CN113689052A - Travel demand prediction method based on tensor product neural network - Google Patents
Travel demand prediction method based on tensor product neural network Download PDFInfo
- Publication number
- CN113689052A CN113689052A CN202111036058.8A CN202111036058A CN113689052A CN 113689052 A CN113689052 A CN 113689052A CN 202111036058 A CN202111036058 A CN 202111036058A CN 113689052 A CN113689052 A CN 113689052A
- Authority
- CN
- China
- Prior art keywords
- matrix
- vector
- travel demand
- time
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 23
- 239000013598 vector Substances 0.000 claims abstract description 126
- 239000011159 matrix material Substances 0.000 claims abstract description 124
- 230000004927 fusion Effects 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims abstract description 5
- 230000002123 temporal effect Effects 0.000 claims description 20
- 230000004913 activation Effects 0.000 claims description 8
- 239000000126 substance Substances 0.000 claims description 7
- 230000009471 action Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000004891 communication Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 10
- 238000004364 calculation method Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000007619 statistical method Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- 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"
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06315—Needs-based resource requirements planning or analysis
-
- 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
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Abstract
The invention relates to the field of intelligent transportation, in particular to a travel demand prediction method based on a tensor product neural network, which comprises the following steps: dividing a road network into a plurality of areas and constructing graph data; establishing a travel demand characteristic matrix X at the t moment according to the divided two dimensions of the region and the time period(t)(ii) a Establishing a travel demand characteristic matrix X 'at the time t according to the divided two dimensions of the area and the date'(t)(ii) a Mixing X(t)And X'(t)Inputting the space characteristic vector and the time characteristic vector into a graph convolution neural network as input; when saidInputting the inter-feature vector and the space feature vector into a tensor product neural network unit for fusion, and outputting a traffic travel demand; and training to minimize the error between the predicted travel demand and the actual travel demand to obtain the network parameters. The travel demand is accurately predicted through the method.
Description
Technical Field
The invention relates to the field of intelligent transportation, in particular to a travel demand prediction method based on a tensor product neural network.
Background
The rapid development of road traffic in China brings convenience for traveling, and the demand of people on traveling is increased. The travel demand which is possibly generated in the area is accurately predicted, traffic resources can be distributed in advance, traffic supply and demand pressure is relieved, and important monitoring and important management and control of managers can be facilitated.
The existing traffic travel demand prediction method mainly comprises a statistical method, a machine learning method, a deep learning method and the like. Due to the fact that the trip data are difficult to collect comprehensively and strong in data sparsity, certain errors may exist when the traditional statistical method is used for predicting the trip demand, and the traditional statistical method can only be used for predicting in a conventional state. Traffic travel data naturally have space-time attributes, and the machine learning method can mine space-time characteristics to achieve the purpose of travel demand prediction, but prediction accuracy is more dependent on characteristic selection. The deep learning method can predict travel demands relatively accurately, but most models essentially divide time dependence and space dependence mining into two independent processes without considering the association between the two processes. The characteristics represented by the traffic travel data generally change in synchronization with the time dimension and the space dimension, and a certain correlation exists between the time characteristics and the space characteristics. Therefore, how to characterize the correlation between the spatio-temporal features and make full use of the correlation in the deep learning model is a key problem to be solved.
Disclosure of Invention
The trip demand prediction method based on the tensor product neural network is provided based on the above requirements in the prior art, and the technical problem to be solved is to provide the trip demand prediction method based on the tensor product neural network so as to realize more accurate trip demand prediction.
