CN116363874A - Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation - Google Patents

Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation Download PDF

Info

Publication number
CN116363874A
CN116363874A CN202310268121.3A CN202310268121A CN116363874A CN 116363874 A CN116363874 A CN 116363874A CN 202310268121 A CN202310268121 A CN 202310268121A CN 116363874 A CN116363874 A CN 116363874A
Authority
CN
China
Prior art keywords
hypergraph
traffic
matrix
semantic
traffic state
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
CN202310268121.3A
Other languages
Chinese (zh)
Other versions
CN116363874B (en
Inventor
唐坤
徐添
郭唐仪
何流
徐永能
刘英舜
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 of Science and Technology
Original Assignee
Nanjing University of Science and Technology
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 of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202310268121.3A priority Critical patent/CN116363874B/en
Publication of CN116363874A publication Critical patent/CN116363874A/en
Application granted granted Critical
Publication of CN116363874B publication Critical patent/CN116363874B/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
    • G08G1/0129Traffic data processing for creating historical data or processing based on historical data
    • 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
    • G06N3/042Knowledge-based neural networks; Logical representations of neural 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
    • G06N3/0464Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/0985Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • 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
    • 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 discloses an urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation, which comprises the following steps: extracting static space topology data and dynamic traffic state data among all traffic areas, and constructing a traffic area adjacency matrix, a recent traffic state feature matrix and a historical traffic state feature matrix; establishing a space topology hypergraph, a recent semantic hypergraph and a historical semantic hypergraph according to the extracted data; splicing the three hypergraphs to obtain a new fusion hypergraph, constructing a depth hypergraph convolution network on the basis, and performing hypergraph convolution layer by layer; and training the model by using the training set to obtain optimal model parameters, and predicting the traffic condition at the next moment by using the optimal model. According to the invention, the hypergraph data structure is introduced into the complex correlation modeling of the traffic condition, so that the second-order spatial adjacency of the traffic area is considered, the multi-mode high-order semantic correlation of the traffic state is considered, and the accuracy of road network traffic state prediction is improved.

