CN111091712A - Traffic flow prediction method based on cyclic attention dual graph convolution network - Google Patents

Traffic flow prediction method based on cyclic attention dual graph convolution network Download PDF

Info

Publication number
CN111091712A
CN111091712A CN201911355366.XA CN201911355366A CN111091712A CN 111091712 A CN111091712 A CN 111091712A CN 201911355366 A CN201911355366 A CN 201911355366A CN 111091712 A CN111091712 A CN 111091712A
Authority
CN
China
Prior art keywords
node
graph
network
edge
traffic flow
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
Application number
CN201911355366.XA
Other languages
Chinese (zh)
Inventor
陈岭
陈纬奇
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.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
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 Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN201911355366.XA priority Critical patent/CN111091712A/en
Publication of CN111091712A publication Critical patent/CN111091712A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/042—Detecting movement of traffic to be counted or controlled using inductive or magnetic detectors
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/065—Traffic control systems for road vehicles by counting the vehicles in a section of the road or in a parking area, i.e. comparing incoming count with outgoing count

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a traffic flow prediction method based on a convolution network of a circular attention dual graph, which comprises the following steps: 1) constructing a traffic flow time sequence according to traffic flow data collected by road coil sensors on an urban traffic network, and preprocessing the traffic flow time sequence; 2) constructing a dual graph to represent a spatial dependency relationship, wherein the dual graph comprises a node graph and an edge graph, wherein a single sensor is regarded as a node, the node graph is constructed according to the road network distance between the sensors, edges in the node graph represent the relationship between the sensors, the edges in the node graph are regarded as the nodes to construct the edge graph, and the edges in the edge graph represent the mutual influence of the relationship between the sensors; 3) inputting the preprocessed traffic flow time sequence into a convolution network of the circular attention dual graph, and predicting the traffic flow of a future traffic network. The traffic flow prediction method can realize the prediction of the traffic flow of the traffic network, and has wide application prospect in the fields of travel planning, traffic management and the like.