In order to solve the above problem, the technical scheme provided by the patent comprises:
the travel demand prediction method based on the tensor product neural network comprises the following steps: s1, dividing the road network into a plurality of areas and constructing graph data, wherein the divided areas cover the whole road network, and the graph data comprises a set V representing the divided areas, a set E representing the edges connecting each node and an adjacent matrix A representing the communication condition between the divided areas; s2, dividing the time of day into T time periods, and establishing a travel demand characteristic matrix X at the time T according to the divided areas and two dimensions of the time periods(t)(ii) a Establishing a travel demand characteristic matrix X 'at the time t according to the divided two dimensions of the area and the date'(t)(ii) a S3, combining the adjacency matrix A and the travel demand characteristic matrix X(t)Inputting the space characteristic vector into a graph convolution neural network as input to obtain a space characteristic vector
Wherein H(l)Denotes the convolution result of the l-th layer, σ is the activation function, and H is the activation function when l is 1(l)＝X(t)，
Is a matrix of the degrees, and the degree matrix,
representation matrix
The ith row and the jth column of (1), wherein the matrix
Is a matrix with self-connection added, INIs an identity matrix, and W is a trainable parameter; mixing the adjacent matrix A 'and a feature matrix X'(t)As input, inputting into graph convolution neural network to obtain time characteristic directionAn amount;
wherein H(l)′Denotes the convolution result of the l-th layer, when l is 1, H(l)′＝X′(t)，
Is a matrix of the degrees, and the degree matrix,
representation matrix
Row i and column j of the matrix
Is a matrix with self-connection added, and W' is a trainable parameter; s4, inputting the time characteristic vector and the space characteristic vector into a tensor product neural network unit for fusion, and outputting a transportation travel demand; and S5, inputting the trip data into the neural network in batches, and training to minimize the error between the predicted trip demand and the actual trip demand to obtain network parameters.
Preferably, the S4 includes S401, obtaining an initial temporal attention vector and a spatial attention vector through a random initialization mode; s402, obtaining a time coding vector according to the time attention vector and the time feature vector; obtaining a space coding vector according to the space attention vector and the space feature vector; s403, calculating according to the time coding vector and the space coding vector after the attention vector is fused to obtain the travel demand; and S404, updating the time attention vector and the space attention vector at the next moment according to the travel demand.
Preferably, the set V is represented by V ═ (V)1，v2，v3，…，vN) Representing, wherein N is the number of divided areas;the set E is defined by E ═ EijI 1 ≦ i, j ≦ N, the graph data is directed graph G, G ═ V, E, a, where,
representing the adjacency matrix of diagram G, the element a in the adjacency matrixijRepresenting the connectivity between region i and region j: if the two regions are connected, then aij1 is ═ 1; if the two regions are not connected, then aij＝0。
Preferably, the adjacency matrix a and the travel demand feature matrix X are combined(t)Inputting the space characteristic vector into the graph convolution neural network to obtain a space characteristic vector
Wherein H(l)Denotes the convolution result of the l-th layer, σ is the activation function, and H is the activation function when l is 1(l)＝X(t)，
Is a matrix of the degrees, and the degree matrix,
representation matrix
The ith row and the jth column of (1), wherein the matrix
Is a matrix with self-connection added, INIs an identity matrix, and W is a trainable parameter; combining the adjacency matrix A' and the travel demand characteristic matrix X(t)' input into the graph convolution neural network to obtain the time characteristics
Wherein H(l)′Denotes the convolution result of the l-th layer when l ═1 hour, H(l)′＝X′(t)，
Is a matrix of the degrees, and the degree matrix,
representation matrix
Row i and column j of the matrix
Is a matrix with self-join added, W' is a trainable parameter.
Preferably, a temporal coding vector is derived from the temporal attention vector and the temporal feature vector, and the temporal coding vector can be expressed as:
wherein the content of the first and second substances,
a time attention vector, which indicates the multiplication of the corresponding elements; deriving a temporal coding vector from the spatial attention vector and the spatial feature vector,
wherein the content of the first and second substances,
is the spatial attention vector.
Preferably, the travel demand coding feature at the time t is expressed by a tensor product of the time coding vector and the space coding vector at the time t
Wherein the operation sign
Expressing tensor product, expressing multiplication of each element in left matrix and right matrix, expressing connection matrix between time characteristic vector and space characteristic vector by matrix C, encoding soft connection relation between time characteristic and space characteristic, and the dimension of matrix C is
Traffic travel demand coding characteristic V(t)With the same dimensions.