Description

Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation
Technical Field
The invention belongs to urban traffic prediction technology, and particularly relates to an urban traffic hypergraph convolution prediction method integrating multi-mode high-order semantic correlation.
Background
Providing active traffic condition information in urban networks is significant and challenging in traffic management and operation. Due to the nature and intuitiveness of traffic networks, the potential of graph roll-up network (GCN) based models in traffic prediction has been at a brand-new angle, with increasing attention. Typical GCN models use a pair-wise connection to capture only the second-order correlation between vertices. However, the potential dependencies hidden in the actual application data may go beyond the range of pairs, and they appear more highly correlated and even more complex in the face of multi-source data.
Disclosure of Invention
In order to solve the technical defects in the prior art, the invention provides an urban traffic hypergraph convolution prediction method integrating multi-mode high-order semantic correlation.
The technical scheme for realizing the purpose of the invention is as follows: a city traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation comprises the following steps:
step 1: extracting static space topological structure data and dynamic traffic state data among all traffic areas in a target area, and constructing a traffic area adjacent matrix, a recent traffic state feature matrix taking a time period as an interval and a historical traffic state feature matrix taking a day as an interval;
step 2: establishing a topological hypergraph based on second-order spatial correlation, a semantic hypergraph based on continuous time period traffic state characteristics and a semantic hypergraph based on daily traffic state characteristics according to the extracted data;
step 3: the method comprises the steps of constructing a depth hypergraph convolution network by taking a time period traffic state feature matrix, a daily traffic state feature matrix and a new hypergraph incidence matrix as inputs, and performing hypergraph convolution layer by layer;
step 4: training the model by using the training set to obtain optimal model parameters, inputting traffic state characteristics at the next moment into the trained optimal model, and predicting the traffic state at the next moment.
Preferably, the specific steps of extracting static space topological structure data and dynamic traffic state data of each traffic area in the target area are as follows:
step 11: extracting the space topological structure among all traffic areas of the target area and marking the space topological structure as an adjacent matrix A g Adjacency matrix A g The data are expressed as:
Figure BDA0004133697230000021
wherein a is g (i, j) is matrix A g Elements of row i, column j, v i The node is the ith node in the graph, and corresponds to the ith traffic area;
step 12: constructing a recent traffic state feature matrix X taking time periods as intervals at the current moment t according to traffic state data of each traffic region in different time periods r Expressed as:
Figure BDA0004133697230000022
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure BDA0004133697230000023
the traffic conditions of N areas at the time t are represented, N represents the number of traffic areas, and p represents a time period; feature matrix X r Is +.>
Figure BDA0004133697230000024
i=1, 2, l, n, representing the recent traffic feature vector of the i-th region, column vector +.>
Figure BDA0004133697230000025
j=1, 2, l, p denote traffic state information of all areas at different time periods;
step 13: acquiring historical data of the period t+1 to be predicted in the previous q days to form a historical traffic state characteristic matrix X with the days as intervals d Expressed as:
Figure BDA0004133697230000026
wherein the method comprises the steps of
Figure BDA0004133697230000027
Representing the traffic condition of N areas in the next time period t+1, T d The number of time slots included per day is indicated, and q indicates the number of days considered.
Preferably, the specific methods for establishing the topological hypergraph based on the second-order spatial correlation, the recent semantic hypergraph based on the continuous time period traffic state characteristics and the historical semantic hypergraph based on the daily traffic state characteristics according to the extracted data are as follows:
step 21: according to the characteristic matrix X r The KNN nearest neighbor algorithm is used for constructing the superficiality, which is specifically as follows:
for each node v i I=1, 2, l, n, using KNN algorithm in matrix X r Calculating k-1 nearest neighbor nodes in the row vectors of the row number; node v i And k-1 nearest neighbor nodes together constitute superedge
Figure BDA0004133697230000028
Obtaining N superflimit;
splice all supersides into one
Figure BDA00041336972300000213
Is denoted as an association matrix H r Expressed as:
Figure BDA0004133697230000029
wherein h is r (i, j) is a matrix H r Elements of the ith row, jth column;
constructing semantic hypergraphs of recent traffic conditions according to hyperedges
Figure BDA00041336972300000210
Wherein->
Figure BDA00041336972300000211
Representing a set of traffic areas, +.>
Figure BDA00041336972300000212
Representing a hyperedge set in a recent semantic hypergraph;
step 22: according to the characteristic matrix X d Constructing supersides by a KNN nearest neighbor algorithm, and splicing all the supersides to form an association matrix H d Expressed as
Figure BDA0004133697230000031
Wherein the method comprises the steps of
Figure BDA0004133697230000032
Is a matrix H d The j th superside of (2);
semantic hypergraph for constructing historical traffic states according to hyperedges
Figure BDA0004133697230000033
Wherein->
Figure BDA0004133697230000034
Representing a hyperedge set in a history semantic hypergraph;
step 23: based on the extracted adjacency matrix A g Constructing topological hypergraph based on second-order spatial correlation
Figure BDA0004133697230000035
Second order spatial correlation for reflecting traffic conditions;
for adjacency matrix A g Is regarded as a superside comprising only two nodes, thereby constructing a space topology supergraph
Figure BDA0004133697230000036
The hypergraph is composed of a +.>
Figure BDA0004133697230000037
Is the correlation matrix H of (1) g Description of the drawings;
Figure BDA0004133697230000038
wherein the method comprises the steps of
Figure BDA0004133697230000039
Is H g The j th superside of (2) represents the second-order correlation, epsilon, between different traffic areas g Representing a set of hyperedges in a spatial topology hypergraph.
Preferably, the new hypergraph obtained after the three hypergraphs are spliced is used as the input of the depth network to carry out hypergraph convolution, and the specific method is as follows:
step 31: splicing the incidence matrixes based on the spatial topological hypergraph and corresponding to the recent traffic state semantic hypergraph and the historical traffic state semantic hypergraph obtained in the step 3 to obtain a comprehensive hypergraph integrating the multi-mode high-order semantic correlation
Figure BDA00041336972300000310
Comprehensive hypergraph->
Figure BDA00041336972300000311
From a dimension of->
Figure BDA00041336972300000312
Is described by an association matrix H; the recent traffic established in the step 1 is processedThe state characteristic matrix is spliced with the historical traffic state characteristic matrix to form a comprehensive traffic state characteristic matrix
Figure BDA00041336972300000313
Step 32: taking the incidence matrix H and the feature matrix X obtained in the step 31 as deep network input, and performing hypergraph convolution;
the depth network is formed by stacking a plurality of hypergraph convolution layers;
the calculation formula of each hypergraph convolution layer is as follows:
Figure BDA00041336972300000314
wherein the method comprises the steps of
Figure BDA00041336972300000315
Is the input signal of the first layer in the hypergraph, is->
Figure BDA00041336972300000316
Is a parameter matrix of the first layer, sigma is a nonlinear activation function, D v ,D e And W is a node degree matrix, a superside degree matrix and a superside weight matrix respectively.
Preferably, the training set is used for training the prediction model, the error between the predicted value and the true value is minimized, and the specific method for calculating the optimal parameters of the prediction model is as follows:
constructing a training set according to traffic state observed values of N traffic areas in T time periods
Figure BDA0004133697230000041
Wherein->
Figure BDA0004133697230000042
As a matrix of features,
Figure BDA0004133697230000043
for predictive labels, p is the number of features;
inputting the training set into a prediction model for training, wherein the model training objective function is as follows:
Figure BDA0004133697230000044
wherein the first term is a predicted mean square error empirical loss term, the second term is a model parameter regularization term,
Figure BDA0004133697230000045
representing hypergraph ++>
Figure BDA00041336972300000411
The upper parameter is the predictive model of Θ, +.>
Figure BDA0004133697230000046
As a loss function, m=t-p is the number of samples.
After model training, the feature matrix at the current moment is obtained
Figure BDA0004133697230000047
Inputting the value to the optimal model obtained after training, and obtaining the prediction of the traffic condition at the next moment as follows:
Figure BDA0004133697230000048
wherein the method comprises the steps of
Figure BDA0004133697230000049
For the feature matrix at the current moment, < > and->
Figure BDA00041336972300000410
And the parameters are the learned optimal model parameters.
Compared with the prior art, the invention has the remarkable advantages that:
(1) According to the invention, a new data structure, namely a hypergraph data structure, is introduced into complex correlation modeling of traffic conditions, so that high-order semantic correlation can be modeled, multi-mode correlation is incorporated, and the interpretability and the accuracy of a prediction model are improved by comprehensively considering related information;
(2) Based on the hypergraph learning theory, the invention provides a new convolution operator to learn the characteristic representation on the hypergraph, and the information transmission between the vertexes is more universal by fully utilizing the correlation between the high order and the multiple modes, and a basic component similar to the graph convolution is provided for the construction of the depth hypergraph network, so that the accuracy of the prediction result is improved;
(3) The traffic condition framework based on hypergraph is used for cooperatively predicting the traffic condition of the urban road network, and the geographic space association and the multimode high-order semantic association of the traffic condition are fused in a unified framework.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, like reference numerals being used to refer to like parts throughout the several views.
FIG. 1 is a flow chart of the urban traffic hypergraph convolution prediction method which is provided by the invention and is fused with the multimode high-order semantic correlation.
Fig. 2 is an input schematic of the present invention.
Fig. 3 is a hypergraph convolution schematic of the present invention.
Detailed Description
It is easy to understand that various embodiments of the present invention can be envisioned by those of ordinary skill in the art without altering the true spirit of the present invention in light of the present teachings. Accordingly, the following detailed description and drawings are merely illustrative of the invention and are not intended to be exhaustive or to limit or restrict the invention. Rather, these embodiments are provided so that this disclosure will be thorough and complete by those skilled in the art. Preferred embodiments of the present invention are described in detail below with reference to the attached drawing figures, which form a part of the present application and are used in conjunction with embodiments of the present invention to illustrate the innovative concepts of the present invention.
The invention relates to an end-to-end method, which consists of four main parts: data input, hypergraph construction, hypergraph convolution and traffic prediction. Based on the static topography data and the dynamic traffic data of the multimode, two semantic hypergraphs and one geospatial graph are constructed and integrated to describe the high-order semantic association and the second-order geospatial association of the fine granularity and the coarse granularity of the traffic condition. To learn the feature representation on hypergraphs, a new hypergraph convolution operator is derived from graph convolution and hypergraph learning theory. By using the proposed hypergraph convolution as a deep network of building blocks, a predicted advanced feature representation is learned and then traffic conditions are predicted. The invention captures the second-order geospatial correlation and the multi-mode high-order semantic correlation of traffic conditions together in a unified framework, and is quite attractive.
The invention will be further described with reference to the accompanying drawings and examples.
Referring to fig. 1, an urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation includes the following steps:
step 1: extracting static space topological structure data and dynamic traffic state data among all traffic areas in a target area, and constructing a traffic area adjacent matrix, a recent traffic state feature matrix taking a time period as an interval and a historical traffic state feature matrix taking a day as an interval;
step 2: establishing a topological hypergraph based on second-order spatial correlation, a semantic hypergraph based on continuous time period traffic state characteristics and a semantic hypergraph based on daily traffic state characteristics according to the extracted data;
step 3: the method comprises the steps of constructing a depth hypergraph convolution network by taking a time period traffic state feature matrix, a daily traffic state feature matrix and a new hypergraph incidence matrix as inputs, and performing hypergraph convolution layer by layer;
step 4: training the model by using the training set to obtain optimal model parameters, inputting traffic state characteristics at the next moment into the trained optimal model, and predicting the traffic state at the next moment.
In a further embodiment, as shown in the input part of fig. 1, step 1 extracts static space topology structure data and dynamic traffic state data between each traffic area in the target area, and the specific method for constructing the traffic area adjacency matrix, the recent traffic state feature matrix with time slots as intervals, and the historical traffic state feature matrix with days as intervals is as follows:
step 11: extracting the space topological structure among all traffic areas of the target area and marking the space topological structure as an adjacent matrix A g Adjacency matrix A g The data are expressed as:
Figure BDA0004133697230000061
wherein a is g (i, j) is matrix A g Elements of row i, column j, v i The node is the ith node in the graph, and corresponds to the ith traffic area;
step 12: as shown in fig. 2, a recent traffic state characteristic matrix X with time slots as intervals at the current time t is constructed according to traffic state data of each traffic region in different time slots r Expressed as:
Figure BDA0004133697230000062
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure BDA0004133697230000063
the traffic conditions of N areas at the time t are represented, N represents the number of traffic areas, and p represents a time period; feature matrix X r Is +.>
Figure BDA0004133697230000064
i=1, 2, l, n, representing the recent traffic feature vector of the i-th region, column vector +.>
Figure BDA0004133697230000065
j=1, 2, l, p denote traffic state information of all areas at different time periods;
step 13: as shown in fig. 2, historical data of the previous q days of the period t+1 to be predicted is acquired to form a historical traffic state characteristic matrix X with the days as intervals d Expressed as:
Figure BDA0004133697230000066
wherein the method comprises the steps of
Figure BDA0004133697230000067
Representing the traffic condition of N areas in the next time period t+1, T d The number of time slots included per day is indicated, and q indicates the number of days considered.
In a further embodiment, as shown in the hypergraph building block of FIG. 1, step 2 is specifically
Step 21: according to the characteristic matrix X r The KNN nearest neighbor algorithm is used for constructing the superficiality, which is specifically as follows:
for each node v i I=1, 2, l, n, using KNN algorithm in matrix X r Calculating k-1 nearest neighbor nodes in the row vectors of the row number; node v i And k-1 nearest neighbor nodes together constitute superedge
Figure BDA0004133697230000071
Obtaining N superflimit;
splice all supersides into one
Figure BDA00041336972300000715
Is denoted as an association matrix H r Expressed as:
Figure BDA0004133697230000072
wherein h is r (i, j) is a matrix H r Elements of the ith row, jth column;
constructing semantic hypergraphs of recent traffic conditions according to hyperedges
Figure BDA0004133697230000073
Wherein->
Figure BDA0004133697230000074
Representing a set of traffic areas, +.>
Figure BDA0004133697230000075
Representing a hyperedge set in a recent semantic hypergraph;
step 22: according to the characteristic matrix X d Constructing supersides by a KNN nearest neighbor algorithm, and splicing all the supersides to form an association matrix H d Expressed as
Figure BDA0004133697230000076
Wherein the method comprises the steps of
Figure BDA00041336972300000713
Is a matrix H d The j th superside of (2);
semantic hypergraph for constructing historical traffic states according to hyperedges
Figure BDA0004133697230000077
Wherein->
Figure BDA0004133697230000078
Representing a hyperedge set in a history semantic hypergraph;
step 23: based on the extracted adjacency matrix A g Constructing topological hypergraph based on second-order spatial correlation
Figure BDA0004133697230000079
Second order spatial correlation for reflecting traffic conditions;
for adjacency matrix A g Is regarded as a superside comprising only two nodes, thereby constructing a space topology supergraph
Figure BDA00041336972300000710
The hypergraph is composed of a +.>
Figure BDA00041336972300000714
Is the correlation matrix H of (1) g Description of the drawings;
Figure BDA00041336972300000711
wherein the method comprises the steps of
Figure BDA00041336972300000712
Is H g The j th superside of (2) represents the second-order correlation, epsilon, between different traffic areas g Representing a set of hyperedges in a spatial topology hypergraph.
In a further embodiment, as shown in the hypergraph convolution part in fig. 1, the specific method for performing hypergraph convolution by using the new hypergraph obtained by splicing three hypergraphs as the input of the depth network is as follows:
step 31: splicing the incidence matrixes based on the spatial topological hypergraph and corresponding to the recent traffic state semantic hypergraph and the historical traffic state semantic hypergraph obtained in the step 3 to obtain a comprehensive hypergraph integrating the multi-mode high-order semantic correlation
Figure BDA0004133697230000081
Comprehensive hypergraph->
Figure BDA0004133697230000082
From a dimension of->
Figure BDA0004133697230000083
Is described by an association matrix H; the recent traffic state established in the step 1 is processedThe characteristic matrix is spliced with the characteristic matrix of the historical traffic state to form the comprehensive traffic state characteristic matrix
Figure BDA0004133697230000084
Step 32: taking the incidence matrix H and the feature matrix X obtained in the step 31 as deep network input, and performing hypergraph convolution;
hypergraph convolution consists of two parts, hypergraph fusion and hypergraph convolution. Considering the relevance of multiple modes, the hypergraph is synthesized
Figure BDA0004133697230000085
By->
Figure BDA0004133697230000086
Three sub hypergraphs. Hypergraph->
Figure BDA00041336972300000816
Is epsilon g ,ε r ,ε d Is of dimension +.>
Figure BDA0004133697230000087
To learn the feature representation of the comprehensive hypergraph, the correlation matrix H, X r And X d Takes the feature matrix of the figure as input to carry out hypergraph convolution. Hypergraph convolution is a novel feature learning method specially designed for hypergraphs. Inspired by classical graph convolution and derived from hypergraph learning theory.
Comprehensive hypergraph
Figure BDA00041336972300000817
Laplacian delta expression of (A) is as follows
Figure BDA0004133697230000088
Wherein D is v ,D e And W is the diagonal matrix of vertex power, edge power and edge weight, respectivelyH is hypergraph
Figure BDA0004133697230000089
Is used for the correlation matrix of the (a). Which is an N x N semi-positive definite matrix.
Through characteristic decomposition, obtain
Figure BDA00041336972300000818
Wherein Φ= (Φ) 12 ,L,φ N ) Is a matrix composed of normal eigenvectors, Λ=diag (λ 12 ,L,λ N ) Is a diagonal matrix of eigenvalues.
Since the eigenvectors in Φ form the orthogonal basis of space, the eigenvectors alone in space
Figure BDA00041336972300000810
Fourier transform of (a) into
Figure BDA00041336972300000811
Then with convolution kernel
Figure BDA00041336972300000812
Frequency domain convolution of x of (2) is
Figure BDA00041336972300000813
Wherein the method comprises the steps of
Figure BDA00041336972300000814
Is the inverse fourier transform, e is the hadamard product, where the eigenvalue function g θ (Λ) is defined as
Figure BDA00041336972300000819
Fourier transformAnd inverse fourier transform are of time complexity
Figure BDA00041336972300000815
g θ (Λ) can be approximated using chebyshev's K-order expansion, and the frequency domain convolution in step 35 can be reduced to
Figure BDA0004133697230000091
Wherein θ is k Is the chebyshev coefficient and,
Figure BDA0004133697230000092
is a Laplace scaling operator, T k (x) Chebyshev polynomials. T (T) k (x) Can be made of T k (x)=2xT k-1 (x)-T k-2 (x),T 0 (x)=1,T 1 (x) Calculated =x.
Since the higher order associations between vertices can be well represented by the laplacian in hypergraph, the first order ChebNet is employed to further increase computational efficiency, i.e., k=1, λ max =2. The convolution process may be simplified as follows:
Figure BDA0004133697230000093
wherein θ is 0 And theta 1 The two coefficients can be reduced to using one coefficient
Figure BDA0004133697230000094
So the hypergraph convolution is finally
Figure BDA0004133697230000095
Based on the hypergraph convolution operator, a hypergraph convolution neural network can be constructed, and the characteristic expression of the hypergraph can be learned layer by layer. By stacking multiple layers of hypergraph convolutions to build a deep network, the convolution layers on the network can be calculated as follows:
Figure BDA0004133697230000096
the exploded view of each part of the formula is shown in FIG. 3
Wherein the method comprises the steps of
Figure BDA0004133697230000097
Is the input signal of the first layer in the hypergraph, is->
Figure BDA0004133697230000098
Is the parameter matrix of the first layer, and is a nonlinear activation function. The core is theta (l) Is the next layer +.>
Figure BDA0004133697230000099
Mid-vertex feature representations are also available.
In a further embodiment, as shown in the prediction part in fig. 1, the training set is used to train the prediction model, the error between the predicted value and the true value is minimized, the optimal parameter of the prediction model is calculated, the feature representation learned by the last layer of the depth network is brought into the optimal parameter prediction model, and the specific method for predicting the traffic condition at the next moment is as follows:
constructing a training set according to traffic state observed values of N traffic areas in T time periods
Figure BDA00041336972300000910
Wherein->
Figure BDA00041336972300000911
As a matrix of features,
Figure BDA0004133697230000101
for predictive labels, p is the number of features;
the learning of the whole prediction model is to minimize the error between the predicted value and the true value on the training set, the training set is input into the prediction model for training, and the objective function of model training is as follows:
Figure BDA0004133697230000102
wherein the first term is a predicted mean square error empirical loss term, the second term is a model parameter regularization term,
Figure BDA0004133697230000103
representing hypergraph ++>
Figure BDA0004133697230000109
The upper parameter is the predictive model of Θ, +.>
Figure BDA0004133697230000104
As a loss function, m=t-p is the number of samples.
Based on the learning model f, the feature matrix of the current moment is obtained
Figure BDA0004133697230000105
Inputting the value to the optimal model obtained after training, and obtaining the prediction of the traffic condition at the next moment as follows:
Figure BDA0004133697230000106
wherein the method comprises the steps of
Figure BDA0004133697230000107
For the feature matrix at the current moment, < > and->
Figure BDA0004133697230000108
And the parameters are the learned optimal model parameters.
The invention introduces a new data representation method, namely hypergraph, to simulate complex relations in traffic data.
The above description is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto,
any changes or substitutions that would be easily recognized by those skilled in the art within the technical scope of the present disclosure are intended to be covered by the present invention.
It should be appreciated that in the above description of exemplary embodiments of the invention, various features of the invention are sometimes described in the context of a single embodiment or with reference to a single figure in order to streamline the invention and aid those skilled in the art in understanding the various aspects of the invention. The present invention should not, however, be construed as including features that are essential to the patent claims in the exemplary embodiments.
It should be understood that modules, units, components, etc. included in the apparatus of one embodiment of the present invention may be adaptively changed to arrange them in an apparatus different from the embodiment. The different modules, units or components comprised by the apparatus of the embodiments may be combined into one module, unit or component or they may be divided into a plurality of sub-modules, sub-units or sub-components.

Claims (5)

1. The urban traffic hypergraph convolution prediction method integrating the multimode high-order semantic correlation is characterized by comprising the following steps of:
step 1: extracting static space topology data and dynamic traffic state data among all traffic areas in a target area, and constructing a traffic area adjacency matrix, a recent traffic state feature matrix taking a time period as an interval and a historical traffic state feature matrix taking a day as an interval;
step 2: establishing a space topology hypergraph based on second-order space adjacency, a recent semantic hypergraph based on continuous time period traffic state characteristics and a historical semantic hypergraph based on daily traffic state characteristics according to the extracted data;
step 3: the three hypergraphs are spliced to obtain a new fusion hypergraph, a depth hypergraph convolution network is built by taking a time period traffic state feature matrix, a daily traffic state feature matrix and an incidence matrix of the fusion hypergraph as inputs, and hypergraph convolution is carried out layer by layer;
step 4: training the model by using the training set to obtain optimal model parameters, inputting traffic state characteristics at the next moment into the trained optimal model, and predicting the traffic state at the next moment.
2. The urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation according to claim 1, wherein the specific steps of extracting static space topology data and dynamic traffic state data of each traffic area in a target area are as follows:
step 11: extracting the space topological structure among all traffic areas of the target area and marking the space topological structure as an adjacent matrix A g Adjacency matrix A g The data are expressed as:
Figure FDA0004133697210000011
wherein a is g (i, j) is matrix A g Elements of row i, column j, v i The node is the ith node in the graph, and corresponds to the ith traffic area;
step 12: constructing a recent traffic state feature matrix X taking time periods as intervals at the current moment t according to traffic state data of each traffic region in different time periods r Expressed as:
Figure FDA0004133697210000012
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure FDA0004133697210000013
the traffic conditions of N areas at the time t are represented, N represents the number of traffic areas, and p represents a time period; feature matrix X r Is +.>
Figure FDA0004133697210000014
Recent traffic feature vector representing the ith region, column vector +.>
Figure FDA0004133697210000015
Traffic state information representing all areas at different time periods;
step 13: acquiring historical data of the period t+1 to be predicted in the previous q days to form a historical traffic state characteristic matrix X with the days as intervals d Expressed as:
Figure FDA0004133697210000021
wherein the method comprises the steps of
Figure FDA0004133697210000022
Representing the traffic condition of N areas in the next time period t+1, T d The number of time slots included per day is indicated, and q indicates the number of days considered.
3. The urban traffic hypergraph convolution prediction method with fusion of multimode high-order semantic relevance according to claim 2, wherein the specific method for establishing the spatial topological hypergraph based on the second-order spatial adjacency, the recent semantic hypergraph based on the continuous time period traffic state characteristics and the historical semantic hypergraph based on the daily traffic state characteristics according to the extracted data is as follows:
step 21: according to the characteristic matrix X r The KNN nearest neighbor algorithm is utilized to construct the superside, which is specifically as follows:
for each node v i I=1, 2, l, n, using KNN algorithm in matrix X r Calculating k-1 nearest neighbor nodes in the row vectors of the row number; node v i And k-1 nearest neighbor nodes together constitute superedge
Figure FDA0004133697210000023
Obtaining N superflimit;
splice all supersides into one
Figure FDA0004133697210000024
Is denoted as an association matrix H r Expressed as:
Figure FDA0004133697210000025
wherein h is r (i, j) is a matrix H r Elements of the ith row, jth column;
constructing semantic hypergraphs of recent traffic conditions according to hyperedges
Figure FDA0004133697210000026
Wherein->
Figure FDA0004133697210000027
Representing a set of traffic areas, +.>
Figure FDA0004133697210000028
Representing a hyperedge set in a recent semantic hypergraph;
step 22: according to the characteristic matrix X d Constructing supersides by using KNN nearest neighbor algorithm, and splicing all the supersides to form an association matrix H d Expressed as
Figure FDA0004133697210000029
Wherein the method comprises the steps of
Figure FDA00041336972100000213
Is a matrix H d The j th superside of (2);
semantic hypergraph for constructing historical traffic states according to hyperedges
Figure FDA00041336972100000210
Wherein->
Figure FDA00041336972100000211
Representing a hyperedge set in a history semantic hypergraph;
step 23: based on the extracted adjacency matrix A g Construction of a second-order spatial adjacency-based spatial topological hypergraph
Figure FDA00041336972100000212
Second order spatial correlation for reflecting traffic conditions;
for adjacency matrix A g Is regarded as a superside comprising only two nodes, thereby constructing a space topology supergraph
Figure FDA0004133697210000031
The hypergraph is composed of a +.>
Figure FDA0004133697210000032
Is the correlation matrix H of (1) g Description of the drawings;
Figure FDA0004133697210000033
wherein the method comprises the steps of
Figure FDA0004133697210000034
Is H g The j th superside of (2) represents the second-order correlation, epsilon, between different traffic areas g Representing a set of hyperedges in a spatial topology hypergraph.
4. The urban traffic hypergraph convolution prediction method integrating the multi-mode high-order semantic correlation according to claim 3, wherein a new hypergraph obtained by splicing three hypergraphs is used as an input of a depth network to carry out hypergraph convolution, and the specific method comprises the following steps:
step 31: the space topology hypergraph obtained in the step 3, the recent traffic state semantic hypergraph and the historical traffic state semantic hypergraph are processedSplicing the corresponding incidence matrixes to obtain a fusion hypergraph fusing the multi-mode high-order semantic relativity
Figure FDA0004133697210000035
The fusion hypergraph->
Figure FDA00041336972100000314
From a dimension of->
Figure FDA0004133697210000036
Is described by an association matrix H; splicing the recent traffic state feature matrix established in the step 1 with the historical traffic state feature matrix to form a comprehensive traffic state feature matrix +.>
Figure FDA0004133697210000037
Step 32: taking the incidence matrix H and the feature matrix X obtained in the step 31 as deep network input, and performing hypergraph convolution;
the depth network is formed by stacking a plurality of hypergraph convolution layers;
the calculation formula of each hypergraph convolution layer is as follows:
Figure FDA0004133697210000038
wherein the method comprises the steps of
Figure FDA0004133697210000039
Is the input signal of the first layer in the hypergraph, is->
Figure FDA00041336972100000310
Is a parameter matrix of the first layer, sigma is a nonlinear activation function, D v ,D e And W is a node degree matrix, a superside degree matrix and a superside weight matrix respectively.
5. The urban traffic hypergraph convolution prediction method with fusion of multimode high-order semantic relevance according to claim 1, wherein the training set is used for training a prediction model, errors between a predicted value and a true value are minimized, and the specific method for calculating the optimal parameters of the prediction model is as follows:
constructing a training set according to traffic state observed values of N traffic areas in T time periods
Figure FDA00041336972100000311
Wherein->
Figure FDA00041336972100000312
Is a feature matrix->
Figure FDA00041336972100000313
For predictive labels, p is the number of features;
inputting the training set into a prediction model for training, wherein the model training objective function is as follows:
Figure FDA0004133697210000041
wherein, the first term is a predicted mean square error empirical loss term, the second term is a model parameter regularization term, f:
Figure FDA0004133697210000042
representing hypergraph ++>
Figure FDA0004133697210000048
The upper parameter is the predictive model of Θ, +.>
Figure FDA0004133697210000043
As a loss function, m=t-p is the number of samples.
After model training, the feature matrix at the current moment is obtained
Figure FDA0004133697210000044
Inputting the value to the optimal model obtained after training, and obtaining the prediction of the traffic condition at the next moment as follows:
Figure FDA0004133697210000045
wherein the method comprises the steps of
Figure FDA0004133697210000046
For the feature matrix at the current moment, < > and->
Figure FDA0004133697210000047
And the parameters are the learned optimal model parameters.
CN202310268121.3A 2023-03-20 2023-03-20 Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation Active CN116363874B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310268121.3A CN116363874B (en) 2023-03-20 2023-03-20 Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310268121.3A CN116363874B (en) 2023-03-20 2023-03-20 Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation

Publications (2)

Publication Number Publication Date
CN116363874A true CN116363874A (en) 2023-06-30
CN116363874B CN116363874B (en) 2024-04-23

Family

ID=86934952

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310268121.3A Active CN116363874B (en) 2023-03-20 2023-03-20 Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation

Country Status (1)

Country Link
CN (1) CN116363874B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116845889A (en) * 2023-09-01 2023-10-03 东海实验室 Hierarchical hypergraph neural network-based power load prediction method

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110161261A1 (en) * 2009-12-28 2011-06-30 Nec(China) Co., Ltd. Method and system for traffic prediction based on space-time relation
US20200234581A1 (en) * 2019-01-17 2020-07-23 International Business Machines Corporation Vehicle traffic information analysis and traffic jam management
US20210233391A1 (en) * 2018-10-18 2021-07-29 Alibaba Group Holding Limited Method and device for predicting traffic flow or travel time period
CN114220271A (en) * 2021-12-21 2022-03-22 南京理工大学 Traffic flow prediction method, equipment and storage medium based on dynamic space-time graph convolution cycle network
CN114495500A (en) * 2022-01-26 2022-05-13 北京工业大学 Traffic prediction method based on dual dynamic space-time diagram convolution
CN114944053A (en) * 2022-03-16 2022-08-26 浙江工业大学 Traffic flow prediction method based on spatio-temporal hypergraph neural network

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110161261A1 (en) * 2009-12-28 2011-06-30 Nec(China) Co., Ltd. Method and system for traffic prediction based on space-time relation
US20210233391A1 (en) * 2018-10-18 2021-07-29 Alibaba Group Holding Limited Method and device for predicting traffic flow or travel time period
US20200234581A1 (en) * 2019-01-17 2020-07-23 International Business Machines Corporation Vehicle traffic information analysis and traffic jam management
CN114220271A (en) * 2021-12-21 2022-03-22 南京理工大学 Traffic flow prediction method, equipment and storage medium based on dynamic space-time graph convolution cycle network
CN114495500A (en) * 2022-01-26 2022-05-13 北京工业大学 Traffic prediction method based on dual dynamic space-time diagram convolution
CN114944053A (en) * 2022-03-16 2022-08-26 浙江工业大学 Traffic flow prediction method based on spatio-temporal hypergraph neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
张永凯 等: "面向交通流量预测的时空超关系图卷积网络", 《计算机应用》, vol. 41, no. 12, 10 December 2021 (2021-12-10), pages 3578 - 3584 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116845889A (en) * 2023-09-01 2023-10-03 东海实验室 Hierarchical hypergraph neural network-based power load prediction method
CN116845889B (en) * 2023-09-01 2023-12-22 东海实验室 Hierarchical hypergraph neural network-based power load prediction method

Also Published As

Publication number Publication date
CN116363874B (en) 2024-04-23

Similar Documents

Publication Publication Date Title
CN112925989B (en) Group discovery method and system of attribute network
CN116363874B (en) Urban traffic hypergraph convolution prediction method integrating multimode high-order semantic correlation
CN114944053B (en) Traffic flow prediction method based on space-time hypergraph neural network
He et al. A three-stage automated modal identification framework for bridge parameters based on frequency uncertainty and density clustering
CN112614336B (en) Traffic flow modal fitting method based on quantum random walk
CN112187554B (en) Operation and maintenance system fault positioning method and system based on Monte Carlo tree search
CN106203628A (en) A kind of optimization method strengthening degree of depth learning algorithm robustness and system
CN114970715A (en) Variable working condition fault diagnosis method and system under small sample and unbalanced data constraint
CN114118375A (en) Continuous dynamic network characterization learning method based on time sequence diagram Transformer
CN111582468B (en) Photoelectric hybrid intelligent data generation and calculation system and method
Yao et al. A stability criterion for discrete-time fractional-order echo state network and its application
Liang et al. Remaining useful life prediction via a deep adaptive transformer framework enhanced by graph attention network
Zhang et al. Ctfnet: Long-sequence time-series forecasting based on convolution and time–frequency analysis
CN116524197B (en) Point cloud segmentation method, device and equipment combining edge points and depth network
CN116992334A (en) Academic-oriented network node classification method and device
CN115374191B (en) Multi-source data-driven cluster method for heterogeneous equipment of data center
CN115828990A (en) Time-space diagram node attribute prediction method for fused adaptive graph diffusion convolution network
Hu et al. Time-dependent reliability analysis in operation: prognostics and health management
CN115130663A (en) Heterogeneous network attribute completion method based on graph neural network and attention mechanism
Borge-Holthoefer et al. Topological versus dynamical robustness in a lexical network
CN117496702A (en) Urban traffic prediction method based on hypergraph neural network under structural data loss
Zhang et al. Traffic flow forecasting of graph convolutional network based on spatio-temporal attention mechanism
Huang et al. Multi-scale aggregation with self-attention network for modeling electrical motor dynamics
Verhaegen et al. Data-Driven Identification of Networks of Dynamic Systems
Yang et al. A high-order tensor completion algorithm based on fully-connected tensor network weighted optimization

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