CN115512548B - Method and system for road side detection unit layout and road traffic sparse sensing - Google Patents

Method and system for road side detection unit layout and road traffic sparse sensing Download PDF

Info

Publication number
CN115512548B
CN115512548B CN202211463075.4A CN202211463075A CN115512548B CN 115512548 B CN115512548 B CN 115512548B CN 202211463075 A CN202211463075 A CN 202211463075A CN 115512548 B CN115512548 B CN 115512548B
Authority
CN
China
Prior art keywords
traffic
detector
data
real
network
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
CN202211463075.4A
Other languages
Chinese (zh)
Other versions
CN115512548A (en
Inventor
周海波
查曹怡
薛鉴哲
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.)
Nanjing University
Original Assignee
Nanjing University
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 Nanjing University filed Critical Nanjing University
Priority to CN202211463075.4A priority Critical patent/CN115512548B/en
Publication of CN115512548A publication Critical patent/CN115512548A/en
Application granted granted Critical
Publication of CN115512548B publication Critical patent/CN115512548B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/18—Network design, e.g. design based on topological or interconnect aspects of utility systems, piping, heating ventilation air conditioning [HVAC] or cabling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • Y—GENERAL 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
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • Analytical Chemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Hardware Design (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Optimization (AREA)
  • Medical Informatics (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pure & Applied Mathematics (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a method and a system for arranging road side detection units and sensing road traffic sparsity, and provides a low-cost solution for a modern urban traffic sensing system. Firstly, a complex urban road network system provided with a road sensor is modeled by using traffic flow simulation software, and the running data of all vehicles in the simulation and the monitoring data of a traffic detector are obtained according to the simulation running result; secondly, removing a part of traffic detectors in the urban road network, and recovering the traffic data of the removed part of the detectors on the basis of the existing detectors by using a sparse perception method; and finally, replacing the removed real detector in the original road network with the recovered traffic detector data as a virtual detector to obtain complete road network traffic detector data. The method can restore data of all detectors in the urban road network by using the sparse roadside detector, realizes a low-cost roadside detector arrangement scheme, and has high estimation precision.

Description

Method and system for arranging road side detection units and sensing road traffic sparsity
Technical Field
The invention belongs to the technical field of traffic perception, and relates to a low-cost road side detection unit arrangement and road traffic sparse perception method and system.
Background
The future urban development can not leave the modern intelligent traffic system, and the modern intelligent traffic system can not efficiently run and leave various road network data collected by the traffic detection system. Accurate and reliable traffic data can provide real-time urban road network conditions, help to plan vehicle routes and manage traffic lights, and the like. The detector system at present is the main means for acquiring traffic flow data in the urban road network. The more expensive detector equipment and the greater range of detectors typically deployed in the road network will result in more accurate and comprehensive collected traffic data. However, the urban road network is complicated in complexity, the road density is often large, and due to the limitation of physical conditions and budget expenditure, the traffic detectors cannot be densely covered in the whole urban road network. It becomes important to achieve low cost traffic sensing by properly laying out road detectors.
Through the search discovery of the existing literature, a great deal of research in the past focuses on the problem of optimizing the layout of detectors in urban road networks. A great deal of literature studies the effect of the location or spacing of traffic detectors on predicting various traffic performances. The document "The effects of detector spacing on travel time prediction from freeways" assesses The effect of detector spacing on The estimation of highway travel time by genetic algorithms. The document "Determining optimal sensor locations in free using genetic algorithm-based optimization" proposes a method of optimizing the location of sensors on a highway to improve the accuracy of time estimation of travel. The interest of these documents is to change the layout of sensors in the existing road network to obtain more traffic flow information. Documents "Data-driven estimation method for traffic Data in sectional units of road links" and "Multilayered LSTM with parameter transfer for vertical speed Data estimation (multilayer long-short-term memory network with parameter transfer for vehicle speed Data interpolation)" both recover sensor Data by a speed interpolation method, but only for existing sensors in a road network, and do not involve sensor links that are not laid out. However, what is needed in modern intelligent traffic systems is a low-cost traffic awareness scheme that recovers the complete traffic flow data of the entire road network from sparse detector data.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a low-cost road side detection unit arrangement and road traffic sparse sensing method and system, which can restore data of all detectors in a city road network by using sparse road side detectors and realize a low-cost road side detector arrangement scheme.
The technical scheme is as follows: in order to realize the purpose, the invention adopts the following technical scheme: a road side detection unit arrangement method based on traffic sparse perception comprises the following steps:
modeling an urban traffic network by using traffic flow simulation software, equipping a real traffic detector on each road section of the traffic network, and obtaining running data of all vehicles in simulation and monitoring data of the real traffic detectors according to a simulation running result;
removing a part of real traffic detectors in a traffic network, and recovering the removed traffic detector data on the basis of the traffic detector data which are not removed by using a traffic sparse perception method; the traffic sparse sensing method maps sparse traffic detector data to removed traffic detector data through a deep learning model;
and taking the recovered traffic detector data as monitoring data of a virtual traffic detector, replacing the removed real traffic detector in the original traffic network with the virtual traffic detector, and combining the real traffic detector and the virtual traffic detector to obtain a deployment mode of the sparse traffic detector in the traffic network.
Furthermore, the topological structure of the urban traffic network is abstracted from the city, and modeling is carried out according to the lengths of all roads in the city in a real map; the topological structure is defined as a directed graph, intersections in the directed graph serve as nodes, and roads serve as links.
Further, in traffic flow simulation software, dynamic distribution of traffic demands is carried out by setting source point-destination point (OD) matrixes, numerical values in the OD matrixes represent travel times between OD pairs in a simulation time period, and the OD matrixes of travel demands of different levels are respectively set for simulation.
Further, a complete traffic detector provided in a traffic network is composed of a real traffic detector and a virtual traffic detector, the complete traffic detector being defined as
Figure 492041DEST_PATH_IMAGE001
Each traffic detector
Figure 37423DEST_PATH_IMAGE002
Is shown as
Figure 932436DEST_PATH_IMAGE003
In which
Figure 978889DEST_PATH_IMAGE004
And
Figure 42791DEST_PATH_IMAGE005
respectively representing traffic detectors
Figure 149287DEST_PATH_IMAGE002
The latitude and longitude position of the mobile phone,
Figure 954432DEST_PATH_IMAGE006
traffic detector
Figure 319423DEST_PATH_IMAGE002
The captured instantaneous speed of the vehicle as it passes,
Figure DEST_PATH_IMAGE007
the time of the simulation is represented by,nis the total traffic detector count; definition ofXIs a sparse set of detector data consisting of a plurality of real traffic detectors,Yis a complete set of detector data; learning the mapping equation through a deep learning model:
Figure 300149DEST_PATH_IMAGE008
；Ga directed graph representing a traffic network topology is shown.
In specific implementation, the deep learning model can adopt a Transformer model, a long-short term memory network model or a full-link neural network model.
Preferably, the deep learning model adopts a Transformer model, position codes are embedded into the Transformer model, and a multi-head attention mechanism is used for learning the relation between data; and a linear transformation is used as a decoder, the obtained average speed data of each real traffic detector is input into a Transformer model, and the output result is an estimated value of the average speed of the virtual traffic detector.
Preferably, all real intersections of the Transformer model in a preset time step are adopted as the optimizationDetecting the average speed of the vehicle speed as input through a preset time window of the detector, and obtaining output through an encoder
Figure 515230DEST_PATH_IMAGE009
As input of decoder, linear conversion is carried out by decoder to obtain output
Figure 401146DEST_PATH_IMAGE010
As an estimate of the average speed of all virtual traffic detectors; wherein
Figure 727085DEST_PATH_IMAGE011
And
Figure 467377DEST_PATH_IMAGE012
are the parameter matrix and the offset vector of the linear transformation.
Preferably, real traffic detectors are randomly removed from a traffic network to obtain a sparsely deployed real traffic detector layout, traffic detector data at the removed real traffic detectors are estimated according to a traffic sparsity perception method, and the real traffic detector layout meeting set conditions including the number range of the real traffic detectors and the accuracy of recovered virtual traffic detector data is selected as a final deployment scheme.
Based on the method for arranging the roadside detection units, the invention provides a road traffic sparse sensing method, which comprises the following steps:
acquiring monitoring data of a real traffic detector in a real traffic network; the deployment mode of the real traffic detectors in the real traffic network is determined according to the deployment mode of the sparse traffic detectors combined with the real and the imaginary obtained by the road side detection unit deployment method based on traffic sparse perception;
and inputting the data of the real traffic detector into the trained deep learning model to obtain the data of the virtual traffic detector, thereby obtaining the complete data of the traffic detector.
The invention also provides a computer system which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the computer program realizes the steps of the road side detection unit arrangement method based on traffic sparse perception or the steps of the road traffic sparse perception method when being loaded to the processor.
Has the advantages that: compared with the prior art, the invention has the following advantages: the low-cost road side detection unit arrangement and road traffic sparse sensing method can estimate complete road traffic data by using sparse traffic detector data in a road network, and accurately recover data of a virtual traffic detector on the basis of sparse real traffic detector arrangement. The mechanism provided by the invention is applied to a complex urban road network existing in a real map for simulation, the effectiveness of the invention is verified by a simulation result, and the invention can provide a low-cost solution for an urban traffic perception system by combining a traffic flow simulation technology and a deep learning technology. Compared with a long-short term memory network algorithm model and a full-connection neural network algorithm model, the effectiveness and the superiority of the Transformer algorithm model are better, the real-time traffic condition of the whole urban road network can be monitored by using 35% of all real traffic detectors in the original road network, and the estimation precision of the traffic speed can reach 3m/s.
Drawings
Fig. 1 is a general flowchart of a method for laying roadside detection units according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of a low-cost traffic data estimation process on the sufoss network according to an embodiment of the present invention.
FIG. 3 is a comparison of the number of real detectors required to achieve the same target error under three algorithmic models in accordance with an embodiment of the present invention.
FIG. 4 is a general flowchart of a road traffic sparsity perception method according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the following detailed description of the embodiments of the present invention is provided with reference to the accompanying drawings: the embodiment is implemented on the premise of the technical scheme of the invention, and gives a detailed implementation mode and a specific operation process. It should be understood that the specific examples described herein are merely illustrative of the invention and that the scope of the invention is not limited to the examples described below.
As shown in fig. 1, in the roadside detection unit arrangement method based on traffic sparse sensing disclosed in the embodiment of the present invention, firstly, a traffic flow simulation software is used to model an urban traffic network, a real traffic detector is equipped on each road section of the traffic network, and according to a simulation operation result, driving data of all vehicles in the simulation and monitoring data of the real traffic detector are obtained; then removing a part of real traffic detectors in a traffic network, and recovering the removed traffic detector data on the basis of the traffic detector data which are not removed by using a traffic sparse perception method; and finally, taking the recovered traffic detector data as monitoring data of a virtual traffic detector, replacing the removed real traffic detector in the original traffic network with the virtual traffic detector, and combining the real traffic detector and the virtual traffic detector to obtain a deployment mode of the sparse traffic detector in the traffic network.
The topological structure of the urban traffic network in the embodiment of the invention is abstracted from the city, modeling is carried out according to the lengths of all roads in the city in a real map, and the urban traffic network can be properly simplified and corrected to obtain the applied urban traffic network; the topological structure of a urban traffic network is defined as a directed graph, intersections serve as nodes, and roads serve as links. The complete traffic detector consists of a real traffic detector and a virtual traffic detector, wherein the real traffic detector is an entity sensor arranged in the urban road network, and the virtual detector is monitoring data of a substitute detector estimated by an algorithm and is not a sensor actually existing in the urban road network. The traffic sparsity perception method in the embodiment of the invention maps sparse traffic detector data to removed traffic detector data through a deep learning model to serve as virtual traffic detector data.
FIG. 2 illustrates the Sulfris network-based network of the present embodimentA low cost traffic data estimation process. The experimental scenario in this embodiment is a road network structure in sufoss city, south dakota, usa, which is a classic urban road network model in the field of traffic research. Fig. 2 shows a left diagram of a network topology formed by abstracting a part of trunk roads from a real map of sufoss city in the united states, which is called a traffic network of sufoss city. The network comprises 24 intersections and 76 roads in total. Defining the road network structure in the left graph of FIG. 2 as a directed graph
Figure 977993DEST_PATH_IMAGE013
The intersection is used as a node, the road is used as a link, and the map is
Figure 492151DEST_PATH_IMAGE014
On behalf of the node(s) of which,
Figure 231568DEST_PATH_IMAGE015
representing a link. Wherein
Figure 577098DEST_PATH_IMAGE014
Can be expressed as
Figure 399561DEST_PATH_IMAGE016
，
Figure 509337DEST_PATH_IMAGE017
The number of the nodes at the intersection is,
Figure 504975DEST_PATH_IMAGE015
can be expressed as
Figure 642695DEST_PATH_IMAGE018
Wherein
Figure 245846DEST_PATH_IMAGE019
Representing a slave node
Figure 62492DEST_PATH_IMAGE020
To the node
Figure 268346DEST_PATH_IMAGE021
To simplify the model, the instruction in this embodiment
Figure 837736DEST_PATH_IMAGE022
，
Figure 126635DEST_PATH_IMAGE019
And
Figure 571523DEST_PATH_IMAGE023
respectively representing two links of the same length but opposite directions. The road network of the right graph of fig. 2 is a local road network graph of a certain area selected from the left graph of fig. 2, and is a typical schematic diagram showing the road detector thinning and recovery. The method for arranging a traffic detector for each link in a road network is a high-density traditional detector arrangement mode in the urban road network, and all traffic detectors in the road network are real entity detectors arranged in roads. The deployment mode of the sparse detector can be obtained through sparsification, namely, after a part of real detectors are removed. Based on the real detectors which are sparsely arranged, real-time traffic flow data at other positions in the road network where no detectors are arranged can be estimated through a sparse perception method, and the data can form virtual detectors in the road network. By combining the recovered virtual detector with the real detector in the original road network, a low-cost sparse detector deployment mode as shown in the right diagram of fig. 2 can be obtained. All of these detectors form a complete set of traffic detector nodes
Figure 456433DEST_PATH_IMAGE001
，nIs the total number of traffic detectors, the firstiA detector is defined as
Figure 896642DEST_PATH_IMAGE002
。
Figure 700650DEST_PATH_IMAGE002
Can be expressed as
Figure 537894DEST_PATH_IMAGE003
Wherein
Figure 85550DEST_PATH_IMAGE004
And
Figure 380265DEST_PATH_IMAGE005
respectively representing detectors
Figure 230540DEST_PATH_IMAGE002
The latitude and longitude position of the mobile phone,
Figure 181179DEST_PATH_IMAGE006
representing the instantaneous speed of the vehicle passing captured by the detector,trepresenting the simulation time.
The traffic flow speed is continuous and is influenced by factors such as road topology and road speed limit, so that the traffic state has space-time correlation in an urban road network. The attention mechanism in the deep learning can effectively capture the dynamic change of traffic flow data in time. Convolution structures in neural networks can extract their spatial features. Traffic flow information is data that can be measured. The full detector data of the right image of fig. 2 can be estimated from the sparse detector by a sparse sensing method. Definition ofXIs composed of a plurality of
Figure 860422DEST_PATH_IMAGE002
A composed set of sparse detector data is formed,Yis a complete set of detector data. By this mapping equation:
Figure 321228DEST_PATH_IMAGE008
the estimation of complete detector data based on sparse data can be realized, and the real-time traffic flow perception with low cost can be realized.
The Transformer model was introduced in 2017 by an article entitled "Attention is all you needed". The self-attention mechanism is a key concept of the model, and by introducing the self-attention mechanism, the Transformer model can learn the relation in the input, the relation in the output and the relation between the input and the output, and then respectively learn the three relations by using the multi-head attention mechanism.
The Transformer model consists of an encoder and a decoder, wherein the encoder has two sublayers, one is a multi-head attention layer, and the other is a full-connection network layer. Because the Transformer model does not adopt a recurrent neural network structure with sequential memory, the sequence of input information cannot be known. The input data then needs to be position coded first, which can be done by two formulas:
Figure 732617DEST_PATH_IMAGE024
Figure 967290DEST_PATH_IMAGE025
wherein
Figure 997694DEST_PATH_IMAGE026
In order to input the location of the data,
Figure 267001DEST_PATH_IMAGE027
is a dimension of the model that is,
Figure 849292DEST_PATH_IMAGE028
the dimensions of the even number are indicated,
Figure 86107DEST_PATH_IMAGE029
representing an odd dimension of
Figure 372732DEST_PATH_IMAGE030
And
Figure 247278DEST_PATH_IMAGE031
. By the two formulas, the system can be usedInput speed data is encoded into
Figure 593946DEST_PATH_IMAGE027
Vectors of dimensions, and thus order information of the original data.
The position encoded data is then input into a multi-head attention mechanism. The multi-head attention mechanism is composed of a plurality of time windows smoothed to obtain a self-attention mechanism, wherein the output of the self-attention mechanism can be expressed as the following formula:
Figure 341495DEST_PATH_IMAGE032
wherein the matrix
Figure 962969DEST_PATH_IMAGE033
、
Figure 692022DEST_PATH_IMAGE034
、
Figure 881695DEST_PATH_IMAGE035
Is derived from multiplying the input of the self-attention mechanism by a weight matrix,
Figure 374993DEST_PATH_IMAGE036
representation matrix
Figure 987109DEST_PATH_IMAGE033
、
Figure 492039DEST_PATH_IMAGE034
I.e. their vector dimensions. In the formula
Figure 711668DEST_PATH_IMAGE037
Is divided by
Figure 833208DEST_PATH_IMAGE036
The square root of (a) is to prevent the matrix inner product from being too large. When the long-distance interdependence characteristics are extracted, the cyclic neural network and the long-short term memory network need to be usedIt takes several time steps for information to accumulate to link the two. However, the self-attention mechanism can capture the remote interdependent features in the sequence data more easily, and directly represents the relation between two data through one calculation result in the calculation process, so that all features are effectively utilized.
The formula for the multi-head attention mechanism can be expressed as follows:
Figure 281638DEST_PATH_IMAGE038
wherein
Figure 906654DEST_PATH_IMAGE039
The neural network parameters for the multi-head attention mechanism,
Figure 500447DEST_PATH_IMAGE040
for the output of a single-head attention,hthe number of the single heads is the number of the single heads,
Figure 217605DEST_PATH_IMAGE041
the calculation formula is as follows:
Figure 859939DEST_PATH_IMAGE042
wherein
Figure 464095DEST_PATH_IMAGE043
、
Figure 979521DEST_PATH_IMAGE044
、
Figure 544495DEST_PATH_IMAGE045
Are respectively a matrix
Figure 583995DEST_PATH_IMAGE033
、
Figure 449183DEST_PATH_IMAGE034
、
Figure 430783DEST_PATH_IMAGE035
The weight matrix of (a). Then, a "sum and normalization" layer is input, where the network degradation is prevented by residual concatenation, and the input data is normalized to data with a mean of 0 and a variance of 1. The following formula for the fully connected network layer is as follows:
Figure 810949DEST_PATH_IMAGE046
after a further "sum and normalization" layer,
Figure 529506DEST_PATH_IMAGE047
、
Figure 124567DEST_PATH_IMAGE048
、
Figure 903167DEST_PATH_IMAGE049
、
Figure 567366DEST_PATH_IMAGE050
are the neural network parameters of that layer. Finally, a linear transformation is used as the decoder, the output of which
Figure 666778DEST_PATH_IMAGE051
The expression of (c) is:
Figure 975400DEST_PATH_IMAGE010
wherein
Figure 315114DEST_PATH_IMAGE011
、
Figure 76397DEST_PATH_IMAGE012
Are the neural network parameters of the linear transformation layer,
Figure 480965DEST_PATH_IMAGE009
is the output of the transform encoder, i.e., the input of the decoder. And inputting the obtained average speed data of each real traffic detector into a Transformer model to carry out the calculation of the process, wherein the output result is the estimated value of the speed of the virtual traffic detector.
In addition to the Transformer model, two deep learning models, namely a long-short term memory network and a fully-connected neural network, are used in the embodiment of the invention to estimate the traffic data of the detector. Both the long-short term memory network and the Transformer perform well in modeling sequence information. But compared with a long-term and short-term memory network, the Transformer can process a plurality of input data in parallel, thereby remarkably reducing the training time of the model and reducing the problem of model performance degradation caused by long-term dependence. The long-term and short-term memory network is obtained by continuously optimizing a full-connection neural network, a convolution neural network and a circulation neural network one step by one step, and solves the problems that the model parameters of the full-connection neural network are too much, the change on a time sequence cannot be modeled, the gradient disappears and the like.
Simulation experiment:
the invention carries out simulation verification on the road traffic perception method for low-cost estimation of road network detector data according to the flow shown in figure 1. The AIMSUN is traffic flow simulation software providing complete traffic analysis, and vehicle data simulation is performed by adopting the AIMSUN software in simulation. Each intersection in the urban road network is provided with a traffic signal lamp, a green light, a yellow light and a red light are sequentially displayed, a period of 90 seconds is taken as one period, and each road of the network is provided with a traffic detector. In order to simplify the model, all vehicle types are assumed to be private cars in the simulation, the same car following model is adopted for all the vehicles, traffic accidents do not occur, and road traffic is not influenced by factors such as weather, holidays and the like. The traffic demands are dynamically distributed by setting source point-destination (OD) matrixes, numerical values in the OD matrixes represent travel times between OD pairs in a simulation time period, and the OD matrixes in three conditions of low demand, medium demand and high demand are set in the experiment respectively.
During the simulation, the detectors on each road segment record the instantaneous speed of the passing vehicle every 0.8 seconds. After the simulation is finished, the AIMSUN outputs a file in an FZP format for storing data, wherein the file comprises information such as the index, the speed, the two-dimensional coordinates, the simulation time and the like of each vehicle. Firstly, screening all vehicle data acquired by each detector in simulation time according to the coordinate position of the detector. Then, a sliding window is used for calculating the average speed of the vehicle in the road section where the detector in a certain time slot is positioned, and the microscopic information is converted into macroscopic data. And then, performing sparsification treatment on all traffic detectors, namely randomly removing a part of detector data, then respectively recovering the removed detector data by using three different deep learning models on the basis of the residual detector data, and distributing the detector data serving as virtual detectors to a road network, so that a scheme for monitoring the real-time traffic condition of the urban road network at low cost can be obtained by using the sparse detectors.
In fig. 3, a comparison of the number of detectors required in the road network when three models are used to achieve equal estimation accuracy is obtained. Compared with a long-short-term memory network and a fully-connected neural network, the complete road network velocity data can be recovered by using a sparser detector by adopting a Transformer model. In the case of a total number of detectors of 30, the estimation accuracy can reach 3m/s by using 35% detectors, which shows that the transform model has feasibility, effectiveness and superiority.
Through the simulation experiment, the complete traffic flow information can be accurately estimated by using a transform deep learning model in a complex urban road network according to sparse detector data. When a real application road section is specifically designed, a Transformer model can be selected for virtual traffic detector data recovery, real traffic detectors are removed randomly to obtain sparsely deployed real traffic detector layout, then the Transformer model estimates virtual traffic detector data, and when the requirements of the number range of the real traffic detectors and the accuracy of virtual traffic detector data recovery (evaluation by indexes such as MSE) are met, a final sparse traffic detector deployment and deployment scheme combining virtuality and reality can be obtained.
As shown in fig. 4, after the road test detection units are arranged in the real traffic network according to the sparse traffic detector deployment scheme, a specific road traffic sparse sensing method is that after the monitoring data of the real traffic detector in the real traffic network is acquired, the data of the real traffic detector is input into the trained deep learning model to obtain the data of the virtual traffic detector, so that the complete data of the traffic detector is obtained. The deep learning model can adopt a model trained when a layout scheme is selected in a simulation environment, and can be further optimized if real data of the corresponding position of the virtual traffic detector can be acquired.
Based on the same inventive concept, the embodiment of the invention discloses a computer system, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the computer program realizes the steps of the road side detection unit arrangement method based on traffic sparse sensing or the steps of the road traffic sparse sensing method when being loaded to the processor.
It will be understood by those skilled in the art that the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer system (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. The storage medium includes: various media capable of storing computer programs, such as a U disk, a removable hard disk, a read only memory ROM, a random access memory RAM, a magnetic disk, or an optical disk.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (9)

1. A road side detection unit arrangement method based on traffic sparse perception is characterized by comprising the following steps:
modeling an urban traffic network by using traffic flow simulation software, equipping a real traffic detector on each road section of the traffic network, and obtaining running data of all vehicles in simulation and monitoring data of the real traffic detectors according to a simulation running result;
removing a part of real traffic detectors in a traffic network, and recovering the removed traffic detector data on the basis of the traffic detector data which are not removed by using a traffic sparse perception method; the traffic sparse sensing method maps sparse traffic detector data to removed traffic detector data through a deep learning model; the complete traffic detector equipped in traffic network is composed of real traffic detector and virtual traffic detector, and is defined byXIs a sparse set of detector data consisting of a plurality of real traffic detectors,Yis a complete set of detector data; learning the mapping equation through a deep learning model:
Figure DEST_PATH_IMAGE002
；Gis a directed graph representing the topological structure of a traffic network;
taking the recovered traffic detector data as monitoring data of a virtual traffic detector, replacing a removed real traffic detector in the original traffic network with the virtual traffic detector, and combining the real traffic detector and the virtual traffic detector to obtain a deployment mode of the sparse traffic detector in the traffic network;
the method comprises the steps of removing real traffic detectors from a traffic network randomly to obtain a sparsely deployed real traffic detector layout, estimating traffic detector data at the removed real traffic detectors according to a traffic sparsity perception method, and selecting the real traffic detector layout meeting set conditions as a final deployment scheme, wherein the set conditions comprise the number range of the real traffic detectors and the accuracy of recovered virtual traffic detector data.
2. The method for arranging the roadside detection units based on traffic sparse perception according to claim 1, wherein: the topological structure of the urban traffic network is abstracted from a city, and modeling is carried out according to the lengths of all roads in the city in a real map; the topological structure is defined as a directed graph, intersections serve as nodes in the directed graph, and roads serve as links.
3. The traffic sparsity perception-based roadside detection unit arrangement method as claimed in claim 1, wherein: in traffic flow simulation software, dynamic distribution is carried out on traffic demands by setting source point-destination OD matrixes, numerical values in the OD matrixes represent travel times between OD pairs in a simulation time period, and the OD matrixes with travel demands of different levels are respectively set for simulation.
4. The traffic sparsity perception-based roadside detection unit arrangement method as claimed in claim 1, wherein: a complete traffic detector is defined as
Figure DEST_PATH_IMAGE004
Each traffic detector
Figure DEST_PATH_IMAGE006
Is shown as
Figure DEST_PATH_IMAGE008
Wherein
Figure DEST_PATH_IMAGE010
And
Figure DEST_PATH_IMAGE012
respectively representing traffic detectors
Figure 857510DEST_PATH_IMAGE006
The latitude and longitude position of the mobile phone,
Figure DEST_PATH_IMAGE014
traffic detector
Figure 305809DEST_PATH_IMAGE006
The captured instantaneous speed of the vehicle as it passes,
Figure DEST_PATH_IMAGE016
the time of the simulation is represented by,nis the total traffic detector count.
5. The method for arranging the roadside detection units based on traffic sparse perception according to claim 1, wherein: the deep learning model adopts a Transformer model, a long-short term memory network model or a full-connection neural network model.
6. The method for arranging the roadside detection units based on traffic sparse perception according to claim 1, wherein: the deep learning model adopts a Transformer model, position codes are embedded into the Transformer model, and a multi-head attention mechanism is used for learning the relation between data; and a linear transformation is used as a decoder, the obtained average speed data of each real traffic detector is input into a Transformer model, and the output result is an estimated value of the average speed of the virtual traffic detector.
7. The method for arranging the roadside detection units based on the traffic sparse perception according to claim 6, wherein: the Transformer model takes the average speed of the vehicle speed detected by the preset time windows of all real traffic detectors with preset time step as input, and obtains output through an encoder
Figure DEST_PATH_IMAGE018
As input to a decoder, is linearly transformed by the decoder to an output->
Figure DEST_PATH_IMAGE020
As an average of all virtual traffic detectorsAn estimate of velocity; wherein->
Figure DEST_PATH_IMAGE022
And &>
Figure DEST_PATH_IMAGE024
Is a parameter matrix and an offset vector of the linear transformation.
8. A road traffic sparse sensing method is characterized by comprising the following steps:
acquiring monitoring data of a real traffic detector in a real traffic network; the deployment mode of real traffic detectors in the real traffic network is determined according to the deployment mode of the virtual-real combined sparse traffic detectors obtained by the traffic sparse perception-based road side detection unit deployment method according to any one of claims 1 to 7;
and inputting the real traffic detector data into the trained deep learning model to obtain virtual traffic detector data, thereby obtaining complete traffic detector data.
9. A computer system comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the computer program when loaded into the processor implements the steps of the traffic sparsity perception based roadside detection unit arrangement method according to any one of claims 1-7 or implements the steps of the road traffic sparsity perception method according to claim 8.
CN202211463075.4A 2022-11-22 2022-11-22 Method and system for road side detection unit layout and road traffic sparse sensing Active CN115512548B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211463075.4A CN115512548B (en) 2022-11-22 2022-11-22 Method and system for road side detection unit layout and road traffic sparse sensing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211463075.4A CN115512548B (en) 2022-11-22 2022-11-22 Method and system for road side detection unit layout and road traffic sparse sensing

Publications (2)

Publication Number Publication Date
CN115512548A CN115512548A (en) 2022-12-23
CN115512548B true CN115512548B (en) 2023-04-07

Family

ID=84514030

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211463075.4A Active CN115512548B (en) 2022-11-22 2022-11-22 Method and system for road side detection unit layout and road traffic sparse sensing

Country Status (1)

Country Link
CN (1) CN115512548B (en)

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104657199B (en) * 2015-02-10 2017-09-22 交通运输部科学研究院 The Forecasting Methodology of freeway traffic event coverage based on microscopic simulation
CN108595846A (en) * 2018-04-26 2018-09-28 苏州城方信息技术有限公司 The traffic simulation method docked with VISSIM based on SCATS
CN110895878B (en) * 2019-10-09 2020-10-30 浙江工业大学 Traffic state virtual detector generation method based on GE-GAN
CN114462233A (en) * 2022-01-18 2022-05-10 广州方纬智慧大脑研究开发有限公司 Microscopic traffic simulation method, computer device and storage medium

Also Published As

Publication number Publication date
CN115512548A (en) 2022-12-23

Similar Documents

Publication Publication Date Title
CN108629978B (en) Traffic track prediction method based on high-dimensional road network and recurrent neural network
CN111400620B (en) User trajectory position prediction method based on space-time embedded Self-orientation
CN110781266B (en) Urban perception data processing method based on time-space causal relationship
CN112633602B (en) Traffic congestion index prediction method and device based on GIS map information
CN116596151B (en) Traffic flow prediction method and computing device based on time-space diagram attention
CN114360239A (en) Traffic prediction method and system for multilayer space-time traffic knowledge map reconstruction
CN114925836A (en) Urban traffic flow reasoning method based on dynamic multi-view graph neural network
CN115099328A (en) Traffic flow prediction method, system, device and storage medium based on countermeasure network
CN115311860B (en) Online federal learning method of traffic flow prediction model
CN115346372A (en) Multi-component fusion traffic flow prediction method based on graph neural network
Lv et al. Digital twins based VR simulation for accident prevention of intelligent vehicle
CN116307152A (en) Traffic prediction method for space-time interactive dynamic graph attention network
CN116052427A (en) Inter-city inter-regional mobility prediction method and device based on private car travel track data
CN115862324A (en) Space-time synchronization graph convolution neural network for intelligent traffic and traffic prediction method
CN115457081A (en) Hierarchical fusion prediction method based on graph neural network
CN113971496A (en) Urban traffic network state evolution trend prediction method and system under influence of activities
CN117636626A (en) Heterogeneous map traffic prediction method and system for strengthening road peripheral space characteristics
CN115512548B (en) Method and system for road side detection unit layout and road traffic sparse sensing
CN113744541A (en) Road network discharge loss space-time distribution reconstruction method and system for confrontation graph convolution network
CN115984634A (en) Image detection method, apparatus, device, storage medium, and program product
Xia et al. Deeprailway: a deep learning system for forecasting railway traffic
CN116080681A (en) Zhou Chehang identification and track prediction method based on cyclic convolutional neural network
CN115905434A (en) Road network track completion method based on learning interpolation prediction
CN115563652A (en) Track embedding leakage prevention method and system
CN111371609B (en) Internet of vehicles communication prediction method based on deep learning

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