Preferably, the time attention vector of the next moment is obtained from the travel demand
Wherein f () is a logistic sigmoid function,
is shown as acting on p(t)The matrix of coefficients of (a) is,
indicating action on vec (V)(t)) Vec () represents a vectorization operation, with matrix V(t)Conversion into vectors, bTRepresenting a bias vector;
and spatial attention vector
Wherein the content of the first and second substances,
to act on s(t)The matrix of coefficients of (a) is,
to act on vec (V)(t)) Coefficient matrix of bSRepresenting a bias vector.
Preferably, the expression minimizing the error between the predicted travel demand and the actual travel demand is:
y represents the actual travel demand at the next moment,
the travel demand predicted by the neural network at the next moment is represented, | | | | represents a two-norm, and parameters in the network are updated by minimizing a Loss function.
Compared with the prior art, the time characteristic and the space characteristic are associated in a tensor product mode, so that the time and space dependency relationship can be better captured by the deep learning network model, and the travel demand can be more accurately predicted.
Drawings
In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the embodiments of the present specification, and other drawings can be obtained by those skilled in the art according to the drawings.
Fig. 1 is a flow chart illustrating steps of a travel demand prediction method based on a tensor product neural network according to the present invention;
FIG. 2 is a flowchart illustrating the step of S4 according to the present invention;
fig. 3 is a network element of the tensor product neural network of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
For the purpose of facilitating understanding of the embodiments of the present application, the following description will be made in terms of specific embodiments with reference to the accompanying drawings, which are not intended to limit the embodiments of the present application.
Example 1
The embodiment provides a travel demand prediction method based on a tensor product neural network, and reference is made to fig. 1.
The travel demand prediction method based on the tensor product neural network comprises the following steps:
s1, dividing the road network into a plurality of regions, and constructing graph data, the divided regions covering the entire road network, the graph data including a set V representing the divided regions, a set E representing edges connecting the nodes, and an adjacency matrix a representing the connectivity between the divided regions.
The whole road network is divided into a plurality of areas according to the prediction requirement, the areas can be regular gridding areas or irregular polygonal areas, and the divided areas need to cover the whole road network.
In the present embodiment, a 300m × 300m grid is used to divide the road network into a plurality of regions.
Dividing the plurality of divided regions into a set V ═ V1,v2,v3,…,vN) Representing, wherein N is the number of divided areas; setting a set E for edges connecting each region node to { E }ijI is less than or equal to 1 and N is less than or equal to j.
Constructing a road network directed graph G, G ═ V, E, A), wherein
Representing the adjacency matrix of diagram G, the element a in the adjacency matrixijRepresenting the connectivity between region i and region j: if the two regions are connected, then aij1 is ═ 1; if the two regions are not connected, then aij＝0。
S2, dividing the time of day into T time periods, and establishing a travel demand characteristic matrix X at the time T according to the divided areas and two dimensions of the time periods(t)(ii) a Establishing a travel demand characteristic matrix X 'at the time t according to the divided two dimensions of the area and the date'(t)。
A day is divided into T time sections, and the amount of traffic generated by each area is calculated for each time section in turn.
In the present embodiment, one day is divided into 48 time periods, 30 minutes are taken as time intervals, and the amount of traffic in each time period is counted.
Establishing a travel demand characteristic matrix X by two dimensions of the divided region and the divided time(t)Each element x in the matrixijRepresents the number of I areas going out in the j time period in the first day, wherein j is less than or equal to T.
And counting a daily travel demand characteristic matrix in the historical data for training according to the prediction demand and the data quantity.
Slicing the counted travel demand feature matrix according to the dimension of the time period to obtain a travel demand feature matrix X 'with two dimensions of area and date'(t)X 'of each element in the matrix'ijIndicating the number of trips of the ith area within a certain period of time within the jth day.
