CN114254214A - Traffic prediction method and system based on space-time hierarchical network - Google Patents

Traffic prediction method and system based on space-time hierarchical network Download PDF

Info

Publication number
CN114254214A
CN114254214A CN202111656768.0A CN202111656768A CN114254214A CN 114254214 A CN114254214 A CN 114254214A CN 202111656768 A CN202111656768 A CN 202111656768A CN 114254214 A CN114254214 A CN 114254214A
Authority
CN
China
Prior art keywords
traffic
network
hierarchical
region
matrix
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
CN202111656768.0A
Other languages
Chinese (zh)
Inventor
刘星星
黄玲
王昌栋
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.)
Sun Yat Sen University
Original Assignee
Sun Yat Sen University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sun Yat Sen University filed Critical Sun Yat Sen University
Priority to CN202111656768.0A priority Critical patent/CN114254214A/en
Publication of CN114254214A publication Critical patent/CN114254214A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2465—Query processing support for facilitating data mining operations in structured databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/29—Graphical models, e.g. Bayesian 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
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • 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
    • G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Analytical Chemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Remote Sensing (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Traffic Control Systems (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a traffic prediction method and a system based on a space-time hierarchical network, wherein the method comprises the following steps: acquiring traffic data, preprocessing the traffic data, and constructing to obtain a hierarchical regional enhanced network and a traffic characteristic matrix; taking the hierarchical regional enhancement network and the traffic characteristic matrix as the input of a prediction model, learning the spatial correlation and the temporal correlation, and outputting a prediction result; the predictive models include a region-aware spatial correlation model and a region-aware temporal correlation model. The system comprises: a preprocessing module and a prediction module. By using the method and the device, the time-space correlation in traffic data is effectively captured, and the accuracy of traffic flow prediction is improved. The invention is used as a traffic prediction method and system based on the space-time hierarchical network, and can be widely applied to the field of traffic prediction.

Description

Traffic prediction method and system based on space-time hierarchical network
Technical Field
The invention relates to the field of traffic prediction, in particular to a traffic prediction method and system based on a space-time hierarchical network.
Background
The traffic prediction task is a key task for realizing traffic management, traffic planning and traffic control in an intelligent traffic system. The traffic prediction task aims to realize the prediction of traffic information such as traffic flow, traffic speed, traffic density and the like at the future time by analyzing the traffic information at the current and historical times and traffic-related external conditions.
The current prediction method mainly models traffic prediction into a time series mining problem, but only considers time correlation, and spatial correlation in traffic data is largely ignored (for example, spatial correlation between different roads or traffic monitoring points), so that the improvement of traffic prediction accuracy is greatly limited.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a traffic prediction method and system based on a space-time hierarchical network, which aims at a space-time data mining algorithm of a hierarchical region structure and improves the accuracy of a traffic prediction result by means of deep learning.
The first technical scheme adopted by the invention is as follows: a traffic prediction method based on a space-time hierarchical network comprises the following steps:
acquiring traffic data, preprocessing the traffic data, and constructing to obtain a hierarchical regional enhanced network and a traffic characteristic matrix;
taking the hierarchical regional enhancement network and the traffic characteristic matrix as the input of a prediction model, learning the spatial correlation and the temporal correlation, and outputting a prediction result;
the predictive models include a region-aware spatial correlation model and a region-aware temporal correlation model.