Description

Traffic flow prediction method based on cyclic attention dual graph convolution network
Technical Field
The invention relates to the field of intelligent traffic systems, in particular to a traffic flow prediction method based on a convolution network of a circular attention dual map.
Background
With the continuous promotion of urbanization and industrialization, the number of automobiles is continuously increased, and urban traffic gradually becomes congested. The urban disease affects daily trips of people, and brings great challenges to urban road planning and traffic management of relevant departments. The method can effectively guide people to plan a travel route by accurately predicting the future traffic flow of the traffic network, and can also provide powerful data support for traffic management, so that the traffic flow prediction becomes an extremely valuable research direction in an intelligent traffic system.
Early traffic flow prediction methods mostly use traditional linear sequence models to predict the traffic flow of a single node, such as autoregressive moving average model (ARMIA), Kalman Filtering (Kalman Filtering), and the like. However, such methods ignore non-linear relationships in traffic flow data, and do not take into account spatial dependencies between multiple nodes.
In order to model complex space-time dependency, researchers have proposed deep learning based traffic flow prediction methods. Some methods model temporal dependencies using Recurrent Neural Networks (RNNs) while modeling spatial dependencies using Convolutional Neural Networks (CNNs). Such a traffic flow prediction method based on CNN can only represent the spatial dependency relationship in a regular grid structure, but the spatial dependency relationship between nodes is often represented as a non-european relationship subject to the constraint of an irregular traffic network. Therefore, the latest deep learning-based method uses a Graph structure to represent the spatial dependency relationship among nodes, edges in a Graph represent the connection relationship of the nodes in a traffic Network, and a Graph Convolutional Network (GCNs) is introduced to aggregate information of a certain node and its neighbor nodes, so as to model the non-european dependency relationship, and the prediction accuracy is higher compared with the CNN-based method.
However, most of the recent traffic flow prediction methods based on the GCN use an unweighted graph or a weighted graph with fixed weight to represent the relationship between nodes, thereby excessively simplifying the complex spatial dependency relationship in the actual traffic network. Additionally, these methods aggregate information within a given neighborhood range (e.g., nodes within a k-hop range), however, different neighborhood ranges tend to exhibit different traffic characteristics, e.g., neighbors within a small range can represent local spatial dependencies and neighbors within a large range tend to reflect overall traffic patterns over a relatively large area. The existing method ignores the influence of different neighbor ranges and cannot model a multi-range spatial dependency relationship.
Disclosure of Invention
The invention aims to solve the technical problem of how to model the complex space-time dependency relationship in traffic flow data, provides a traffic flow prediction method based on a convolution network of a circular attention dual graph, and aims to solve the defects of the prior art in the background technology.
In order to solve the above problems, the present invention provides a traffic flow prediction method based on a convolutional network of a cyclic attention dual graph, which comprises the following steps:
step 1, constructing a traffic flow time sequence according to traffic flow data collected by road coil sensors on an urban traffic network, and preprocessing the traffic flow time sequence;
step 2, calculating the road network distance between the coil sensors, and constructing a node map according to the road network distance
Figure BDA0002335757840000021
Wherein, VnIs a set of nodes, EnIs a set of edges, AnIs a contiguous matrix;
step 3, node map
Figure BDA0002335757840000022
Edge E innDefining two edge influence modes of upstream and downstream connection relation and competition relation as the nodes of the edge graph, constructing the edges of the edge graph according to the two edge influence modes, and then constructing an edge graph Ge＝(Ve,Ee,Ae) Wherein V iseIs a set of nodes, EeIs a set of edges, AeIs a contiguous matrix;
step 4, according to the node map
Figure BDA0002335757840000034
And edge graph GeConstructing a convolution network of a dual graph, and dividing the time tTraffic flow XtInputting k layers of the convolutional network of the dual graph, and expressing the node of each layer output as
Figure BDA0002335757840000031
Figure BDA0002335757840000032
Step 5, the output nodes of each layer of the convolution network of the dual graph represent input multi-range attention network for fusion, and the output fusion represents Ut；
Step 6, the past T is processed′Fused representation of individual moments U(s-T′+1):sThe coding and decoding structure long-time memory network shared among the input nodes outputs traffic flow predicted values at T moments in the future
Figure BDA0002335757840000033
According to the method, not only the dependency relationship among the nodes is considered when the spatial dependency relationship is modeled, but also the mutual influence of the relationships among the nodes is considered, and meanwhile, an attention mechanism is introduced to model the multi-range spatial dependency relationship. Compared with the prior method, the method has the advantages that:
1) and constructing a node graph according to the road network distance between the sensors, constructing an edge graph according to the edge influence mode, and modeling the relationship between the nodes and the relationship between the edges by using a dual graph convolution network display mode, thereby modeling a more complex spatial dependence relationship.
2) The multi-range attention mechanism can aggregate information of a plurality of neighbor ranges, and simultaneously learn a self-adaptive weight for different neighbor ranges, so that the expression capability of the model is further improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings 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 of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is an overall flowchart of a traffic flow prediction method based on a convolutional network of a cyclic attention dual graph according to an embodiment of the present invention;
fig. 2 is an edge influence mode provided by an embodiment of the present invention, in which (a) is an upstream and downstream connection relationship edge influence mode, and (b) is a competition relationship edge influence mode;
