CN112562312B - GraphSAGE traffic network data prediction method based on fusion features - Google Patents

GraphSAGE traffic network data prediction method based on fusion features Download PDF

Info

Publication number
CN112562312B
CN112562312B CN202011129295.4A CN202011129295A CN112562312B CN 112562312 B CN112562312 B CN 112562312B CN 202011129295 A CN202011129295 A CN 202011129295A CN 112562312 B CN112562312 B CN 112562312B
Authority
CN
China
Prior art keywords
network
road network
traffic
road
node
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011129295.4A
Other languages
Chinese (zh)
Other versions
CN112562312A (en
Inventor
徐东伟
商学天
魏臣臣
林臻谦
丁加丽
彭航
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 of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
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 of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202011129295.4A priority Critical patent/CN112562312B/en
Publication of CN112562312A publication Critical patent/CN112562312A/en
Application granted granted Critical
Publication of CN112562312B publication Critical patent/CN112562312B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G08G1/0133—Traffic data processing for classifying traffic situation

Landscapes

  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Traffic Control Systems (AREA)

Abstract

A method for predicting data of a GraphSAGE traffic network based on fusion characteristics comprises the steps of firstly, carrying out correlation coefficient calculation on historical traffic flow data of the network, constructing a correlation matrix of the network, redefining the communication state between nodes according to the correlation between the nodes of the network, and obtaining a topological network based on time correlation; and then, respectively extracting road network characteristic information of the original traffic road network and the reconstructed topological road network by using GraphSAGE, and predicting the future traffic state of the road network by fusing the road network space-time characteristic information extracted by two different road networks. The invention integrates the road network space-time characteristic information extracted from two different road networks, predicts the future traffic state of the road network and improves the accuracy of predicting the traffic network state data.