Further, the step of acquiring traffic data and preprocessing the traffic data to construct a hierarchical regional enhanced network and a traffic feature matrix includes:
acquiring original traffic data;
from the original road network
Figure BDA0003446154150000011
And hierarchical region structure
Figure BDA0003446154150000012
Adding the region as a virtual node into the network, extracting the connection relation between the node and the region and between the regions, and establishing a hierarchical regional enhanced network
Figure BDA0003446154150000013
Modeling traffic time sequence information in original traffic data as node attributes, and constructing a traffic characteristic matrix
Figure BDA0003446154150000014
Further, the step of using the hierarchical region enhancement network and the traffic feature matrix as input of the prediction model, learning the spatial correlation and the temporal correlation, and outputting the prediction result specifically includes:
graph attention network enhancing network according to hierarchical regions
Figure BDA0003446154150000021
Spatial topology of, traffic characteristic matrix to be input
Figure BDA0003446154150000022
Conversion into a new feature matrix Xt；
Gated cyclic unit based on new feature matrix XtCalculating a corresponding hidden state matrix Ht；
Alternately passing through a graph attention network and a gate control circulation unit and recursively learning hidden states for multiple times, and then based on a hidden state matrix H of the last steptPerforming dimension transformation via a full link layer to obtain final prediction result
Figure BDA0003446154150000023
Further, the graph attention network enhances the network according to the hierarchical region
Figure BDA0003446154150000024
Spatial topology of, traffic characteristic matrix to be input
Figure BDA0003446154150000025
Conversion into a new feature matrix XtThis step, in particular, comprises:
for traffic characteristic matrix to be input
Figure BDA0003446154150000026
At each time step, modeling the spatial dependence relationship of the inter-node, the inter-node and inter-region perception based on the attention mechanism to obtain a corresponding new feature matrix Xt。
Further, the gated cyclic unit is based on the new feature matrix XtCalculating a corresponding hidden state matrix HtThis step, in particular, comprises:
the gate control circulation unit simultaneously models the time dependence relationship of region perception among nodes, between nodes and regions and between regions;
based on new feature matrix XtAnd the hidden state matrix H of the last time stept-1Obtaining the hidden state matrix H of the current time stept。
Further, the gated loop unit includes a reset gate that determines a degree of forgetting information for a past time step, and an update gate that determines a degree of transferring the information for the past time step to a next hidden state.
The second technical scheme adopted by the invention is as follows: a traffic prediction system based on a spatiotemporal hierarchical network, comprising:
the preprocessing module is used for acquiring traffic data, preprocessing the traffic data and constructing a hierarchical regional enhanced network and a traffic characteristic matrix;
and the prediction module is used for taking the hierarchical region enhancement network and the traffic characteristic matrix as the input of the prediction model, learning the spatial correlation and the temporal correlation and outputting the prediction result.
The method and the system have the beneficial effects that: the invention further constructs a spatial correlation model and a time correlation model of regional perception based on the network by expanding the original road network into a hierarchical regional enhanced network so as to effectively capture the space-time correlation in traffic data and improve the accuracy of traffic flow prediction.
Drawings
FIG. 1 is a flow chart of the steps of a traffic prediction method based on a spatio-temporal hierarchical network according to the present invention;
FIG. 2 is a schematic diagram of a data processing procedure of a prediction method according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating the establishment of a hierarchical regional enhanced network in accordance with an embodiment of the present invention;
FIG. 4 is a data processing diagram of the attention network in accordance with an embodiment of the present invention;
FIG. 5 is a schematic diagram of a gate unit according to an embodiment of the present invention;
FIG. 6 is a block diagram of a traffic prediction system based on a spatiotemporal hierarchical network according to the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and the specific embodiments. The step numbers in the following embodiments are provided only for convenience of illustration, the order between the steps is not limited at all, and the execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.
The traffic space-time data relates to coding of space information in a road network and traffic data time sequence information at different moments, and are respectively represented by the traffic network and an attribute characteristic matrix: traffic roads are modeled as a undirected, unweighted graph
Figure BDA0003446154150000031
Wherein
Figure BDA0003446154150000032
Is a node set consisting of n nodes, viAnd epsilon is an edge set corresponding to the ith traffic monitoring point. Besides modeling the topological structure among the traffic monitoring points as a road network, the traffic timing information also needs to be modeled as node attributes, namely a feature matrix of the road network, as follows:
the traffic information of the time step t is modeled as a node attribute characteristic of the road network, expressed as
Figure BDA0003446154150000033
Wherein the column vector ftThe ith item of (a) represents the traffic information of the ith monitoring point in the time step t. Traffic characteristic matrix
Figure BDA0003446154150000034
And representing the traffic information in p historical steps and the current time step t, wherein each column represents the traffic information of all traffic monitoring points in a corresponding single step.