FIG. 3 is a block diagram of a convolutional trellis diagram for a dual graph according to an embodiment of the present invention;
fig. 4 is a schematic diagram illustrating a fusion of a multi-range attention network according to an embodiment of the present invention;
fig. 5 is a coding and decoding LSTM network according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is an overall flowchart of a traffic flow prediction method based on a convolutional network of a cyclic attention dual graph according to an embodiment of the present invention. The traffic flow prediction method is intended to realize the following tasks: let the current time be s, and according to the past T of N nodes on the traffic network′Traffic flow of every moment
Figure BDA0002335757840000041
Predicting traffic flow at T moments in the future
Figure BDA0002335757840000042
Figure BDA0002335757840000043
Referring to fig. 1, the traffic flow prediction method includes the steps of:
step 1, constructing a traffic flow time sequence according to traffic flow data collected by road coil sensors on an urban traffic network, and preprocessing the traffic flow time sequence.
Road coil sensors are widely used for road vehicle detection, vehicle type identification. Counting the number of vehicles detected by the coil sensor in a time window, and obtaining the traffic flow passing through the road node where the coil sensor is located in the time window. The number of coil sensors on the road network is N,
Figure BDA0002335757840000044
is represented by the formula (T ═ s-T)′+1,s-T′+2, …, s) time, and processing missing values and abnormal values for the traffic flow by using a linear interpolation method. Let the current time be s, count the past T′The traffic flow of each moment and the time sequence of the traffic flow is constructed
Figure BDA0002335757840000045
This is taken as input data.
Step 2, calculating the road network distance between the coil sensors, and constructing a node map according to the road network distance
Figure BDA0002335757840000059
Wherein, VnIs a set of nodes, EnIs a set of edges, AnIs a contiguous matrix;
wherein a node map is constructed
Figure BDA0002335757840000052
The method comprises the following specific steps:
(a) building a node graph
Figure BDA0002335757840000053
Node set V ofn＝{v1,v2,…,vNIn which, | VnN, node set VnOne element in (1) corresponds to one road node;
(b) calculating the road network distance between any two road nodes, dist (v)i,vj) The shortest road network distance from road node i to road node j is shown, and attention is paid to the fact that the road nodes areThe way tends to be directional, dist (v)i,vj)≠dist(vj,vi)；
(c) Calculation based on road network distance between nodes
Figure BDA0002335757840000054
Adjacent matrix A ofnWherein the adjacent matrix AnThe calculation formula of (2) is as follows:
Figure BDA0002335757840000051
wherein σ2D is a threshold value set manually, and when the shortest road network distance from a road node i to a road node j is greater than d, the graph is ignored
Figure BDA0002335757840000055
The edge from the middle road node i to the road node j, the guarantee graph
Figure BDA0002335757840000056
Sparsity of (3) to prevent over-fitting of the model.
(d) Building a node graph
Figure BDA0002335757840000057
Edge set E ofn＝{(i→j)|0≤i,j≤N,Ai,j>0, where (i → j) represents an edge with i as the head node and j as the tail node.
Step 3, node map
Figure BDA0002335757840000058
Edge E innDefining two edge influence modes of upstream and downstream connection relation and competition relation as the nodes of the edge graph, constructing the edges of the edge graph according to the two edge influence modes, and then constructing an edge graph Ge＝(Ve,Ee,Ae) Wherein V iseIs a set of nodes, EeIs a set of edges, AeIs a contiguous matrix;
wherein, an edge graph G is constructedeComprises the specific steps of：
(a) Constructing a boundary graph GeNode set V ofeEdge graph G ═ EeThe node in (1) corresponds to the node map
Figure BDA00023357578400000510
An edge of (1);
(b) for the path: defining the connection relationship between the upstream and the downstream as follows: (i → j) is the upstream side of (j → k), and (j → k) is the downstream side of (i → j), and defines AeThe weight of the upstream and downstream connection relationship between (i → j) and (j → k);
considering that a road segment (represented as an edge in a node map) in a traffic network is influenced by the road segments upstream and downstream, the upstream and downstream connection relation is introduced when the edge map is constructed;
taking fig. 2(a) as an example, when the degree of the connecting node j between (i → j) and (j → k) is larger, the influence relationship between (i → j) and (j → k) is weaker because the influence relationship is susceptible to influence of other neighbors, and in view of the above characteristics, a is definedeThe weight of the upstream and downstream connection relationship between (i → j) and (j → k) is as follows:
Figure BDA0002335757840000061
wherein deg is-(. and deg)+(. to) denote the in-degree and out-degree of the node, respectively2Representing the variance of the node degrees.
(c) For the path: a road node i to a road node k denoted as (i → k), a road node j to a road node k denoted as (j → k), i.e., (i → k) and (j → k) share the same end node, are defined as competing relationships, and define AeThe weight of the competitive relationship between (i → k) and (j → k);
considering that road segments connected to the same node in a traffic network compete for downstream traffic resources, a competitive relationship is introduced when constructing the edge graph. Taking fig. 2(b) as an example, when the out-degree of the upstream nodes i and j is large, the vehicle will have multiple routes to select when passing through the nodes i and j, so that the vehicle will have multiple routes to select when passing through the nodes i and jThe competition relationship is weak. In view of the above characteristics, define AeThe weight of the competitive relationship between (i → k) and (j → k) is as follows:
Figure BDA0002335757840000062
(d) constructing a boundary graph GeEdge set E ofeWherein E iseWherein the element is GeThe edge of (2).
Step 4, according to the node map
Figure BDA00023357578400000711
And edge graph GeConstructing a convolution network of a dual graph and calculating the traffic flow X at the time ttInputting k layers of the convolutional network of the dual graph, and expressing the node of each layer output as
Figure BDA0002335757840000071
Figure BDA0002335757840000072
The graph convolutional network is a deep neural network for processing graph structure data, can model message transmission among nodes, and is widely applied to application scenarios such as social network analysis and chemical molecular modeling. Let graph G be (V, E, a), a typical graph convolution network is calculated as follows:
Figure BDA0002335757840000073
wherein the content of the first and second substances,
Figure BDA0002335757840000074
for the input node characteristics, N is the number of nodes, P is the characteristic dimension of each node,
Figure BDA0002335757840000075
for the parameters of the graph convolution network,
Figure BDA0002335757840000076
to take into account self-connected adjacency matrices, INIs an identity matrix of size N x N,
Figure BDA0002335757840000077
is composed of
Figure BDA0002335757840000078
P (-) is a non-linear activation function. The one-layer graph convolution network can aggregate the messages of the 1-hop neighbors for each node, and the range of the message-passing neighbors can be enlarged by stacking the multi-layer graph convolution networks.
As shown in FIG. 3, the invention designs a dual graph convolution network with graph convolution network as a component, which comprises k layers of node graph convolution network and k-1 layers of edge graph convolution network, can simultaneously model the message transmission of nodes and edges, and uses the dual graph convolution network to process the traffic flow at t moment
Figure BDA0002335757840000079
The specific steps (for simplicity of notation, the variables of steps (a) - (d) omit the subscript t without causing ambiguity) are:
(a) constructing node-edge mapping matrices
Figure BDA00023357578400000710
To represent the correspondence between road nodes and edges, where each row of M represents a road node, each column represents an edge, and M is defined as: mi,(i→j)＝M j,(i→j)1, and the other positions are 0;
(b) m pairs of node maps according to node-edge mapping matrix
Figure BDA00023357578400000712
Input X of(0)Linear transformation is performed on X and mapped to edge graph GeInput Z of(0)：
Z(0)＝MTX(0)Wb(5)
Wherein, WbIs a learnable mapping matrix;
(c) side map GeInput Z of(0)Inputting k-1 layers of edge graph convolution networks, and outputting edge representations of each layer:
Figure BDA0002335757840000081
wherein ★ G represents a graph convolution operation,
Figure BDA0002335757840000082
parameters for convolution of the l +1 th layer edge map, Z(l)And Z(l+1)Edge representations respectively representing the output of the l < th > layer and l +1 < th > layer edge graph convolutional networks;
(d) mixing X(0)Inputting a k-layer node graph convolution network, and outputting node representations of all layers, wherein the first-layer node graph convolution network does not consider edge representation, and a back k-1-layer network considers edge representation:
Figure BDA0002335757840000083
Figure BDA0002335757840000084
wherein the content of the first and second substances,
Figure BDA0002335757840000085
is the parameter of the l +1 level node graph convolution [, ]]Indicating a splicing operation, X(l)And X(l+1)Node representations representing the output of the l-th layer and l + 1-th layer node graph convolution network respectively;
(e) taking the output of the node graph convolution network as the output of the dual graph convolution network, and outputting node representation of k layers of the dual graph convolution network at t time
Figure BDA0002335757840000086
The output of each layer represents information for a different neighbor range, where F is the dimension of the output node representation.
Step 5, the nodes output by each layer of the convolution network of the dual graph represent input multi-range attention networks to be fused, and output fusion is carried outCombined expression Ut。
The multi-range attention network is shown in fig. 4, and the specific steps of fusing node representations output by k layers of the convolutional network of the dual graph by using the attention network (for simplifying the labeling, the variable from the step (a) to the step (c) omits a subscript t without causing ambiguity) are as follows:
(a) node representation X for each layer output of the convolutional network of the dual graph(l)A linear transformation is performed, mapping it to the metric space:
Q(l)＝X(l)Wa(9)
wherein the content of the first and second substances,
Figure BDA0002335757840000087
in order for the mapping matrix to be learnable,
Figure BDA0002335757840000088
is a node representation after mapping to the metric space.
(b) Representation after mapping to metric space for node i of l-th layer
Figure BDA0002335757840000091
Figure BDA0002335757840000092
Is Q(l)Is computed with a learnable global neighbor-wide context representation
Figure BDA0002335757840000093
And performing SoftMax normalization between layers to obtain the normalized weight represented by each layer of nodes:
Figure BDA0002335757840000094
Figure BDA0002335757840000095
wherein the content of the first and second substances,
Figure BDA0002335757840000096
and
Figure BDA0002335757840000097
expressed as non-normalized and normalized weights, respectively.
(c) Using normalized weights
Figure BDA0002335757840000098
Carrying out weighted summation on the single node representations of each layer and outputting a fused representation of the single nodes
Figure BDA0002335757840000099
Figure BDA00023357578400000910
Wherein the content of the first and second substances,
Figure BDA00023357578400000911
is X(l)Represents the representation of the level i node.
(d) Splicing the fusion expression of N nodes at the t moment to finally obtain
Figure BDA00023357578400000912
Step 6, the past T is processed′Fused representation of individual moments U(s-T′+1):sInputting a coding and decoding structure Long-time Memory (LSTM) network shared among nodes, and outputting traffic flow predicted values at T moments in the future
Figure BDA00023357578400000913
LSTM networks are a class of recurrent neural networks that can be used to model temporal dependencies in time series data. To effectively extract the features of time series data, a multi-layer LSTM network is generally adopted to enhance the nonlinear capability of the model. In order to balance the fitting capability and complexity of the model, the invention adopts two layers of LSTM networks to model the traffic flow time sequence. As shown in fig. 5, the present invention is based onLSTM constructs a coding and decoding frame shared among nodes, wherein the step length of an encoder LSTM network is T′The step size of the decoder LSTM network is T. Will pass T′Fused representation of individual moments U(s-T′+1):sInputting the encoder and the decoder to output predicted traffic flow values at T time points in the future
Figure BDA00023357578400000914
The traffic flow prediction method uses the dual graph to display and represent the dependency relationship between the nodes and the mutual influence of the relationships between the nodes, and simultaneously introduces the attention mechanism to model the multi-range spatial dependency relationship, so as to model the complex spatial dependency relationship in the traffic flow data, thereby predicting the traffic flow of a traffic network, and having wide application prospects in the fields of trip planning, traffic management and the like.
The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.

Claims (6)

1. A traffic flow prediction method based on a convolution network of a cyclic attention dual graph comprises the following steps:
step 1, constructing a traffic flow time sequence according to traffic flow data collected by road coil sensors on an urban traffic network, and preprocessing the traffic flow time sequence;
step 2, calculating the road network distance between the coil sensors, and constructing a node map according to the road network distance
Figure FDA0002335757830000015
Wherein, VnIs a set of nodes, EnIs a set of edges, AnIs a contiguous matrix;
step 3, node map
Figure FDA0002335757830000016
Edge E innDefining two edge influence modes of upstream and downstream connection relation and competition relation as the nodes of the edge graph, constructing the edges of the edge graph according to the two edge influence modes, and then constructing an edge graph Ge＝(Ve，Ee，Ae) Wherein V iseIs a set of nodes, EeIs a set of edges, AeIs a contiguous matrix;
step 4, according to the node map
Figure FDA0002335757830000017
And edge graph GeConstructing a convolution network of a dual graph and calculating the traffic flow X at the time ttInputting k layers of the convolutional network of the dual graph, and expressing the node of each layer output as
Figure FDA0002335757830000011
Figure FDA0002335757830000012
Step 5, the output nodes of each layer of the convolution network of the dual graph represent input multi-range attention network for fusion, and the output fusion represents Ut；
Step 6, representing the fusion of the past T' time points as U(s-T′+1)：sThe coding and decoding structure long-time memory network shared among the input nodes outputs traffic flow predicted values at T moments in the future
Figure FDA0002335757830000013
2. The traffic flow prediction method based on the cyclic attention pair graph convolutional network according to claim 1, wherein in step 1, the number of coil sensors on the road network is N,
Figure FDA0002335757830000014
indicating that the traffic path ended within the time window at time tAnd (3) processing the missing value and the abnormal value of the traffic flow of N nodes on the network by using a linear interpolation method.
3. The traffic flow prediction method based on the convolutional network of circular attention dual map as claimed in claim 1, wherein in step 2, a node map is constructed
Figure FDA0002335757830000018
The method comprises the following specific steps:
(a) building a node graph
Figure FDA0002335757830000023
Node set V ofn＝{v1，v2，…，vNIn which, | VnN, node set VnOne element in (1) corresponds to one road node;
(b) calculating the road network distance between any two road nodes, dist (v)i，vj) Representing the shortest road network distance from the road node i to the road node j;
(c) calculation based on road network distance between nodes
Figure FDA0002335757830000024
Adjacent matrix A ofnWherein the adjacent matrix AnThe calculation formula of (2) is as follows:
Figure FDA0002335757830000021
wherein σ2D is a manually set threshold value for the variance of the road network distance between all road nodes;
(d) building a node graph
Figure FDA0002335757830000025
Edge set E ofn＝{(i→j)|0≤i，j≤N，Ai，j> 0, where (i → j) represents an edge with i as the head node and j as the tail node.
4. The traffic flow prediction method based on the convolutional network of circular attention dual map as claimed in claim 1, wherein in step 3, a boundary map G is constructedeThe method comprises the following specific steps:
(a) constructing a boundary graph GeNode set V ofeEdge graph G ═ EeThe node in (1) corresponds to the node map
Figure FDA0002335757830000026
An edge of (1);
(b) for the path: defining the connection relationship between the upstream and the downstream as follows: (i → j) is the upstream side of (j → k), and (j → k) is the downstream side of (i → j), and defines AeThe weight of the upstream and downstream connection relationship between (i → j) and (j → k) is as follows:
Figure FDA0002335757830000022
wherein deg is-(. and deg)+(. to) denote the in-degree and out-degree of the node, respectively2A variance representing a degree of a node;
(c) for the path: a road node i to a road node k denoted as (i → k), a road node j to a road node k denoted as (j → k), i.e., (i → k) and (j → k) share the same end node, are defined as competing relationships, and define AeThe weight of the competitive relationship between (i → k) and (j → k) is as follows:
Figure FDA0002335757830000031
(d) constructing a boundary graph GeEdge set E ofeWherein E iseWherein the element is GeThe edge of (2).