Each element a 'of the adjacent matrix A'ijRepresenting connection condition of elements within region connection and date difference of not more than 1 day, if connected'ij1, otherwise'ij＝0。
S3, combining the adjacency matrix A and the travel demand characteristic matrix X(t)Inputting the space characteristic vector into a graph convolution neural network as input to obtain a space characteristic vector; mixing the adjacent matrix A 'and a feature matrix X'(t)The time characteristic vector is input into a graph convolution neural network as input to obtain a time characteristic vector.
The adjacency matrix A and the travel demand characteristic matrix X are combined(t)Inputting into the graph convolution neural network.
The graph convolution calculation is represented by the following equation:
wherein H(l)And H(l+1)Denotes the convolution results of the l-th and l + 1-th layers, respectively, σ is the activation function, and H is the activation function when l is 1(l)＝X(t)。
Is a matrix of the degrees, and the degree matrix,
representation matrix
The ith row and the jth column of (1), wherein the matrix
Is a matrix with self-connection added, INIs an identity matrix, WlIs a trainable parameter of a level l.
The adjacent matrix A 'and the travel demand characteristic matrix X'(t)Inputting into the graph convolution neural network.
The graph convolution calculation is represented by the following equation:
wherein H(l)′And H(l+1)′Denotes the convolution results of the l-th and l + 1-th layers, respectively, and H is H when l is 1(l)′＝X′(t)。
Is a matrix of the degrees, and the degree matrix,
representation matrix
Row i and column j of the matrix
Is a matrix with self-connection added, INIs an identity matrix, Wl′Is a trainable parameter of a level l.
In this embodiment, the calculation is performed using a one-layer graph convolution neural network.
Using the adjacent matrix A and the travel demand characteristic matrix X at the time t(t)Inputting the space characteristic vector into the graph convolution neural network to obtain a space characteristic vector at the time t
Wherein X(t)And (4) a travel demand characteristic matrix at the time t.
The adjacent matrix A ' and the travel demand characteristic matrix X ' at the time t are combined '(t)Inputting the time characteristic vector into the graph convolution neural network to obtain the time characteristic vector at the time t
Wherein X'(t)And (4) a travel demand characteristic matrix at the time t.
And S4, inputting the time characteristic vector and the space characteristic vector into a tensor product neural network unit for fusion, and outputting the travel demand, referring to FIGS. 2 and 3.
S401, obtaining an initial time attention vector and a space attention vector through a random initialization mode.
S402, obtaining a time coding vector according to the time attention vector and the time feature vector; and obtaining a spatial coding vector according to the spatial attention vector and the spatial feature vector.
Deriving a temporal coding vector from the temporal attention vector and the temporal feature vector, the temporal coding vector q(t)Can be expressed as:
wherein q is(t)A time encoded vector, a indicates a multiplication of corresponding elements,
a temporal feature vector representing the time of the t instant,
representing the temporal attention vector at time t.
Deriving a time-coded vector s from the spatial attention vector and the spatial feature vector(t)Can be expressed as:
wherein s is(t)In order to spatially encode the vector(s),
is the spatial feature vector at time t,
is the spatial attention vector at time t.
And S403, calculating according to the time coding vector and the space coding vector after the attention vector is fused to obtain the travel demand.
And the transport travel demand at the time t is expressed by the tensor product of the time coding vector and the space coding vector at the time t. The traffic travel demand coding features are expressed as follows:
wherein the operation sign
A tensor product is represented, representing the multiplication of each element in the left matrix by the right matrix. The matrix C represents a connection matrix between the temporal and spatial eigenvectors that encodes the soft-connected relationship between the spatio-temporal features. The dimension of the matrix C is
Same traffic travel demand coding characteristic V(t)Features also have the same dimensions.
And S404, updating the time attention vector and the space attention vector at the next moment according to the travel demand.