Description

GraphSAGE traffic network data prediction method based on fusion features
Technical Field
The invention relates to a method for predicting GraphSAGE traffic network data based on fusion characteristics, and belongs to the field of intelligent transportation.
Background
With the rapid development of modern cities, the number of people and vehicles is rapidly increased, so that the problem of urban traffic road congestion is more and more severe, and people and society are not disturbed, therefore, the traffic state is better adjusted in order to ensure that the road traffic has stronger liquidity, and the prediction of future traffic state data is of great significance.
The road traffic prediction method at the present stage mainly comprises the following steps: the method comprises the following steps of graph convolution neural network, noise reduction self-encoder, support vector machine, feedback neural network and the like, but most of the methods are direct-push learning and cannot be directly generalized to unknown roads.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a method for predicting the GraphSAGE traffic network data based on fusion characteristics.
The technical scheme adopted by the invention for solving the technical problem is as follows:
a method for predicting GraphSAGE traffic network data based on fusion characteristics comprises the following steps:
1) Constructing a topological network based on time correlation: calculating correlation coefficients among different road network nodes according to historical traffic state data of the road network, redefining the communication relation among the nodes according to the correlation coefficients among the different nodes, and constructing a logic correlation road network based on time correlation;
2) Extracting road network space-time characteristics based on GraphSAGE and performing characteristic fusion: adopting GraphSAGE to respectively extract space-time characteristics of the original road network and the reconstructed logic-related road network based on time correlation, and performing characteristic fusion on the extracted different characteristics;
3) Defining a network model loss function, continuously training and adjusting model parameters by taking a minimum loss function as a target, and finally realizing the prediction of the traffic state of the road network: and defining a model loss function, continuously carrying out iterative training by adopting a back propagation algorithm to reduce the loss function, and finally storing optimal model parameters to realize the prediction of future traffic state data of the road network based on the historical traffic state data of the road network.
The technical conception of the invention is as follows: firstly, performing correlation coefficient calculation on historical traffic flow data of a road network, constructing a road network correlation matrix, redefining the communication state between nodes according to the correlation between the road network nodes, and obtaining a topological road network based on time correlation; and then, respectively extracting road network characteristic information of the original traffic road network and the reconstructed topological road network by using GraphSAGE, and predicting the future traffic state of the road network by fusing the road network space-time characteristic information extracted by two different road networks. The method has a crucial role in the field of intelligent traffic, realizes the extraction of the space-time characteristics of the traffic flow state, and improves the accuracy of the traffic network state prediction.
The invention has the following beneficial effects: (1) And fully mining the space-time characteristics of the road network by adopting a GraphSAGE graph aggregation algorithm. (2) The original road network and the reconstructed topological road network are subjected to space-time feature extraction and feature fusion respectively, so that the prediction precision of the traffic road network is effectively improved.
Drawings
FIG. 1 is a diagram of a GraphSAGE network model architecture.
FIG. 2 is the prediction result of GraphSAGE traffic network model based on fusion features (3, 10, 2017).
Detailed Description
The invention is further described below with reference to the accompanying drawings.
Referring to fig. 1 and 2, a method for predicting GraphSAGE traffic network data based on fusion characteristics comprises the following steps:
1) Constructing a topological network based on time correlation, and the process is as follows:
1.1 ) construct an original road network of traffic
Constructing a traffic original road network G = (V, E), wherein: v = { V) 1 ,v 2 ,v 3 ,…,v N }，|V|＝N，
Figure BDA0002734622440000031
N is the number of detectors in a node of the traffic network, E is the adjacency matrix of the traffic network state, i.e. the spatial relationship between the nodes of the traffic network, v i ( i epsilon 1,2,3, \8230;, N) represents the ith detector which detects the traffic node, and the selected and node v i The node sets in the space are the connection relation and are marked as
Figure BDA0002734622440000032
If the i-th detector v i Representative road node and jth detector v j If the represented road nodes have adjacent relation, e ij =1, otherwise: e.g. of a cylinder ij ＝0；
1.2 Computing correlation coefficients between different road network nodes
Road node v for each detector using Pearson's correlation coefficient calculation formula i (i ∈ 1,2,3, \8230;, N), the historical road state data of which are as follows: x is the number of i ＝[x i1 ,x i2 ,x i3 ,…,x iT ]T is the data quantity in the historical data, the correlation between the nodes of each road network is calculated, and the ith detector v is used for calculating the correlation i Representative road node and jth detector v j Pearson's correlation coefficient r between representative road nodes ij The calculation formula is as follows:
Figure BDA0002734622440000033
where K is the length of the traffic network state node data represented by the detector chosen in the calculation of the pearson correlation coefficient. Obtaining a Pearson correlation coefficient matrix of x of the road network G by obtaining Pearson correlation coefficients among different detectors
Figure BDA0002734622440000041
1.3 Constructing a logical correlation road network based on time correlation according to the Pearson correlation coefficient matrix
For each detector node v i Belongs to V, the Pearson correlation coefficient between detectors is obtained by calculation, and the Pearson correlation coefficient is selected to be connected with a node V i M detectors with larger coefficient (noted as
Figure BDA0002734622440000042
) For the continuous-edge relation, a coefficient matrix of the temporal correlation is constructed, wherein,
Figure BDA0002734622440000043
v ik representation and node v i Establishing the kth (k =1,2, \ 8230;, m) node, v, of the continuous edge relation im ∈V，
Figure BDA0002734622440000044
Represents rounding down; p is the proportion of the more relevant detector nodes chosen, p ∈ (0, 1). The constructed traffic road network H = (V, A), wherein
Figure BDA0002734622440000045
a ij Denotes the ith detector v i And the jth detector v j The connection edge relationship existing between the two is as follows:
Figure BDA0002734622440000046
2) Extracting the space-time characteristics of the road network based on GraphSAGE and fusing the characteristics, wherein the process is as follows:
according to an original road network and a constructed logic correlation road network based on time correlation, space information is aggregated by adopting a mean aggregation method aiming at the neighbor node characteristics of each detector node, and if T-layer mean aggregation is carried out, an aggregation calculation formula is as follows:
Figure BDA0002734622440000047
Figure BDA0002734622440000048
Figure BDA0002734622440000049
Figure BDA00027346224400000410
wherein
Figure BDA0002734622440000051
And
Figure BDA0002734622440000052
respectively represents the node v for the original road network and the logically related road network i Extracting the t-th layer characteristics through the characteristics of the GraphSAGE traffic network model;
Figure BDA0002734622440000053
respectively represent
Figure BDA0002734622440000054
Features obtained by polymerization at layer t by means of a mean polymerization process, in which
Figure BDA0002734622440000055
And
Figure BDA0002734622440000056
respectively expressed as original road network and logically related road network, and node v i A node set with a connection edge relation, namely a neighbor node set; mean is expressed as solving the characteristic mean value of different nodes on different characteristic attributes; CONCAT is expressed as feature merge; sigma 1 ，σ 2 Expressed as an activation function; w is a group of 1 t ，W 2 t Weight parameters to be trained for the model;
after T-layer GraphSAGE mean value aggregation is carried out on all nodes in the road network, aggregation characteristics representing adjacent matrixes and based on correlation coefficient matrixes are obtained respectively
Figure BDA0002734622440000057
And
Figure BDA0002734622440000058
Figure BDA0002734622440000059
Figure BDA00027346224400000510
and performing feature fusion on the two aggregated features, wherein the calculation mode of the feature fusion is as follows:
Figure BDA00027346224400000511
wherein
Figure BDA00027346224400000512
Representing the feature after feature fusion, W T ，σ 1 ，σ 2 B is the parameter to be learned of the model, σ 1 As ReLU function, σ 2 As a Sigmoid function, the function expression is:
Figure BDA00027346224400000513
Figure BDA00027346224400000514
3) Defining a network model loss function, and predicting traffic network state data, wherein the process is as follows:
defining a network model loss function L G ：
Figure BDA0002734622440000061
Defining a model loss function L G The model is that the output data is subjected to anti-standardization operation to obtain the predicted traffic network state data, and the anti-standardization calculation formula is as follows:
Figure BDA0002734622440000062
wherein the content of the first and second substances,
Figure BDA0002734622440000063
respectively represent the minimum value of the speed of the ith road section,
Figure BDA0002734622440000064
respectively represent the maximum speed of the ith link, F i(t+q) The speed of the ith road at the (t + q) th time is respectively predicted.
Example (c): the data in the actual experiment were carried out as follows:
1) Selecting experimental data
The experimental data set adopts the speed data of 323 detectors in the Seattle expressway network in 2017 all the year, and the data sampling interval is 5 minutes.
2) Parameter determination
The number of nodes of the traffic network detector is N =323, and the number of features of each node is F =12; the division ratio a =0.8 of the training set and the test set, when a road network structure based on time correlation is constructed and a Pearson correlation coefficient is calculated, the length K =288 × 12=3456 of the historical traffic state data of each selected detector node, and the selection ratio of the detector node with the larger Pearson correlation coefficient is set as p =0.25; the number of layers of GraphSAGE mean aggregation is T =3, the number of hidden units in each layer is 128,64,32, and the activation function sigma is a ReLU activation function; reconstruction error coefficient α =100; the Adam optimizer was used to optimize the model parameters.
3) Results of the experiment
The model evaluation index selects the Root Mean Square Error (RMSE), mean Absolute Error (MAE) and Mean Absolute Percent Error (MAPE). The functional expressions are respectively:
Figure BDA0002734622440000071
Figure BDA0002734622440000072
Figure BDA0002734622440000073
wherein
Figure BDA0002734622440000074
For the real traffic status data at the kth time,
Figure BDA0002734622440000075
predicted traffic state data for the kth time.
In the result analysis, the time error of the data of one whole year is evaluated, the experimental result is shown in table 1, and table 1 is the prediction experimental result of the data of the traffic road of 2017 year all the year in seattle:
Time RMSE MAE MAPE(％)
2017 3.72 2.51 5.59
Table 1.
The embodiments described in this specification are merely exemplary of implementations of the inventive concepts and are provided for illustrative purposes only. The scope of the present invention should not be construed as being limited to the particular forms set forth in the embodiments, but is to be accorded the widest scope consistent with the principles and equivalents thereof as contemplated by those skilled in the art.