5. The traffic flow prediction method based on the convolutional network of the cyclic attention pair graph according to claim 1, wherein step 4 isThe dual graph convolution network comprises a k-layer node graph convolution network and a k-1 layer edge graph convolution network, simultaneously models the message transmission of nodes and edges, and processes the traffic flow at the t moment by using the dual graph convolution network
Figure FDA0002335757830000032
The method comprises the following specific steps:
(a) constructing node-edge mapping matrices
Figure FDA0002335757830000033
To represent the correspondence between road nodes and edges, where each row of M represents a road node, each column represents an edge, and M is defined as: mi，(i→j)＝Mj，(i→j)1, and the other positions are 0;
(b) m pairs of node maps according to node-edge mapping matrix
Figure FDA0002335757830000036
Input X of(0)Linear transformation is performed on X and mapped to edge graph GeInput Z of(0)：
Z(0)＝MTX(0)Wb(4)
Wherein, WbIs a learnable mapping matrix;
(c) side map GeInput Z of(0)Inputting k-1 layers of edge graph convolution networks, and outputting edge representations of each layer:
Figure FDA0002335757830000034
wherein ★ G represents a graph convolution operation,
Figure FDA0002335757830000035
parameters for convolution of the l +1 th layer edge map, Z(l)And Z(l+1)Edge representations respectively representing the output of the l < th > layer and l +1 < th > layer edge graph convolutional networks;
(d) mixing X(0)Inputting k-layer node graph convolution network, outputting eachNode representations of layers, wherein the first layer node graph convolutional network considers edge representations, and the k-1 layer network considers edge representations:
Figure FDA0002335757830000041
Figure FDA0002335757830000042
wherein the content of the first and second substances,
Figure FDA0002335757830000043
is the parameter of the l +1 level node graph convolution [, ]]Indicating a splicing operation, X(l)And X(l+1)Node representations representing the output of the l-th layer and l + 1-th layer node graph convolution network respectively;
(e) taking the output of the node graph convolution network as the output of the dual graph convolution network, and outputting node representation of k layers of the dual graph convolution network at t time
Figure FDA0002335757830000044
The output of each layer represents information for a different neighbor range, where F is the dimension of the output node representation.
6. The traffic flow prediction method based on the circular attention dual graph convolutional network as claimed in claim 1, wherein in step 5, the specific step of fusing the node representations of the k-layer output of the dual graph convolutional network by using the attention network is as follows:
(a) node representation X for each layer output of the convolutional network of the dual graph(l)A linear transformation is performed, mapping it to the metric space:
Q(l)＝X(l)Wa(8)
wherein the content of the first and second substances,
Figure FDA0002335757830000045
in order for the mapping matrix to be learnable,
Figure FDA0002335757830000046
is the node representation after mapping to the metric space;
(b) representation after mapping to metric space for node i of l-th layer
Figure FDA0002335757830000047
Figure FDA0002335757830000048
Is Q(l)Is computed with a learnable global neighbor-wide context representation
Figure FDA0002335757830000049
And performing SoftMax normalization between layers to obtain the normalized weight represented by each layer of nodes:
Figure FDA00023357578300000410
Figure FDA00023357578300000411
wherein the content of the first and second substances,
Figure FDA00023357578300000412
and
Figure FDA00023357578300000413
expressed as non-normalized and normalized weights, respectively;
(c) using normalized weights
Figure FDA00023357578300000414
Carrying out weighted summation on the single node representations of each layer and outputting a fused representation of the single nodes
Figure FDA0002335757830000051
Figure FDA0002335757830000052
Wherein the content of the first and second substances,
Figure FDA0002335757830000053
is X(l)Represents the representation of the l-th level node i;
(d) splicing the fusion expression of N nodes at the t moment to finally obtain
Figure FDA0002335757830000054
CN201911355366.XA 2019-12-25 2019-12-25 Traffic flow prediction method based on cyclic attention dual graph convolution network Pending CN111091712A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911355366.XA CN111091712A (en) 2019-12-25 2019-12-25 Traffic flow prediction method based on cyclic attention dual graph convolution network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911355366.XA CN111091712A (en) 2019-12-25 2019-12-25 Traffic flow prediction method based on cyclic attention dual graph convolution network

Publications (1)

Publication Number Publication Date
CN111091712A true CN111091712A (en) 2020-05-01

Family

ID=70397179

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911355366.XA Pending CN111091712A (en) 2019-12-25 2019-12-25 Traffic flow prediction method based on cyclic attention dual graph convolution network

Country Status (1)

Country Link
CN (1) CN111091712A (en)

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111710154A (en) * 2020-05-15 2020-09-25 湖州师范学院 Expressway traffic flow prediction method
CN111753037A (en) * 2020-06-24 2020-10-09 北京百度网讯科技有限公司 Information representation method and device, electronic equipment and storage medium
CN111815956A (en) * 2020-09-11 2020-10-23 浙江高速信息工程技术有限公司 Expressway traffic flow prediction method
CN111862591A (en) * 2020-05-13 2020-10-30 北京嘀嘀无限科技发展有限公司 Road condition prediction method, road condition prediction device and storage medium
CN111899510A (en) * 2020-07-28 2020-11-06 南京工程学院 Intelligent traffic system flow short-term prediction method and system based on divergent convolution and GAT
CN112071065A (en) * 2020-09-16 2020-12-11 山东理工大学 Traffic flow prediction method based on global diffusion convolution residual error network
CN112382089A (en) * 2020-11-11 2021-02-19 湖南大学 Traffic junction node flow prediction method based on road network directed graph and parallel long-time memory network
CN112382081A (en) * 2020-09-30 2021-02-19 浙江高速信息工程技术有限公司 Traffic flow prediction method based on multiple tasks
CN112529311A (en) * 2020-12-16 2021-03-19 南京中兴力维软件有限公司 Road flow prediction method and device based on graph convolution analysis
CN112669606A (en) * 2020-12-24 2021-04-16 西安电子科技大学 Traffic flow prediction method for training convolutional neural network by utilizing dynamic space-time diagram
CN112748732A (en) * 2020-12-01 2021-05-04 杭州电子科技大学 Real-time path planning method based on improved Kstar algorithm and deep learning
CN112819213A (en) * 2021-01-22 2021-05-18 华南理工大学 Expressway freight volume prediction method and system based on deep learning network
CN113609896A (en) * 2021-06-22 2021-11-05 武汉大学 Object-level remote sensing change detection method and system based on dual-correlation attention
CN113768474A (en) * 2021-11-15 2021-12-10 武汉理工大学 Anesthesia depth monitoring method and system based on graph convolution neural 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
CN114582131A (en) * 2022-03-17 2022-06-03 中远海运科技股份有限公司 Monitoring method and system based on intelligent ramp flow control algorithm
CN115174421A (en) * 2022-06-13 2022-10-11 湖北工业大学 Network fault prediction method and device based on self-supervision unwrapping hypergraph attention
CN115494349A (en) * 2022-11-04 2022-12-20 国网浙江省电力有限公司金华供电公司 Active power distribution network single-phase earth fault section positioning method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180060665A1 (en) * 2016-08-29 2018-03-01 Nec Laboratories America, Inc. Dual Stage Attention Based Recurrent Neural Network for Time Series Prediction
CN109242140A (en) * 2018-07-24 2019-01-18 浙江工业大学 A kind of traffic flow forecasting method based on LSTM_Attention network
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
CN110491129A (en) * 2019-09-24 2019-11-22 重庆城市管理职业学院 The traffic flow forecasting method of divergent convolution Recognition with Recurrent Neural Network based on space-time diagram

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180060665A1 (en) * 2016-08-29 2018-03-01 Nec Laboratories America, Inc. Dual Stage Attention Based Recurrent Neural Network for Time Series Prediction
CN109242140A (en) * 2018-07-24 2019-01-18 浙江工业大学 A kind of traffic flow forecasting method based on LSTM_Attention network
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
CN110491129A (en) * 2019-09-24 2019-11-22 重庆城市管理职业学院 The traffic flow forecasting method of divergent convolution Recognition with Recurrent Neural Network based on space-time diagram

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
WEIQI CHEN 等: "Multi-Range Attentive Bicomponent Graph Convolutional Network for Traffic Forecasting", 《ARXIV》 *
朱凯利: "基于图卷积神经网络的交通流量预测", 《智能计算机与应用》 *
高志强 等: "《深度学习 从入门到实战》", 30 June 2018, 中国铁道出版社 *