The time attention vector at the next moment is obtained according to the travel demand, and the calculation formula is as follows:
wherein f () is a logistic sigmoid function,
is shown as acting on p(t)The matrix of coefficients of (a) is,
indicating action on vec (V)(t)) Vec () represents a vectorization operation, with matrix V(t)And converting into a vector. bTShows a deviationAnd (5) setting a vector.
The formula for calculating the spatial attention vector is as follows:
wherein the content of the first and second substances,
to act on s(t)The matrix of coefficients of (a) is,
to act on vec (V)(t)) Coefficient matrix of bSRepresenting a bias vector.
And S5, inputting the time characteristic vector and the space characteristic vector into a tensor product neural network unit as input for fusion, and outputting the travel demand.
Parameters in the tensor product network are driven by back propagation to minimize the error between the predicted and actual travel demands.
The network parameters include
bT，
bSW and W', etc.
Minimizing the error formula between the predicted travel demand and the actual travel demand is as follows:
y represents the actual travel demand at the next moment,
the travel demand predicted by the neural network at the next moment is represented, and | | | | represents a two-norm. By passingAnd minimizing the Loss function to update parameters in the network, thereby realizing training.
The neural network includes a convolutional neural network and a tensor product neural network.
The input of the graph convolution neural network is to establish a travel demand characteristic matrix X according to two dimensions of the divided regions and the time periods(t)And establishing a travel demand characteristic matrix X 'according to the two dimensions of the divided areas and dates'(t)Its output is a temporal feature vector dtAnd spatial feature vector ds。
The input of the tensor product network is a time eigenvector dtAnd spatial feature vector dsThe output is the traffic travel demand at the next moment of each region of the road network
The above-mentioned embodiments, objects, technical solutions and advantages of the present application are described in further detail, it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present application, and are not intended to limit the scope of the present application, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present application should be included in the scope of the present application.
Claims (7)
1. A travel demand prediction method based on a tensor product neural network is characterized by comprising the following steps:
s1, dividing the road network into a plurality of areas and constructing graph data, wherein the divided areas cover the whole road network, and the graph data comprises a set V representing the divided areas, a set E representing the edges connecting each node and an adjacent matrix A representing the communication condition between the divided areas;
s2, dividing the time of day into T time periods, and establishing a travel demand characteristic matrix X at the time T according to the divided areas and two dimensions of the time periods(t)(ii) a Establishing travel demand characteristics of t moment according to two dimensions of divided region and dateSign matrix X'(t)；
S3, combining the adjacency matrix A and the travel demand characteristic matrix X(t)Inputting the space characteristic vector into a graph convolution neural network as input to obtain a space characteristic vector
Wherein H(l)Denotes the convolution result of the l-th layer, σ is the activation function, and H is the activation function when l is 1(l)＝X(t)，
Is a matrix of the degrees, and the degree matrix,
representation matrix
The ith row and the jth column of (1), wherein the matrix
Is a matrix with self-connection added, INIs an identity matrix, and W is a trainable parameter; mixing the adjacent matrix A 'and a feature matrix X'(t)Inputting the time characteristic vector into a graph convolution neural network as input to obtain a time characteristic vector;
wherein H(L)' denotes the convolution result of the l-th layer, and when l is 1, H(L)′＝X′(t)，
Is a matrix of the degrees, and the degree matrix,
representation matrix
Row i and column j of the matrix
Is a matrix with self-connection added, and W' is a trainable parameter;
s4, inputting the time characteristic vector and the space characteristic vector into a tensor product neural network unit for fusion, and outputting a transportation travel demand;
and S5, inputting the trip data into the neural network in batches, and training to minimize the error between the predicted trip demand and the actual trip demand to obtain network parameters.
2. A travel demand prediction method based on a tensor product neural network as claimed in claim 1, wherein the S4 includes S401, obtaining an initial temporal attention vector and a spatial attention vector by a random initialization mode; s402, obtaining a time coding vector according to the time attention vector and the time feature vector; obtaining a space coding vector according to the space attention vector and the space feature vector; s403, calculating according to the time coding vector and the space coding vector after the attention vector is fused to obtain the travel demand; and S404, updating the time attention vector and the space attention vector at the next moment according to the travel demand.