Claims (1)

1. A method for predicting GraphSAGE traffic network data based on fusion characteristics is characterized by comprising the following steps:
1) Constructing a topological network based on time correlation: calculating correlation coefficients among different road network nodes according to historical traffic state data of the road network, redefining the communication relation among the nodes according to the correlation coefficients among the different nodes, and constructing a logic correlation road network based on time correlation;
2) Extracting road network space-time characteristics based on GraphSAGE and performing characteristic fusion: adopting GraphSAGE to respectively extract space-time characteristics of the original road network and the reconstructed logically-related road network based on time correlation, and performing characteristic fusion on the extracted different characteristics;
3) Defining a network model loss function, continuously training and adjusting model parameters by taking the minimum loss function as a target, and finally realizing the prediction of the traffic state of the road network: defining a model loss function, continuously performing iterative training by adopting a back propagation algorithm to reduce the loss function, and finally storing optimal model parameters to realize prediction of future traffic state data of the road network based on historical traffic state data of the road network;
the process of the step 1) is as follows:
1.1 Constructing an original road network of traffic
Constructing a traffic original road network G = (V, E), wherein: v = { V) 1 ,v 2 ,v 3 ,...,v N }，|V|＝N，
Figure FDA0003684909060000011
N is the number of detectors in a node of the traffic network, E is the adjacency matrix of the traffic network state, i.e. the spatial relationship between the nodes of the traffic network, v i (i epsilon 1,2,3, \8230;, N) represents the ith detector which detects the traffic node, and the selected and node v i The node sets in the connection relation in space are recorded as
Figure FDA0003684909060000012
If the ith detector v i Representative road node and jth detector v j If the represented road nodes have adjacent relation, e ij =1, otherwise: e.g. of the type ij ＝0；
1.2 Computing correlation coefficients between different road network nodes
Road node v for each detector using Pearson's correlation coefficient calculation formula i (i ∈ 1,2,3, \8230;, N), the historical road state data of which are as follows: x is a radical of a fluorine atom i ＝[x i1 ,x i2 ,x i3 ,…,x iT ]T is the data quantity in the historical data, the correlation between each road network node is calculated, and the ith detector v i Representative road node and jth detector v j Pearson's correlation coefficient r between representative road nodes ij The calculation formula is as follows:
Figure FDA0003684909060000021
k is the length of the traffic network state node data represented by the detector selected when the Pearson correlation coefficient is calculated, and the Pearson correlation coefficient matrix of x of the road network G is obtained by obtaining the Pearson correlation coefficient among different detectors
Figure FDA0003684909060000022
1.3 Constructing a logical correlation road network based on time correlation according to the Pearson correlation coefficient matrix
For each detector node v i E, V, calculating to obtain Pearson correlation coefficient between detectors, selecting and connecting with node V i M detectors with larger coefficients between them are taken as a continuous boundary relation, and the m detectors are recorded as
Figure FDA0003684909060000023
A matrix of time-dependent coefficients is constructed, wherein,
Figure FDA0003684909060000024
v ik representation and node v i Establishing the kth (k =1,2, \8230;, m) node of the continuous edge relation，v im ∈V，
Figure FDA0003684909060000025
Represents rounding down; p is the proportion of the selected relatively relevant detector nodes, p is the (0, 1), and the constructed traffic road network H = (V, A), wherein
Figure FDA0003684909060000026
a ij Denotes the ith detector v i And the jth detector v j The connection edge relationship existing between the two is as follows:
Figure FDA0003684909060000027
the process of the step 2) is as follows:
according to an original road network and a constructed logic correlation road network based on time correlation, space information is aggregated by adopting a mean aggregation method aiming at the neighbor node characteristics of each detector node, and if T-layer mean aggregation is carried out, an aggregation calculation formula is as follows:
Figure FDA0003684909060000028
Figure FDA0003684909060000029
Figure FDA00036849090600000210
Figure FDA0003684909060000031
wherein
Figure FDA0003684909060000032
And
Figure FDA0003684909060000033
respectively representing the nodes v for the original road network and the logically related road network i Extracting the characteristics of the t-th layer through the characteristics of the GraphSAGE traffic network model;
Figure FDA0003684909060000034
respectively represent
Figure FDA0003684909060000035
Features obtained by polymerization at layer t by means of a mean polymerization process, in which
Figure FDA0003684909060000036
And
Figure FDA0003684909060000037
respectively expressed as an original road network and a logic-related road network, and a node v i A node set with a connection edge relation, namely a neighbor node set; mean is expressed as solving the characteristic mean value of different nodes on different characteristic attributes; CONCAT is expressed as feature merge; sigma 1 ，σ 2 Expressed as an activation function; w 1 t ，W 2 t Weight parameters which need to be trained for the model;
after T-layer GraphSAGE mean value aggregation is carried out on all nodes in the road network, aggregation characteristics representing adjacent matrixes and based on correlation coefficient matrixes are obtained respectively
Figure FDA0003684909060000038
And
Figure FDA0003684909060000039
Figure FDA00036849090600000310
Figure FDA00036849090600000311
and performing feature fusion on the two aggregated features, wherein the calculation mode of the feature fusion is as follows:
Figure FDA00036849090600000312
wherein
Figure FDA00036849090600000313
Representing the feature after feature fusion, W T ，σ 1 ，σ 2 B is the parameter to be learned of the model, σ 1 As ReLU function, σ 2 As a Sigmoid function, the function expression is:
Figure FDA00036849090600000314
Figure FDA00036849090600000315
the process of the step 3) is as follows:
defining a network model loss function L G ：
Figure FDA00036849090600000316
Defining a model loss function L G Wherein alpha is a reconstruction error coefficient, X is real data of a future traffic state, a loss function is minimized, optimal model parameters are returned finally, a loop iteration back propagation algorithm is adopted, and finally a GraphSAGE traffic network model based on fusion characteristics is reservedThe state data of the access network has an anti-standardization calculation formula as follows:
Figure FDA00036849090600000317
wherein the content of the first and second substances,
Figure FDA0003684909060000041
respectively represent the minimum value of the speed of the ith road section,
Figure FDA0003684909060000042
respectively represent the maximum speed of the ith link, F i(t+q) The predicted sizes are respectively the speed of the ith road at the (t + q) th moment.
CN202011129295.4A 2020-10-21 2020-10-21 GraphSAGE traffic network data prediction method based on fusion features Active CN112562312B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011129295.4A CN112562312B (en) 2020-10-21 2020-10-21 GraphSAGE traffic network data prediction method based on fusion features

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011129295.4A CN112562312B (en) 2020-10-21 2020-10-21 GraphSAGE traffic network data prediction method based on fusion features

Publications (2)

Publication Number Publication Date
CN112562312A CN112562312A (en) 2021-03-26
CN112562312B true CN112562312B (en) 2022-10-28

Family

ID=75042807

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011129295.4A Active CN112562312B (en) 2020-10-21 2020-10-21 GraphSAGE traffic network data prediction method based on fusion features

Country Status (1)

Country Link
CN (1) CN112562312B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113569750B (en) * 2021-07-29 2023-07-07 上海动亦科技有限公司 Road target detection and identification method based on spatial feature aggregation
CN113705873B (en) * 2021-08-18 2024-01-19 中国科学院自动化研究所 Construction method of film and television work score prediction model and score prediction method

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110991713B (en) * 2019-11-21 2022-04-01 杭州电子科技大学 Irregular area flow prediction method based on multi-graph convolution sum GRU
CN111291193B (en) * 2020-01-18 2020-09-29 杭州电子科技大学 Application method of knowledge graph in zero-time learning
CN111612243B (en) * 2020-05-18 2023-11-10 湖南大学 Traffic speed prediction method, system and storage medium
CN111768625A (en) * 2020-07-01 2020-10-13 中国计量大学 Traffic road event prediction method based on graph embedding
CN111756587B (en) * 2020-07-29 2021-06-15 南昌航空大学 Method for predicting time sequence network link by using GraphSAGE

Also Published As

Publication number Publication date
CN112562312A (en) 2021-03-26

Similar Documents

Publication Publication Date Title
CN112418547B (en) Bus stop passenger flow prediction method based on GCN-LSTM combination model
CN111860951B (en) Rail transit passenger flow prediction method based on dynamic hypergraph convolutional network
CN113313947B (en) Road condition evaluation method of short-term traffic prediction graph convolution network
CN109492822B (en) Air pollutant concentration time-space domain correlation prediction method
CN113053115B (en) Traffic prediction method based on multi-scale graph convolution network model
CN111612243B (en) Traffic speed prediction method, system and storage medium
CN112350876A (en) Network flow prediction method based on graph neural network
CN112396587B (en) Method for detecting congestion degree in bus compartment based on collaborative training and density map
CN112906982A (en) GNN-LSTM combination-based network flow prediction method
CN113591380B (en) Traffic flow prediction method, medium and equipment based on graph Gaussian process
CN113129585B (en) Road traffic flow prediction method based on graph aggregation mechanism of reconstructed traffic network
CN112562312B (en) GraphSAGE traffic network data prediction method based on fusion features
CN112949828A (en) Graph convolution neural network traffic prediction method and system based on graph learning
CN111047078B (en) Traffic characteristic prediction method, system and storage medium
CN112241724A (en) Automatic identification method and system based on double-path convolution long-term and short-term neural network
CN112991721A (en) Urban road network traffic speed prediction method based on graph convolution network node association degree
CN112966871A (en) Traffic jam prediction method and system based on convolution long-short term memory neural network
CN115376317B (en) Traffic flow prediction method based on dynamic graph convolution and time sequence convolution network
CN115828990A (en) Time-space diagram node attribute prediction method for fused adaptive graph diffusion convolution network
CN112598165A (en) Private car data-based urban functional area transfer flow prediction method and device
CN114118567A (en) Power service bandwidth prediction method based on dual-channel fusion network
CN116011684A (en) Traffic flow prediction method based on space-time diagram convolutional network
CN115080795A (en) Multi-charging-station cooperative load prediction method and device
CN115966083A (en) Distributed traffic flow prediction method and system based on time-space decoupling
CN114596726B (en) Parking berth prediction method based on interpretable space-time attention mechanism

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant