optimization function, namely the parameter z in the step (4.1) is more than or equal to 2, updating the pareto front, updating the initial set to be X', and then updating the proxy model h;
(4.2.5) repeating the steps (4.2.1) - (4.2.4) in sequence until the termination requirement is met to obtain the optimal solution; if the network signal optimization function is a single-objective optimization function, obtaining an optimal network signal optimization scheme; if the network signal is a multi-objective optimization function, obtaining a pareto optimal leading edge, and obtaining a series of optimal network signal optimization schemes;
the system comprises an optimizer, a proxy model, a simulator, a city traffic microscopic simulation software and a user, wherein the optimizer adopts Bayesian optimization, the proxy model adopts a Gaussian process regression model, the simulator adopts city traffic microscopic simulation software, and the termination condition is set by the user according to actual requirements;
step 5, internal and external circulation double feedback full chain optimization
After the network signal optimization scheme obtained in the step 4 is implemented, the traffic flow state in the network changes, the traffic data acquisition is realized again through the step 1, and the step 2, the step 3 and the step 4 are repeated in sequence; in the process of repeated circulation, data are dynamically updated and accumulated, traffic flow prediction in the step 2 is more accurate, further, the optimization result in the step 4 is more accurate, instantaneous dynamic optimization or long-term steady state optimization is finally realized, and the optimal network signal timing scheme is obtained.
CN202110792525.3A 2021-07-14 2021-07-14 Self-adaptive full-chain urban area network signal control optimization method Active CN113538910B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110792525.3A CN113538910B (en) 2021-07-14 2021-07-14 Self-adaptive full-chain urban area network signal control optimization method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110792525.3A CN113538910B (en) 2021-07-14 2021-07-14 Self-adaptive full-chain urban area network signal control optimization method

Publications (2)

Publication Number Publication Date
CN113538910A CN113538910A (en) 2021-10-22
CN113538910B true CN113538910B (en) 2022-08-30

Family

ID=78098952

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110792525.3A Active CN113538910B (en) 2021-07-14 2021-07-14 Self-adaptive full-chain urban area network signal control optimization method

Country Status (1)

Country Link
CN (1) CN113538910B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113087030B (en) * 2021-03-02 2022-08-02 天津大学 Water resource double-circulation feedback control system and method
US20230185881A1 (en) * 2021-12-15 2023-06-15 International Business Machines Corporation Stepwise uncertainty-aware offline reinforcement learning under constraints
CN114419878B (en) * 2021-12-22 2023-04-07 银江技术股份有限公司 Method, electronic device and storage medium for predicting urban road network global traffic state
CN114333361B (en) * 2022-02-22 2022-06-28 南京慧尔视智能科技有限公司 Signal lamp timing method and device
CN114495506B (en) * 2022-02-23 2023-07-28 复旦大学 Multi-intersection signal lamp control system and method based on traffic flow prediction and reinforcement learning
CN115083173B (en) * 2022-04-27 2023-08-29 同济大学 Single-point timing signal control time interval dividing method based on electric alarm data
CN114973699A (en) * 2022-05-10 2022-08-30 阿波罗智联(北京)科技有限公司 Traffic control signal generation method, edge calculation unit and road side unit

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113112823A (en) * 2021-04-14 2021-07-13 吉林大学 Urban road network traffic signal control method based on MPC

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9076332B2 (en) * 2006-10-19 2015-07-07 Makor Issues And Rights Ltd. Multi-objective optimization for real time traffic light control and navigation systems for urban saturated networks
CN102360532B (en) * 2011-10-19 2014-02-19 太仓市同维电子有限公司 Crossing traffic signal control system and control method thereof
CN104575036B (en) * 2015-01-28 2017-01-04 重庆云途交通科技有限公司 Regional signal control method based on Dynamic OD volume forecasting Yu simulation optimization
CN105809958A (en) * 2016-03-29 2016-07-27 中国科学院深圳先进技术研究院 Traffic control method and system based on intersection group
CN106355885A (en) * 2016-11-24 2017-01-25 深圳市永达电子信息股份有限公司 Traffic signal dynamic control method and system based on big data analysis platform
CN110414747B (en) * 2019-08-08 2022-02-01 东北大学秦皇岛分校 Space-time long-short-term urban pedestrian flow prediction method based on deep learning
CN110580814B (en) * 2019-10-22 2020-11-24 北京航空航天大学 Timing method and device for traffic signal lamp
CN110751834B (en) * 2019-10-23 2020-10-27 长安大学 Method for optimizing signal timing of urban saturated intersection
CN112633500A (en) * 2021-01-18 2021-04-09 天津大学 Multi-objective optimization evolutionary computation method of convolutional neural network proxy model based on decomposition idea
CN113053120B (en) * 2021-03-19 2022-03-22 宁波亮控信息科技有限公司 Traffic signal lamp scheduling method and system based on iterative learning model predictive control
CN113112791A (en) * 2021-03-26 2021-07-13 华南理工大学 Traffic flow prediction method based on sliding window long-and-short term memory network

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113112823A (en) * 2021-04-14 2021-07-13 吉林大学 Urban road network traffic signal control method based on MPC

Also Published As

Publication number Publication date
CN113538910A (en) 2021-10-22

Similar Documents

Publication Publication Date Title
CN113538910B (en) Self-adaptive full-chain urban area network signal control optimization method
CN112265546B (en) Networked automobile speed prediction method based on time-space sequence information
CN112216108A (en) Traffic prediction method based on attribute-enhanced space-time graph convolution model
CN111860951A (en) Rail transit passenger flow prediction method based on dynamic hypergraph convolutional network
CN112767682A (en) Multi-scale traffic flow prediction method based on graph convolution neural network
Ou et al. Learn, assign, and search: real-time estimation of dynamic origin-destination flows using machine learning algorithms
CN107862877A (en) A kind of urban traffic signal fuzzy control method
Bao et al. Spatial–temporal complex graph convolution network for traffic flow prediction
Shao et al. The traffic flow prediction method using the incremental learning-based CNN-LTSM model: the solution of mobile application
CN114944053A (en) Traffic flow prediction method based on spatio-temporal hypergraph neural network
Liu et al. A method for short-term traffic flow forecasting based on GCN-LSTM
Wu et al. Traffic prediction based on GCN-LSTM model
CN116307152A (en) Traffic prediction method for space-time interactive dynamic graph attention network
Lu et al. Leveraging graph neural network with lstm for traffic speed prediction
Huang et al. Network-scale traffic signal control via multiagent reinforcement learning with deep spatiotemporal attentive network
Yu et al. 3D CNN-based accurate prediction for large-scale traffic flow
CN111899511A (en) Bus arrival time prediction method for AVL data of collinear line
CN114694382A (en) Dynamic one-way traffic control system based on Internet of vehicles environment
Zhang et al. Road section traffic flow prediction method based on the traffic factor state network
Zhang et al. Short-term Traffic Flow Prediction With Residual Graph Attention Network.
Zhang et al. Urban traffic flow prediction model based on BP artificial neural network in Beijing area
CN116612633A (en) Self-adaptive dynamic path planning method based on vehicle-road cooperative sensing
CN116665448A (en) Traffic speed real-time prediction method and system based on graph convolution network
CN115034478B (en) Traffic flow prediction method based on field self-adaption and knowledge migration
CN116486624A (en) Traffic flow prediction method and system based on space-time diagram convolutional neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant