CN114743374A - Multi-element traffic flow prediction method based on graph network - Google Patents

Multi-element traffic flow prediction method based on graph network Download PDF

Info

Publication number
CN114743374A
CN114743374A CN202210349264.2A CN202210349264A CN114743374A CN 114743374 A CN114743374 A CN 114743374A CN 202210349264 A CN202210349264 A CN 202210349264A CN 114743374 A CN114743374 A CN 114743374A
Authority
CN
China
Prior art keywords
traffic flow
graph
network
multivariate
road
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.)
Granted
Application number
CN202210349264.2A
Other languages
Chinese (zh)
Other versions
CN114743374B (en
Inventor
黄雷
赵灿
李志恒
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.)
Shenzhen International Graduate School of Tsinghua University
Original Assignee
Shenzhen International Graduate School of Tsinghua 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 Shenzhen International Graduate School of Tsinghua University filed Critical Shenzhen International Graduate School of Tsinghua University
Priority to CN202210349264.2A priority Critical patent/CN114743374B/en
Publication of CN114743374A publication Critical patent/CN114743374A/en
Application granted granted Critical
Publication of CN114743374B publication Critical patent/CN114743374B/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
    • G08G1/0125—Traffic data processing
    • G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Human Resources & Organizations (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Strategic Management (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Analytical Chemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a multivariate traffic flow prediction method based on a graph network, which comprises the following steps: s1, acquiring multivariate traffic flow historical data for traffic flow prediction; s2, constructing an independent dynamic relation graph for each traffic flow attribute based on the multivariate traffic flow historical data; s3, fusing the dynamic relationship graphs of the traffic flow attributes and converting the fused dynamic relationship graphs into a fused dynamic relationship graph network with the attributes mutually linked; s4, based on the fusion dynamic relational graph network, learning potential space dependency relationships between traffic flow attributes and between road sections by using a graph attention network, extracting space dependency characteristics, and constructing a multi-element traffic flow prediction model based on the graph network; s5, training the multi-element traffic flow prediction model based on the graph network; and S6, carrying out short-time traffic flow prediction according to the trained multi-element traffic flow prediction model.

Description

Multi-element traffic flow prediction method based on graph network
Technical Field
The invention relates to the technical field of deep learning, in particular to a multivariate traffic flow prediction method based on a graph network.
Background
The traffic flow prediction is a key problem in the construction of an intelligent traffic system, and the accurate traffic flow prediction has important practical significance for improving the road co-operation efficiency and promoting energy conservation and emission reduction.
At present, a plurality of traffic flow prediction methods are proposed by a plurality of researchers, and the prediction method mainly comprises the step of researching the traffic flow prediction problem from the aspects of time characteristics, space-time dependence and the like of the traffic flow.
The main characteristic attributes of the traffic flow comprise three aspects of traffic flow, average speed and road occupancy rate, which are closely related and different, and most of the current traffic flow prediction methods predict one of the attributes and do not reflect the overall condition of the traffic flow comprehensively.
Traffic flow in the real world often exhibits a high dependency between multiple attributes (flow, speed, etc.), multiple variables (multiple road segments in the road network). Some prediction methods rely on deep learning methods of complex structures, package learning is performed on interaction between multiple attributes and multiple variables by combining multiple network structures, complex space-time characteristics of traffic flow are mined, potential dependency relationships between the attributes and the variables are difficult to fully utilize, and the structures of deep learning prediction networks are often complicated.
Most of the existing methods are difficult to learn the potential spatial dependency between traffic flow attributes and road sections completely and autonomously, if a good prediction effect is to be achieved, the spatial dependency characteristics are often acquired better according to the prior knowledge of the existing spatial structure or domain experts, for example, by means of the prior road network structure, and the prediction algorithm for model autonomous learning of the road section spatial dependency is few under the condition of no prior knowledge.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a multivariate traffic flow prediction method based on a graph network, which can realize more accurate prediction of short-time traffic flow by independently learning the spatial dependence between traffic flow attributes and road sections of multivariate traffic flow by a model under the condition of no prior knowledge.
A multivariate traffic flow prediction method based on a graph network comprises the following steps: s1, acquiring multivariate traffic flow historical data for traffic flow prediction; s2, constructing an independent dynamic relation graph for each traffic flow attribute based on the multi-element traffic flow historical data; s3, fusing the dynamic relationship graphs of the traffic flow attributes and converting the fused dynamic relationship graphs into a fused dynamic relationship graph network with the attributes mutually linked; s4, based on the fusion dynamic relational graph network, learning potential space dependency relationships among traffic flow attributes and road sections by using a graph attention network, extracting space dependency characteristics, and constructing a multi-element traffic flow prediction model based on the graph network; s5, training the multi-element traffic flow prediction model based on the graph network; and S6, carrying out short-time traffic flow prediction according to the trained multi-element traffic flow prediction model.
Further, in step S1, the acquiring the multivariate traffic flow history data includes: obtaining time interval [1, T ] of all road sections in road network]The internal multivariate traffic flow historical data is represented by a time sequence X ═ X1,X2,…,XT]Represents; wherein the content of the first and second substances,
Figure BDA0003578728440000021
the traffic, average speed and road section occupancy of all road sections in the road network at the time T are shown, and T belongs to [1, T ∈]And n represents the number of links.
Further, step S1 further includes preprocessing the multivariate traffic flow history data as follows: extracting multivariate traffic flow historical data at t moment
Figure BDA0003578728440000022
As input, the actual value of the traffic flow characteristic of each road section at the moment t +1
Figure BDA0003578728440000023
As a label.
Further, step S2 includes: aiming at three traffic flow attributes of flow, average speed and road section occupancy rate, a self-attention mechanism is respectively used for learning the mutual adjacency relation among road section nodes, the attention score between two road section nodes is used as the weight of a connecting edge between the nodes in a dynamic relation graph, and the independent dynamic relation graph of each traffic flow attribute is obtained.
Further, step S2 includes: taking the road sections as nodes of the dynamic relationship graph, and calculating the attention scores of any two road section nodes by the following method:
Figure BDA0003578728440000024
Figure BDA0003578728440000025
Figure BDA0003578728440000026
wherein the content of the first and second substances,
Figure BDA0003578728440000027
representing a link node NiAnd NjAttention points on traffic, average speed and road occupancy, i, j ═ 1,2, …, n;
Figure BDA0003578728440000028
and
Figure BDA0003578728440000029
all are matrixes used for linear transformation and also learnable parameters of the model, and the dimensionalities of the matrixes are dk，·TRepresenting a matrix transposition;
Figure BDA0003578728440000031
respectively representing road section nodes NjThe flow rate at time t, the average speed and the road segment occupancy,
Figure BDA0003578728440000032
respectively representing road section nodes NiFlow, average speed and road segment occupancy at time t; the attention score between two road section nodes is used as the weight of the connecting edge between the nodes in the dynamic relationship graph,obtaining a dynamic relation graph of the flow, the average speed and the road section occupancy rate, and respectively recording as: gq,Gk,Gv；
Figure BDA0003578728440000033
I.e. representing link nodes NiAnd NjIn-flow dynamic relationship graph GqThe weight of the middle-connected edge is,
Figure BDA0003578728440000034
i.e. representing link nodes NiAnd NjIn average velocity dynamics graph GkThe weight of the middle-connected edge is,
Figure BDA0003578728440000035
i.e. representing link nodes NiAnd NjDynamic relation graph G of road section occupancyvThe weight of the middle connecting edge.
Further, step S3 includes: for any two road segment nodes NiAnd NjAccording to
Figure BDA0003578728440000036
Calculating two road section nodes NiAnd NjThe adjacent relation in the network of the converged dynamic relational graph is obtained; the calculation formula of the adjacency relation is as follows:
Figure BDA0003578728440000037
wherein the content of the first and second substances,
Figure BDA0003578728440000038
representing road segment nodes N in a converged dynamic relationship graph networkiAnd NjThe adjacent relationship of (a) to (b),
Figure BDA0003578728440000039
a value of 0 represents a link node N in the converged dynamic relationship graph networkiAnd NjThere is no correlation between them and the correlation between them,
Figure BDA00035787284400000310
a value of 1 represents a road segment node N in the converged dynamic relationship graph networkiAnd NjThere is a correlation between them; thereby obtaining the converged dynamic relationship graph network GfExpressed as follows:
Figure BDA00035787284400000311
where n represents the number of nodes, i.e., the number of links.
Further, step S4 includes: the graph attention network is calculated as follows:
Figure BDA00035787284400000312
Figure BDA0003578728440000041
wherein alpha isijDenotes the attention coefficient, exp denotes the exponential function, LeakyRelu denotes the nonlinear activation function, σ denotes the sigmoid activation function, W and aTAll represent a weight matrix of the linear transformation, hi,hjRespectively representing road section nodes N in the graph attention networkiAnd NjIs input by the user, | | represents the splicing operation on the vector, HiRepresenting a link node NiNeighbor nodes in the graph attention network, hi' Note the graph attention network vs. road segment node NiThe neighbor node information is aggregated and then output space dependence characteristics are obtained; and inputting the spatial dependence features extracted by the graph attention network into a 1 x 1 convolutional layer, outputting the convolutional layer as a multi-element traffic flow prediction result, and constructing a multi-element traffic flow prediction model based on the graph network.
Further, step S5 includes: training a multi-element traffic flow prediction model based on a graph network by using multi-element traffic flow historical data, wherein one part of data is used as a training set, and the other part of data is used as a verification set; inputting the training set into a model, training the model according to a loss function:
Figure BDA0003578728440000042
wherein n is the number of road segments in the road network, and T is the number of time intervals included in the training set;
Figure BDA0003578728440000043
representing a link node NiFlow, average speed and road segment occupancy at time t;
Figure BDA0003578728440000044
representation model to road segment node NiPredicted values of flow, average speed and road section occupancy at time t; selecting an optimal prediction model using the validation set a predetermined number of times per iteration; and stopping training after the iteration times reach 100 times to obtain the optimal multi-element traffic flow prediction model based on the graph network.
Further, still include: and evaluating the prediction effect by using the root mean square error on the traffic flow prediction result.
The present invention further provides a computer readable storage medium, on which a computer program is stored, wherein the computer program is executed by a processor to implement the aforementioned method for predicting multivariate traffic flow.
The technical scheme of the invention has the beneficial effects that: forecasting is carried out by jointly considering the dependency relationship of various attributes such as the flow, the average speed, the road section occupancy rate and the like of the traffic flow, so that more space dependency information is considered by the model, and the traffic flow state is more comprehensively depicted by the forecasting result; constructing a dynamic space-time relation graph under the condition of no prior knowledge of a road network spatial structure, and autonomously learning spatial dependence relations among traffic flow attributes and road sections; the degree of dependence between traffic flow segments can be quantified by the weight of the node edges, making the network under study explanatory.
Drawings
FIG. 1 is a flow chart of a multivariate traffic flow prediction method based on graph network according to an embodiment of the invention;
FIG. 2 is a diagram of a multivariate traffic flow prediction model network architecture based on a graph network according to an embodiment of the present invention.
Detailed Description
In order to make the implementation objects, technical solutions and advantages of the present invention more clear, the present invention is further described below with reference to the accompanying drawings and specific embodiments.
The embodiment of the invention provides a multivariate traffic flow prediction method based on a graph network, and with reference to a graph 1, the method comprises the following steps of S1-S6:
and step S1, acquiring multivariate traffic flow historical data for traffic flow prediction.
Acquiring multivariate traffic flow historical data for traffic flow prediction, wherein the multivariate traffic flow historical data is actually a time sequence and is marked as X ═ X1,X2,…,XT]The time sequence represents all road sections in the road network in the time interval [1, T ]]The traffic flow data of (2) includes data of three traffic flow attributes of a flow rate, an average speed, and a link occupancy. Wherein the content of the first and second substances,
Figure BDA0003578728440000051
representing the flow, average speed and road section occupancy of all road sections in the road network at the moment T, and T belongs to [1, T ∈]And n represents the number of links.
Then, the acquired historical data is preprocessed: extracting multivariate traffic flow historical data at t moment
Figure BDA0003578728440000052
As input, the actual value of the traffic flow characteristic of each road segment at the time t +1
Figure BDA0003578728440000053
As a label.
And step S2, constructing an independent dynamic relation graph for each traffic flow attribute based on the multivariate traffic flow historical data.
The spatial relationship of the link variables is described using a graph G ═ (N, E), where node N represents the variable for a link in the traffic flow and E ═ E<Ni,Nj>Representing mutual adjacency between segments, characterised by nodes
Figure BDA0003578728440000054
Indicating the link N at time tiFlow, average speed, and road segment occupancy.
Due to time t +1, link NiIs not only related to the traffic flow attribute of itself at the time t, but also influenced by the traffic flow attributes of other links at the time t, based on which the link NiThe traffic flow state at time t +1 may be expressed as:
Figure BDA0003578728440000061
wherein the content of the first and second substances,
Figure BDA0003578728440000062
respectively represent the removed road sections NiThe flow, the average speed and the road section occupancy of the other road sections at the time t; f (-) represents the information in what way the nodes aggregate the surrounding nodes, i.e. how to consider the influence of traffic flow of other road segments for a particular road segment.
Regarding the road sections as nodes of the dynamic relationship graph, calculating the attention scores of any two road section nodes by the following method:
Figure BDA0003578728440000063
Figure BDA0003578728440000064
Figure BDA0003578728440000065
wherein the content of the first and second substances,
Figure BDA0003578728440000066
showing roadSegment node NiAnd NjAttention points on flow, average speed and road occupancy, i, j being 1,2, …, n, respectively;
Figure BDA0003578728440000067
and
Figure BDA0003578728440000068
all are matrixes used for linear transformation and also learnable parameters of the model, and the dimensionalities of the matrixes are dkT represents a matrix transpose;
Figure BDA0003578728440000069
respectively representing road section nodes NjThe flow rate at time t, the average speed and the link occupancy,
Figure BDA00035787284400000610
respectively representing road section nodes NiFlow at time t, average speed, and link occupancy.
And taking the attention score between two road section nodes as the weight of a connecting edge between the nodes in the dynamic relation graph to obtain the dynamic relation graph of the flow, the average speed and the road section occupancy, and respectively recording as: gq,Gk,Gv；
Figure BDA00035787284400000611
I.e. representing link nodes NiAnd NjIn-flow dynamic relationship graph GqThe weight of the middle-connected edge is,
Figure BDA00035787284400000612
i.e. representing link nodes NiAnd NjIn average velocity dynamics graph GkThe weight of the middle-connected edge is,
Figure BDA00035787284400000613
i.e. representing link nodes NiAnd NjDynamic relation graph G of road section occupancyvThe weight of the middle connecting edge.
And S3, fusing the dynamic relationship graphs of the traffic flow attributes and converting the fused dynamic relationship graphs into a fused dynamic relationship graph network with the attributes mutually linked.
And taking the mutual influence among the flow, the average speed and the road occupancy into consideration. I.e. road section NiThe flow at the time t +1 is not only related to the flow attributes of the self and other road sections at the time t, but also is subjected to the road section N at the time t in the speed relation graph and the road section occupancy rate relation graphiThe value of (c) is affected, and the effect is realized through fusion among various attribute relation graphs.
For any two road segment nodes NiAnd NjAccording to
Figure BDA0003578728440000071
Calculating two road section nodes NiAnd NjAnd (3) the adjacent relations in the converged dynamic relational graph network so as to obtain the converged dynamic relational graph network.
The calculation formula of the adjacency relation is as follows:
Figure BDA0003578728440000072
wherein the content of the first and second substances,
Figure BDA0003578728440000073
representing road segment nodes N in a converged dynamic relationship graph networkiAnd NjThe adjacent relationship of (a) to (b),
Figure BDA0003578728440000074
a value of 0 represents a link node N in the converged dynamic relationship graph networkiAnd NjThere is no correlation between the two or more groups,
Figure BDA0003578728440000075
a value of 1 represents a road segment node N in the converged dynamic relationship graph networkiAnd NjThere is a correlation between them.
Thereby, the converged dynamic relationship graph network G is obtainedfExpressed as follows:
Figure BDA0003578728440000076
where n represents the number of nodes, i.e., the number of links.
And S4, based on the fusion dynamic relational graph network, learning potential space dependency relationships between traffic flow attributes and between road sections by using a graph attention network, and extracting space dependency characteristics to construct a multi-element traffic flow prediction model based on the graph network.
The graph attention network is calculated as follows:
Figure BDA0003578728440000077
Figure BDA0003578728440000078
wherein alpha isijDenotes the attention coefficient, exp denotes the exponential function, LeakyRelu denotes the nonlinear activation function, σ denotes the sigmoid activation function, W and aTAll represent a weight matrix of a linear transformation, hi,hjRespectively representing road section nodes N in the graph attention networkiAnd NjThe input feature of (1), represents the splicing operation on the vector, HiRepresenting a link node NiNeighbor nodes in a graph attention network, hi' Note the graph attention network vs. road segment node NiAnd (4) aggregating the neighbor node information and outputting the aggregated neighbor node information.
The spatial dependence features extracted by the graph attention network are input into a 1 x 1 convolutional layer, and the output of the convolutional layer is a multivariate traffic flow prediction result. And completing the construction of the multivariate traffic flow prediction model based on the graph network.
And S5, training the multi-element traffic flow prediction model based on the graph network.
The multivariate traffic flow historical data is used for training a multivariate traffic flow prediction model based on a graph network, wherein one part of data (such as 80%) serves as a training set, and the other part of data (such as 20%) serves as a verification set. Inputting the training set into a model, training the model according to a loss function:
Figure BDA0003578728440000081
wherein the content of the first and second substances,
Figure BDA0003578728440000082
representing a link node NiFlow, average speed and road segment occupancy at time t;
Figure BDA0003578728440000083
representation model to road segment node NiAnd (4) predicted values of the flow, the average speed and the road section occupancy at the time t.
Selecting an optimal prediction model by using a verification set every iteration for a certain number of times; and stopping training after the iteration times reach 100 times to obtain the optimal multi-element traffic flow prediction model based on the graph network. It should be understood that the number of iterations is not limited to 100, but may be other values around 100, for example only.
And S6, carrying out short-time traffic flow prediction according to the trained multi-element traffic flow prediction model.
Referring to fig. 2, in prediction, multivariate traffic flow historical data are input into a constructed multivariate traffic flow prediction model, a flow dynamic relation graph, an average speed dynamic relation graph and a link occupancy dynamic relation graph are obtained by using a self-attention mechanism according to step S2, then the relation graphs are fused according to step S3, then a graph attention network GAT is used for extracting a spatial dependency feature according to step S4, 1 × 1 convolution is input to convert the dimension of the spatial dependency feature into the dimension of desired output, and a prediction result of the traffic flow is output and expressed as the prediction result of the traffic flow
Figure BDA0003578728440000084
Wherein
Figure BDA0003578728440000085
Road segment N representing model predictionsiFlow, average speed, and link occupancy at time t + 1. Finally, the algorithm effect can be evaluated based on the root mean square error.
The embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method for predicting multivariate traffic flow based on graph network provided in the foregoing embodiment can be implemented.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several equivalent substitutions or obvious modifications can be made without departing from the spirit of the invention, and all the properties or uses are considered to be within the scope of the invention.

Claims (10)

1. A multivariate traffic flow prediction method based on a graph network is characterized by comprising the following steps:
s1, acquiring multivariate traffic flow historical data for traffic flow prediction;
s2, constructing an independent dynamic relation graph for each traffic flow attribute based on the multivariate traffic flow historical data;
s3, fusing the dynamic relationship graphs of the traffic flow attributes and converting the fused dynamic relationship graphs into a fused dynamic relationship graph network with the attributes mutually linked;
s4, based on the fusion dynamic relational graph network, learning potential space dependency relationships between traffic flow attributes and between road sections by using a graph attention network, extracting space dependency characteristics, and constructing a multi-element traffic flow prediction model based on the graph network;
s5, training the multi-element traffic flow prediction model based on the graph network;
and S6, carrying out short-time traffic flow prediction according to the trained multi-element traffic flow prediction model.
2. The multivariate traffic flow prediction method based on graph network as claimed in claim 1, wherein in step S1, the obtaining of the multivariate traffic flow history data comprises:
obtaining time interval [1, T ] of all road sections in road network]The multivariate traffic flow historical data in the time sequence X ═ X1,X2,…,XT]Represents; wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003578728430000011
showing the flow, average speed and road section occupancy of all road sections in the road network at the time T, wherein T belongs to [1, T ]]And n represents the number of links.
3. The multi-element traffic flow prediction method based on graph network as claimed in claim 2, wherein step S1 further comprises the following preprocessing of the multi-element traffic flow history data:
extracting multivariate traffic flow historical data at t moment
Figure FDA0003578728430000012
As input, the actual value of the traffic flow characteristic of each road section at the moment t +1
Figure FDA0003578728430000013
As a label.
4. The map network-based multivariate traffic flow prediction method according to claim 1, wherein the step S2 comprises:
aiming at three traffic flow attributes of flow, average speed and road section occupancy rate, a self-attention mechanism is respectively used for learning the mutual adjacency relation among road section nodes, the attention score between two road section nodes is used as the weight of a connecting edge between the nodes in a dynamic relation graph, and the independent dynamic relation graph of each traffic flow attribute is obtained.
5. The multivariate traffic flow prediction method based on graph network as claimed in claim 4, wherein the step S2 comprises:
taking the road sections as nodes of the dynamic relationship graph, and calculating the attention scores of any two road section nodes by the following method:
Figure FDA0003578728430000021
Figure FDA0003578728430000022
Figure FDA0003578728430000023
wherein the content of the first and second substances,
Figure FDA0003578728430000024
representing a link node NiAnd NjAttention points on traffic, average speed and road occupancy, i, j ═ 1,2, …, n;
Figure FDA0003578728430000025
and Ws q,Ws k,Ws vAll are matrixes used for linear transformation and also are learnable parameters of the model, and the dimensionalities of the matrixes are all dk，·TRepresenting a matrix transposition;
Figure FDA0003578728430000026
Figure FDA0003578728430000027
respectively representing road section nodes NjThe flow rate at time t, the average speed and the road segment occupancy,
Figure FDA0003578728430000028
Figure FDA0003578728430000029
respectively indicate the waySegment node NiThe flow, the average speed and the road section occupancy at the moment t;
and taking the attention score between two road section nodes as the weight of a connecting edge between the nodes in the dynamic relation graph to obtain the dynamic relation graph of the flow, the average speed and the road section occupancy, and respectively recording as: gq,Gk,Gv；
Figure FDA00035787284300000210
I.e. representing link nodes NiAnd NjIn-flow dynamic relationship graph GqThe weight of the middle-connected edge is,
Figure FDA00035787284300000211
i.e. representing link nodes NiAnd NjAt average speed dynamics graph GkThe weight of the middle-connected edge is,
Figure FDA00035787284300000212
i.e. representing link nodes NiAnd NjDynamic relation graph G of road section occupancyvThe weight of the middle connecting edge.
6. The multivariate traffic flow prediction method based on graph network as claimed in claim 5, wherein the step S3 comprises:
for any two road segment nodes NiAnd NjAccording to
Figure FDA00035787284300000213
Calculating two road section nodes NiAnd NjThe adjacent relation in the network of the converged dynamic relational graph is obtained;
the calculation formula of the adjacency relation is as follows:
Figure FDA00035787284300000214
wherein the content of the first and second substances,
Figure FDA00035787284300000215
representing road segment nodes N in a converged dynamic relationship graph networkiAnd NjThe adjacent relationship of (a) to (b),
Figure FDA00035787284300000216
is 0 represents a road segment node N in the converged dynamic relationship graph networkiAnd NjThere is no correlation between them and the correlation between them,
Figure FDA0003578728430000031
a value of 1 represents a road segment node N in the converged dynamic relationship graph networkiAnd NjThere is a correlation between them;
thereby obtaining the converged dynamic relationship graph network GfExpressed as follows:
Figure FDA0003578728430000032
where n represents the number of nodes, i.e., the number of links.
7. The map network-based multivariate traffic flow prediction method according to claim 1, wherein the step S4 comprises:
the graph attention network is calculated as follows:
Figure FDA0003578728430000033
Figure FDA0003578728430000034
wherein alpha isijDenotes the attention coefficient, exp denotes the exponential function, LeakyRelu denotes the nonlinear activation function, σ denotes the sigmoid activation function, W and aTAll represent a weight matrix of the linear transformation, hi,hjRespectively representing road section nodes N in the graph attention networkiAnd NjThe input feature of (1), represents the splicing operation on the vector, HiRepresenting a link node NiNeighbor nodes in a graph attention network, hi' Note the graph attention network vs. road segment node NiThe neighbor node information is aggregated and then output space dependence characteristics are obtained;
and inputting the spatial dependence characteristics of all road section nodes extracted by the graph attention network into a 1 x 1 convolution layer, outputting the convolution layer as a multivariate traffic flow prediction result, and constructing a multivariate traffic flow prediction model based on the graph network.
8. The multivariate traffic flow prediction method based on graph network as claimed in claim 1, wherein the step S5 comprises:
training a multi-element traffic flow prediction model based on a graph network by using multi-element traffic flow historical data, wherein one part of data is used as a training set, and the other part of data is used as a verification set;
inputting the training set into a model, training the model according to a loss function:
Figure FDA0003578728430000041
wherein n is the number of road segments in the road network, and T is the number of time intervals included in the training set;
Figure FDA0003578728430000042
representing a link node NiFlow, average speed and road segment occupancy at time t;
Figure FDA0003578728430000043
representation model to road segment node NiPredicted values of flow, average speed and road section occupancy at the moment t;
selecting an optimal prediction model using the validation set a predetermined number of times per iteration; and stopping training after the iteration times reach 100 times to obtain the optimal multi-element traffic flow prediction model based on the graph network.
9. The map network-based multivariate traffic flow prediction method of claim 1, further comprising: and evaluating the prediction effect by using the root mean square error on the traffic flow prediction result.
10. A computer-readable storage medium having stored thereon a computer program, characterized in that: the computer program when executed by a processor implements the multivariate traffic flow prediction method of any of claims 1-9.
CN202210349264.2A 2022-04-01 2022-04-01 Multi-element traffic flow prediction method based on graph network Active CN114743374B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210349264.2A CN114743374B (en) 2022-04-01 2022-04-01 Multi-element traffic flow prediction method based on graph network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210349264.2A CN114743374B (en) 2022-04-01 2022-04-01 Multi-element traffic flow prediction method based on graph network

Publications (2)

Publication Number Publication Date
CN114743374A true CN114743374A (en) 2022-07-12
CN114743374B CN114743374B (en) 2023-04-07

Family

ID=82279251

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210349264.2A Active CN114743374B (en) 2022-04-01 2022-04-01 Multi-element traffic flow prediction method based on graph network

Country Status (1)

Country Link
CN (1) CN114743374B (en)

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102018201787A1 (en) * 2018-02-06 2019-08-08 Siemens Aktiengesellschaft Method and system for optimizing and predicting a traffic situation
CN110969854A (en) * 2019-12-13 2020-04-07 深圳先进技术研究院 Traffic flow prediction method, system and terminal equipment
CN112241814A (en) * 2020-10-20 2021-01-19 河南大学 Traffic prediction method based on reinforced space-time diagram neural network
CN113053115A (en) * 2021-03-17 2021-06-29 北京工商大学 Traffic prediction method based on multi-scale graph convolution network model
CN113313947A (en) * 2021-05-31 2021-08-27 湖南大学 Road condition evaluation method of short-term traffic prediction graph convolution network
CN113505536A (en) * 2021-07-09 2021-10-15 兰州理工大学 Optimized traffic flow prediction model based on space-time diagram convolution network
CN113673769A (en) * 2021-08-24 2021-11-19 北京航空航天大学 Graph neural network traffic flow prediction method based on multivariate time sequence interpolation
CN113705880A (en) * 2021-08-25 2021-11-26 杭州远眺科技有限公司 Traffic speed prediction method and device based on space-time attention diagram convolutional network
CN114038200A (en) * 2021-11-29 2022-02-11 东北大学 Attention mechanism-based time-space synchronization map convolutional network traffic flow prediction method
CN114120652A (en) * 2021-12-21 2022-03-01 重庆邮电大学 Traffic flow prediction method based on dynamic graph neural network

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102018201787A1 (en) * 2018-02-06 2019-08-08 Siemens Aktiengesellschaft Method and system for optimizing and predicting a traffic situation
CN110969854A (en) * 2019-12-13 2020-04-07 深圳先进技术研究院 Traffic flow prediction method, system and terminal equipment
CN112241814A (en) * 2020-10-20 2021-01-19 河南大学 Traffic prediction method based on reinforced space-time diagram neural network
CN113053115A (en) * 2021-03-17 2021-06-29 北京工商大学 Traffic prediction method based on multi-scale graph convolution network model
CN113313947A (en) * 2021-05-31 2021-08-27 湖南大学 Road condition evaluation method of short-term traffic prediction graph convolution network
CN113505536A (en) * 2021-07-09 2021-10-15 兰州理工大学 Optimized traffic flow prediction model based on space-time diagram convolution network
CN113673769A (en) * 2021-08-24 2021-11-19 北京航空航天大学 Graph neural network traffic flow prediction method based on multivariate time sequence interpolation
CN113705880A (en) * 2021-08-25 2021-11-26 杭州远眺科技有限公司 Traffic speed prediction method and device based on space-time attention diagram convolutional network
CN114038200A (en) * 2021-11-29 2022-02-11 东北大学 Attention mechanism-based time-space synchronization map convolutional network traffic flow prediction method
CN114120652A (en) * 2021-12-21 2022-03-01 重庆邮电大学 Traffic flow prediction method based on dynamic graph neural network

Also Published As

Publication number Publication date
CN114743374B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
US11461654B2 (en) Multi-agent cooperation decision-making and training method
CN112382082B (en) Method and system for predicting traffic running state in congested area
Wang et al. Research and application of a combined model based on multi-objective optimization for multi-step ahead wind speed forecasting
Meng et al. Wind speed forecasting based on wavelet packet decomposition and artificial neural networks trained by crisscross optimization algorithm
CN109241313B (en) Image retrieval method based on high-order deep hash learning
CN111104522A (en) Regional industry association effect trend prediction method based on knowledge graph
Sawale et al. Use of artificial neural network in data mining for weather forecasting
CN106529818B (en) Water quality assessment Forecasting Methodology based on Fuzzy Wavelet Network
CN111967972B (en) Financial product recommendation method and device
CN114299723B (en) Traffic flow prediction method
CN103324954A (en) Image classification method based on tree structure and system using same
CN113762595B (en) Traffic time prediction model training method, traffic time prediction method and equipment
CN114925836B (en) Urban traffic flow reasoning method based on dynamic multi-view graph neural network
CN115545350B (en) Vehicle path problem solving method integrating deep neural network and reinforcement learning
CN112396187A (en) Multi-agent reinforcement learning method based on dynamic collaborative map
CN113190688A (en) Complex network link prediction method and system based on logical reasoning and graph convolution
CN113240086B (en) Complex network link prediction method and system
CN113537580A (en) Public transport passenger flow prediction method and system based on adaptive graph learning
CN114780739A (en) Time sequence knowledge graph completion method and system based on time graph convolution network
CN115376317A (en) Traffic flow prediction method based on dynamic graph convolution and time sequence convolution network
CN113360664B (en) Knowledge graph complementing method
CN113326884B (en) Efficient learning method and device for large-scale heterograph node representation
CN112989539B (en) Traffic prediction method based on intersection transfer calculation
CN113627676A (en) Traffic prediction method and system based on multi-attention causal relationship
CN114743374B (en) Multi-element traffic flow prediction method based on graph 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