3. The method for predicting travel demand based on tensor product neural network as claimed in claim 1, wherein the set V is defined by V ═ (V ═ V)1，v2，v3，…，vN) Representing, wherein N is the number of divided areas; the set E is defined by E ═ EijI 1 ≦ i, j ≦ N ≦ and the graph data is directed graph G, G ═ V, E, a, whichIn (1),
representing the adjacency matrix of diagram G, the element a in the adjacency matrixijRepresenting the connectivity between region i and region j: if the two regions are connected, then aij1 is ═ 1; if the two regions are not connected, then aij＝0。
4. The method for predicting travel demand based on tensor product neural network as claimed in claim 1, wherein a temporal coding vector is obtained according to the temporal attention vector and the temporal feature vector, and the temporal coding vector can be expressed as:
wherein the content of the first and second substances,
a time attention vector, which indicates the multiplication of the corresponding elements; deriving a temporal coding vector from the spatial attention vector and the spatial feature vector,
wherein the content of the first and second substances,
is the spatial attention vector.
5. The travel demand prediction method based on the tensor product neural network as claimed in claim 1, wherein the transport travel demand coding feature at the time t is expressed by a tensor product of a time coding vector and a space coding vector at the time t
Wherein the operation sign
Expressing tensor product, expressing multiplication of each element in left matrix and right matrix, expressing connection matrix between time characteristic vector and space characteristic vector by matrix C, encoding soft connection relation between time characteristic and space characteristic, and the dimension of matrix C is
Traffic travel demand coding characteristic V(t)With the same dimensions.
6. The travel demand prediction method based on the tensor product neural network as claimed in claim 1, wherein a time attention vector at the next moment is obtained from a traffic travel demand
Wherein f () is a logistic sigmoid function,
is shown as acting on p(t)The matrix of coefficients of (a) is,
indicating action on vec (V)(t)) Vec () represents a vectorization operation, with matrix V(t)Conversion into vectors, bTRepresenting a bias vector;
and spatial attention vector
Wherein the content of the first and second substances,
to act on s(t)The matrix of coefficients of (a) is,
to act on vec (V)(t)) Coefficient matrix of bSRepresenting a bias vector.
7. The method for predicting the travel demand based on the tensor product neural network as claimed in claim 1, wherein the expression for minimizing the error between the predicted travel demand and the actual travel demand is as follows:
y represents the actual travel demand at the next moment,
the travel demand predicted by the neural network at the next moment is represented, | | | | represents a two-norm, and parameters in the network are updated by minimizing a Loss function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111036058.8A CN113689052A (en) | 2021-09-06 | 2021-09-06 | Travel demand prediction method based on tensor product neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111036058.8A CN113689052A (en) | 2021-09-06 | 2021-09-06 | Travel demand prediction method based on tensor product neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113689052A true CN113689052A (en) | 2021-11-23 |
Family
ID=78585295
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111036058.8A Pending CN113689052A (en) | 2021-09-06 | 2021-09-06 | Travel demand prediction method based on tensor product neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113689052A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116245183A (en) * | 2023-02-28 | 2023-06-09 | 清华大学 | Traffic scene generalization understanding method and device based on graph neural network |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2017220000A (en) * | 2016-06-07 | 2017-12-14 | 日本電信電話株式会社 | Flow rate prediction device, pattern estimation device, flow rate prediction method, pattern estimation method and program |
| CN109754605A (en) * | 2019-02-27 | 2019-05-14 | 中南大学 | A kind of traffic forecast method based on attention temporal diagram convolutional network |
| CN110264709A (en) * | 2019-05-06 | 2019-09-20 | 北京交通大学 | The prediction technique of the magnitude of traffic flow of road based on figure convolutional network |
| CN110674987A (en) * | 2019-09-23 | 2020-01-10 | 北京顺智信科技有限公司 | Traffic flow prediction system and method and model training method |
| US20200151580A1 (en) * | 2018-11-13 | 2020-05-14 | International Business Machines Corporation | Generating and managing deep tensor neural networks |
| CN111223301A (en) * | 2020-03-11 | 2020-06-02 | 北京理工大学 | Traffic flow prediction method based on graph attention convolution network |
| CN111292562A (en) * | 2020-05-12 | 2020-06-16 | 北京航空航天大学 | Aviation flow prediction method |
| CN112350899A (en) * | 2021-01-07 | 2021-02-09 | 南京信息工程大学 | Network flow prediction method based on graph convolution network fusion multi-feature input |
| US20210064999A1 (en) * | 2019-08-29 | 2021-03-04 | Nec Laboratories America, Inc. | Multi-scale multi-granularity spatial-temporal traffic volume prediction |
| CN112734100A (en) * | 2020-12-31 | 2021-04-30 | 北京航空航天大学 | Road network travel time prediction method based on tensor neural network |
| CN112801404A (en) * | 2021-02-14 | 2021-05-14 | 北京工业大学 | Traffic prediction method based on self-adaptive spatial self-attention-seeking convolution |
| WO2021109318A1 (en) * | 2019-12-03 | 2021-06-10 | 东南大学 | Method for estimating and predicting short-term traffic circulation state of urban road network |
-
2021
- 2021-09-06 CN CN202111036058.8A patent/CN113689052A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2017220000A (en) * | 2016-06-07 | 2017-12-14 | 日本電信電話株式会社 | Flow rate prediction device, pattern estimation device, flow rate prediction method, pattern estimation method and program |
| US20200151580A1 (en) * | 2018-11-13 | 2020-05-14 | International Business Machines Corporation | Generating and managing deep tensor neural networks |
| CN109754605A (en) * | 2019-02-27 | 2019-05-14 | 中南大学 | A kind of traffic forecast method based on attention temporal diagram convolutional network |
| CN110264709A (en) * | 2019-05-06 | 2019-09-20 | 北京交通大学 | The prediction technique of the magnitude of traffic flow of road based on figure convolutional network |
| US20210064999A1 (en) * | 2019-08-29 | 2021-03-04 | Nec Laboratories America, Inc. | Multi-scale multi-granularity spatial-temporal traffic volume prediction |
| CN110674987A (en) * | 2019-09-23 | 2020-01-10 | 北京顺智信科技有限公司 | Traffic flow prediction system and method and model training method |
| WO2021109318A1 (en) * | 2019-12-03 | 2021-06-10 | 东南大学 | Method for estimating and predicting short-term traffic circulation state of urban road network |
| CN111223301A (en) * | 2020-03-11 | 2020-06-02 | 北京理工大学 | Traffic flow prediction method based on graph attention convolution network |
| CN111292562A (en) * | 2020-05-12 | 2020-06-16 | 北京航空航天大学 | Aviation flow prediction method |
| CN112734100A (en) * | 2020-12-31 | 2021-04-30 | 北京航空航天大学 | Road network travel time prediction method based on tensor neural network |
| CN112350899A (en) * | 2021-01-07 | 2021-02-09 | 南京信息工程大学 | Network flow prediction method based on graph convolution network fusion multi-feature input |
| CN112801404A (en) * | 2021-02-14 | 2021-05-14 | 北京工业大学 | Traffic prediction method based on self-adaptive spatial self-attention-seeking convolution |
Non-Patent Citations (6)
| Title |
|---|
| XURAN XU, TONG ZHANG, CHUNYAN XU, ZHEN CUI, JIAN YANG: "Spatial-Temporal Tensor Graph Convolutional Network for Traffic Prediction", JOURNAL OF LATEX CLASS FILES, vol. 14, no. 8, pages 1 - 10 * |
| ZHENG FANG，QINGQING LONG，GUOJIE SONG，KUNQING XIE: "Spatial-Temporal Graph ODE Networks for Traffic Flow Forecasting", PROCEEDINGS OF THE 27TH ACM SIGKDD CONFERENCE ON KNOWLEDGE DISCOVERY & DATA MINING, pages 364 - 373 * |
| ZHENGCHAO ZHANG, MENG LI, XI LIN, YINHAI WANG, FANG HE: "Multistep speed prediction on traffic networks: A deep learning approach considering spatio-temporal dependencies", TRANSPORTATION RESEARCH PART C, pages 297 - 322 * |
| 向敏, 饶华阳, 张进进, 陈梦鑫: "基于图卷积神经网络的软件定义电力通信网络路由控制策略", 电子与信息学报, vol. 43, no. 2, pages 388 - 395 * |
| 柯文前;陆玉麒;朱宇;陈伟;杨青;: "交通流网络的时空特征解析――基于张量分解方法视角", 地理科学, no. 11, pages 78 - 86 * |
| 董伟;张磊磊;金子恒;孙伟;高俊波;: "基于多特征时空图卷积网络的水运通航密度预测", 物联网学报, no. 03, pages 82 - 89 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116245183A (en) * | 2023-02-28 | 2023-06-09 | 清华大学 | Traffic scene generalization understanding method and device based on graph neural network |
| CN116245183B (en) * | 2023-02-28 | 2023-11-07 | 清华大学 | Traffic scene generalization understanding method and device based on graph neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11842271B2 (en) | Multi-scale multi-granularity spatial-temporal traffic volume prediction | |
| CN109887282B (en) | Road network traffic flow prediction method based on hierarchical timing diagram convolutional network | |
| CN111639787B (en) | Spatio-temporal data prediction method based on graph convolution network | |
| CN114299723B (en) | Traffic flow prediction method | |
| CN115240425B (en) | Traffic prediction method based on multi-scale space-time fusion graph network | |
| CN110599766B (en) | Road traffic jam propagation prediction method based on SAE-LSTM-SAD | |
| KR102204337B1 (en) | Method for providing real-time construction estimator service using intuitive step-by-step choice | |
| CN112766597B (en) | Bus passenger flow prediction method and system | |
| CN109344992B (en) | Modeling method for user control behavior habits of smart home integrating time-space factors | |
| CN113762595B (en) | Traffic time prediction model training method, traffic time prediction method and equipment | |
| CN116468186B (en) | Flight delay time prediction method, electronic equipment and storage medium | |
| CN113112793A (en) | Traffic flow prediction method based on dynamic space-time correlation | |
| CN112446489A (en) | Dynamic network embedded link prediction method based on variational self-encoder | |
| CN113689052A (en) | Travel demand prediction method based on tensor product neural network | |
| CN115114128A (en) | Satellite health state evaluation system and evaluation method | |
| CN111898836A (en) | Crime space-time prediction method and system | |
| CN116311939A (en) | OD demand prediction method and system based on dynamic space-time correlation | |
| Amisigo et al. | Using a spatio-temporal dynamic state-space model with the EM algorithm to patch gaps in daily riverflow series | |
| Madhavi et al. | Multivariate deep causal network for time series forecasting in interdependent networks | |
| CN117610734A (en) | Deep learning-based user behavior prediction method, system and electronic equipment | |
| CN116777539A (en) | Hierarchical region structure diagram-based hair tendency prediction system and method for breeder chickens | |
| CN112883292B (en) | User behavior recommendation model establishment and position recommendation method based on spatio-temporal information | |
| CN116822722A (en) | Water level prediction method, system, device, electronic equipment and medium | |
| CN111260121B (en) | Urban-range pedestrian flow prediction method based on deep bottleneck residual error network | |
| CN114372627A (en) | Urban vehicle travel time estimation method based on hybrid deep learning framework |
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 |