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111862591A (en) * 2020-05-13 2020-10-30 北京嘀嘀无限科技发展有限公司 Road condition prediction method, road condition prediction device and storage medium
CN111710154A (en) * 2020-05-15 2020-09-25 湖州师范学院 Expressway traffic flow prediction method
CN111710154B (en) * 2020-05-15 2022-11-18 湖州师范学院 Expressway traffic flow prediction method
CN111753037A (en) * 2020-06-24 2020-10-09 北京百度网讯科技有限公司 Information representation method and device, electronic equipment and storage medium
CN111753037B (en) * 2020-06-24 2023-06-27 北京百度网讯科技有限公司 Information characterization method, information characterization device, electronic equipment and storage medium
CN111899510B (en) * 2020-07-28 2021-08-20 南京工程学院 Intelligent traffic system flow short-term prediction method and system based on divergent convolution and GAT
CN111899510A (en) * 2020-07-28 2020-11-06 南京工程学院 Intelligent traffic system flow short-term prediction method and system based on divergent convolution and GAT
CN111815956A (en) * 2020-09-11 2020-10-23 浙江高速信息工程技术有限公司 Expressway traffic flow prediction method
CN112071065A (en) * 2020-09-16 2020-12-11 山东理工大学 Traffic flow prediction method based on global diffusion convolution residual error network
CN112382081A (en) * 2020-09-30 2021-02-19 浙江高速信息工程技术有限公司 Traffic flow prediction method based on multiple tasks
CN112382089A (en) * 2020-11-11 2021-02-19 湖南大学 Traffic junction node flow prediction method based on road network directed graph and parallel long-time memory network
CN112748732A (en) * 2020-12-01 2021-05-04 杭州电子科技大学 Real-time path planning method based on improved Kstar algorithm and deep learning
CN112748732B (en) * 2020-12-01 2022-08-05 杭州电子科技大学 Real-time path planning method based on improved Kstar algorithm and deep learning
CN112529311A (en) * 2020-12-16 2021-03-19 南京中兴力维软件有限公司 Road flow prediction method and device based on graph convolution analysis
CN112529311B (en) * 2020-12-16 2024-05-24 南京中兴力维软件有限公司 Road flow prediction method and device based on graph convolution analysis
CN112669606A (en) * 2020-12-24 2021-04-16 西安电子科技大学 Traffic flow prediction method for training convolutional neural network by utilizing dynamic space-time diagram
CN112669606B (en) * 2020-12-24 2022-07-12 西安电子科技大学 Traffic flow prediction method for training convolutional neural network by utilizing dynamic space-time diagram
CN112819213A (en) * 2021-01-22 2021-05-18 华南理工大学 Expressway freight volume prediction method and system based on deep learning network
CN112819213B (en) * 2021-01-22 2022-06-14 华南理工大学 Highway freight volume prediction method and system based on deep learning network
CN113609896A (en) * 2021-06-22 2021-11-05 武汉大学 Object-level remote sensing change detection method and system based on dual-correlation attention
CN113609896B (en) * 2021-06-22 2023-09-01 武汉大学 Object-level remote sensing change detection method and system based on dual-related attention
CN113768474A (en) * 2021-11-15 2021-12-10 武汉理工大学 Anesthesia depth monitoring method and system based on graph convolution neural 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
CN114582131B (en) * 2022-03-17 2023-08-29 中远海运科技股份有限公司 Monitoring method and system based on ramp intelligent flow control algorithm
CN114582131A (en) * 2022-03-17 2022-06-03 中远海运科技股份有限公司 Monitoring method and system based on intelligent ramp flow control algorithm
CN115174421B (en) * 2022-06-13 2023-05-23 湖北工业大学 Network fault prediction method and device based on self-supervision unwrapping hypergraph attention
CN115174421A (en) * 2022-06-13 2022-10-11 湖北工业大学 Network fault prediction method and device based on self-supervision unwrapping hypergraph attention
CN115494349A (en) * 2022-11-04 2022-12-20 国网浙江省电力有限公司金华供电公司 Active power distribution network single-phase earth fault section positioning method
CN115494349B (en) * 2022-11-04 2023-04-07 国网浙江省电力有限公司金华供电公司 Active power distribution network single-phase earth fault section positioning method

Similar Documents

Publication Publication Date Title
CN111091712A (en) Traffic flow prediction method based on cyclic attention dual graph convolution network
CN109754605B (en) Traffic prediction method based on attention temporal graph convolution network
Zhou et al. Reinforced spatiotemporal attentive graph neural networks for traffic forecasting
US6470261B1 (en) Automatic freeway incident detection system and method using artificial neural network and genetic algorithms
CN111383452A (en) Method for estimating and predicting short-term traffic running state of urban road network
US20220092418A1 (en) Training method for air quality prediction model, prediction method and apparatus, device, program, and medium
Wang et al. A multi-view bidirectional spatiotemporal graph network for urban traffic flow imputation
CN110047291B (en) Short-term traffic flow prediction method considering diffusion process
CN112489426B (en) Urban traffic flow space-time prediction scheme based on graph convolution neural network
CN114519932B (en) Regional traffic condition integrated prediction method based on space-time relation extraction
US20230334981A1 (en) Traffic flow forecasting method based on multi-mode dynamic residual graph convolution network
CN113673769A (en) Graph neural network traffic flow prediction method based on multivariate time sequence interpolation
Kim et al. Structural recurrent neural network for traffic speed prediction
Das et al. A probabilistic approach for weather forecast using spatio-temporal inter-relationships among climate variables
CN111382686A (en) Lane line detection method based on semi-supervised generation confrontation network
Lu et al. Leveraging graph neural network with lstm for traffic speed prediction
CN113112791A (en) Traffic flow prediction method based on sliding window long-and-short term memory network
CN114841400A (en) Air quality prediction method based on multi-task space-time diagram convolution
CN113487856A (en) Traffic flow combination prediction model based on graph convolution network and attention mechanism
CN114742280B (en) Road condition prediction method and corresponding model training method, device, equipment and medium
CN116168548A (en) Traffic flow prediction method of space-time attention pattern convolution network based on multi-feature fusion
CN116486624A (en) Traffic flow prediction method and system based on space-time diagram convolutional neural network
Jia et al. Forecasting citywide short-term turning traffic flow at intersections using an attention-based spatiotemporal deep learning model
Das et al. Enhanced Bayesian network models for spatial time series prediction
Yang et al. Spatiotemporal DeepWalk gated recurrent neural network: a deep learning framework for traffic learning and forecasting

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20200501

RJ01 Rejection of invention patent application after publication