The traffic prediction problem is defined as, at time t, given traffic topology information
Figure BDA0003446154150000035
Traffic flow information of past p time step lengths and current step length
Figure BDA0003446154150000036
Predicting traffic information in future time steps of length q
Figure BDA0003446154150000037
Namely:
Figure BDA0003446154150000038
referring to fig. 1 and 2, the present invention provides a traffic prediction method based on a spatio-temporal hierarchical network, the method comprising the steps of:
s1, acquiring traffic data, preprocessing the traffic data, and constructing to obtain a hierarchical regional enhanced network and a traffic characteristic matrix;
specifically, the hierarchical region structure existing in the traffic network is considered, after the traffic space-time data are obtained, a hierarchical region enhancement network needs to be constructed through data preprocessing, and the network data obtained through preprocessing are further used as the input of the traffic flow prediction model.
S2, taking the hierarchical regional enhancement network and the traffic characteristic matrix as the input of a prediction model, learning the spatial correlation and the temporal correlation, and outputting a prediction result;
the predictive models include a region-aware spatial correlation model and a region-aware temporal correlation model.
Further, as a preferred embodiment of the method, the step of acquiring traffic data, preprocessing the traffic data, and constructing and obtaining a hierarchical regional enhanced network and a traffic feature matrix specifically includes:
acquiring original traffic data;
modeling traffic time sequence information in original traffic data as node attributes, and constructing a traffic characteristic matrix
Figure BDA0003446154150000041
From the original road network
Figure BDA0003446154150000042
And hierarchical region structure
Figure BDA0003446154150000043
Extracting the connection relation between nodes and regions and between regions, and establishing a hierarchical regional enhanced network
Figure BDA0003446154150000044
Specifically, the setup process is as shown in fig. 3. The left side of fig. 2 shows an original traffic road network with 15 nodes. Original sourceThe 15 nodes in the original network are divided into 5 regions
Figure BDA0003446154150000045
In which 5 regions are further divided into 2 higher level regions
Figure BDA0003446154150000046
And
Figure BDA0003446154150000047
in (1). The right side of fig. 3 shows the process of building a hierarchical regional enhanced network: adding 7 virtual nodes v16,v17,v18,v19,v20,v21And v22Respectively represent regions
Figure BDA0003446154150000048
And
Figure BDA0003446154150000049
and adding new connecting edges into the network according to the hierarchical node-area, area-area membership and the adjacent relation between areas. In the obtained regional enhanced network, newly-established node connecting edges are represented by dotted lines, wherein three types of node-regional membership relations (such as connecting edges 10-19, 11-19, 12-19 and 13-19), regional-regional membership relations (such as connecting edges 16-21, 17-21, 18-22, 19-22 and 20-22) and regional-regional proximity relations (such as connecting edges 16-17, 16-18, 17-19, 18-19, 19-20 and 21-22) are respectively contained.
Specifically, after the original road network is acquired, it needs to be expanded into a hierarchical regional enhanced network. Simultaneously consider: 1) the spatial/temporal correlation between nodes, 2) between nodes and regions, and 3) between regions (including the same and different levels) is extracted. And adding the area into the network as a virtual node, and updating the connecting edges between the nodes.
Use of
Figure BDA0003446154150000051
Representing a hierarchical regional structure of L layers, wherein
Figure BDA0003446154150000052
Indicates the first layer region set (containing k together)lAn area). It is assumed that each node in the original road network is affiliated with and belongs to only one of the primary regions. Likewise, for
Figure BDA0003446154150000053
The area of each layer/belongs to and only belongs to one of the layer/+ 1 areas. Finally, use
Figure BDA0003446154150000054
To indicate that all L layers contain
Figure BDA0003446154150000055
And (4) a region.
Based on original network structure
Figure BDA0003446154150000056
And region set
Figure BDA0003446154150000057
A hierarchical regional enhanced network can be constructed, represented as
Figure BDA0003446154150000058
Wherein the node sets
Figure BDA0003446154150000059
Defined as a set of raw nodes
Figure BDA00034461541500000510
Union with the set of area virtual nodes, namely:
Figure BDA00034461541500000511
wherein the content of the first and second substances,
Figure BDA00034461541500000512
representing the virtual node associated with the k-th region. Use of
Figure BDA00034461541500000513
Indicating the number of nodes in the hierarchical regional enhanced network, then
Figure BDA00034461541500000514
Subsequently, an enhanced edge set is defined
Figure BDA00034461541500000515
In order to model the spatial/temporal correlation between nodes and regions and between different regions (including between the same and different hierarchies), the hierarchical region enhanced edge set must additionally contain node-region edges and region-region edges in addition to the original node-node edges. There are three different situations:
a) node-area membership: if node viDirect membership to an area
Figure BDA00034461541500000516
(i.e. the
Figure BDA00034461541500000517
Is containing node viIs in v), theniAnd
Figure BDA00034461541500000518
a node-area edge is established between the two nodes;
b) region-region membership: if region
Figure BDA00034461541500000519
Is directly affiliated to the area
Figure BDA00034461541500000520
Sub-area (i.e. of)
Figure BDA00034461541500000521
Is an inclusion region
Figure BDA00034461541500000522
The smallest area) of the first and second regions are
Figure BDA00034461541500000523
And
Figure BDA00034461541500000524
a region-region edge is established between the two;
c) region-region proximity relationship: if region
Figure BDA00034461541500000525
And area
Figure BDA00034461541500000526
Belong to the same level and have a proximity relation in the region division, then
Figure BDA0003446154150000061
And
Figure BDA0003446154150000062
between them, a region-region edge is established
Enhanced edge set
Figure BDA0003446154150000063
Is the union of the original network edge set epsilon and the edges added through the steps.
After the virtual node is added, the traffic characteristic matrix is reconstructed
Figure BDA0003446154150000064
To be reflected in the history and current time step
Figure BDA0003446154150000065
Regional awareness traffic information for individual nodes. For the original node, the traffic feature vector remains unchanged. The feature vector of each newly added area virtual node is an original node directly or indirectly attached to the areaMean of feature vectors of points. That is, for the first n rows of the feature matrix, the feature values are consistent with the original feature matrix,
Figure BDA0003446154150000066
for other rows, the eigenvalues are defined as the average of the eigenvectors of the original nodes directly or indirectly belonging to the corresponding region, i.e. the eigenvalues are the average of the eigenvectors of the original nodes directly or indirectly belonging to the corresponding region
Figure BDA0003446154150000067
Further as a preferred embodiment of the method, the region-aware spatial correlation model employs a multilayer graph attention network, the region-aware temporal correlation model employs a multilayer gated cyclic unit, and the step of learning spatial correlation and temporal correlation and outputting a prediction result by using a hierarchical region enhancement network and a traffic feature matrix as inputs of a prediction model specifically includes:
graph attention network enhancing network according to hierarchical regions
Figure BDA0003446154150000068
Spatial topology of, traffic characteristic matrix to be input
Figure BDA0003446154150000069
Conversion into a new feature matrix Xt；
Gated cyclic unit based on new feature matrix XtCalculating a corresponding hidden state matrix Ht；
Alternately passing through a graph attention network and a gate control circulation unit and recursively learning hidden states for multiple times, and then based on a hidden state matrix H of the last steptPerforming dimension transformation via a full link layer to obtain final prediction result
Figure BDA00034461541500000610
Further as a preferred embodiment of the method, the graph attention network enhances the network according to hierarchical regions
Figure BDA00034461541500000611
Spatial topology of, traffic characteristic matrix to be input
Figure BDA00034461541500000612
Conversion into a new feature matrix XtThis step, in particular, comprises:
for traffic characteristic matrix to be input
Figure BDA00034461541500000613
At each time step, modeling the spatial dependence relationship of the inter-node, the inter-node and inter-region perception based on the attention mechanism to obtain a corresponding new feature matrix Xt。
Specifically, fig. 4 shows a schematic diagram of a graph attention network (GAT) used in the present invention, which performs an aggregation operation on neighbor nodes through an attention mechanism to implement adaptive distribution of different neighbor weights. For node v1It first applies to each neighborhood node { v }1,v2,v3,v4,v5,v6Solving attention coefficients alpha respectively11,α12,α13,α14,α15α16Then, features of neighboring nodes are aggregated according to attention coefficients to learn a new feature vector [ X }t]1,:As input to a gated loop unit (GRU). Because the input network data is enhanced based on a hierarchical regionalization structure, the GAT can be used for establishing the node weight self-adaptive distribution of the regional perception, thereby capturing the spatial correlation of the regional perception.
The graph attention network (GAT) is used for modeling complex spatial dependency relationships among nodes, between nodes and regions and between regions, and automatically capturing the importance (attention coefficient) of a node to another node, a node to a region, a region to a node and a region to another region.
Specifically, nodes are encoded using a graph attention layer to convert traffic characteristics of the nodes into higher-level feature vectorsComplex spatial correlations between nodes and regions, and between regions. The input of the graph attention layer is a traffic feature vector of each node in the hierarchical regional enhanced network, namely
Figure BDA0003446154150000071
Wherein
Figure BDA0003446154150000072
The output is a new set of feature vectors, i.e.
Figure BDA0003446154150000073
Wherein
Figure BDA0003446154150000074
The specific calculation steps are as follows:
first, a self-attention mechanism is used for nodes in a network
Figure BDA0003446154150000075
Calculating to obtain the attention value between every two points:
Figure BDA0003446154150000076
wherein the content of the first and second substances,
Figure BDA0003446154150000077
is a weight matrix for linear transformation of node features. Specifically, the self-attention mechanism used in the present invention is a single layer feedforward neural network comprising a linear transformation vector
Figure BDA0003446154150000078
And a leak relu function (parameter α ═ 0.2) as the activation function, i.e.:
Figure BDA0003446154150000079
wherein，.TRepresenting a transpose operation and | l representing a join operation. Using a softmax function for the neighborhood nodes to obtain a normalized attention coefficient:
Figure BDA00034461541500000710
wherein the content of the first and second substances,
Figure BDA00034461541500000711
and representing the neighborhood nodes of the node i in the hierarchical regional enhanced network.
Finally, for each node, linear combination is obtained on the neighborhood node features by using the obtained normalized attention coefficient to update the node feature vector:
Figure BDA00034461541500000712
thus, the output feature matrix for all nodes is constructed as follows:
Figure BDA0003446154150000081
further as a preferred embodiment of the method, the gated loop unit is based on a new feature matrix XtCalculating a corresponding hidden state matrix HtThis step, in particular, comprises:
the gate control unit simultaneously models the time dependence relationship of region perception among nodes, between nodes and regions and between regions;
based on new feature matrix XtAnd the hidden state matrix H of the last time stept-1Obtaining the hidden state matrix H of the current time stept。
In particular, gating cell principle schematic referring to fig. 5, a time-series neural network is constructed using gated round-robin cells (GRUs) to model complex time-dependent relationships between nodes, between nodes and regions, and between regions and regions.
Assuming that the number of cells in the hidden layer is h, use is made of
Figure BDA0003446154150000082
Representing the hidden state matrix at time step t-1. For each layer of the neural network, the input at time t is given
Figure BDA0003446154150000083
And hidden state at time t-1
Figure BDA0003446154150000084
Reset gate for gated cycle unit
Figure BDA0003446154150000085
And a retrofit gate
Figure BDA0003446154150000086
Can be calculated by the following formulas respectively:
Figure BDA0003446154150000087
Figure BDA0003446154150000088
wherein the content of the first and second substances,
Figure BDA0003446154150000089
and
Figure BDA00034461541500000810
in order to be a weight matrix, the weight matrix,
Figure BDA00034461541500000811
is a bias vector. Subsequently, a candidate hidden state matrix in time step t is calculated
Figure BDA00034461541500000812
Figure BDA00034461541500000813
Wherein the content of the first and second substances,
Figure BDA00034461541500000814
and
Figure BDA00034461541500000815
in order to be a weight matrix, the weight matrix,
Figure BDA00034461541500000816
is a bias vector. Reset gate RtThe candidate hidden states for controlling the current time step include the degree of hidden states of the previous time step.
Hidden state H based on previous time stept-1And current candidate hidden state
Figure BDA00034461541500000817
The hidden state matrix of time step t can be updated by the update gate ZtThe weighted linear combination of the controls is calculated as:
Figure BDA0003446154150000091
finally, according to the hidden state matrix HtUsing the fully-connected layer to calculate the final output, i.e. the prediction of the time step t
Figure BDA0003446154150000092
Comprises the following steps:
Figure BDA0003446154150000093
wherein the content of the first and second substances,
Figure BDA0003446154150000094
in order to be a weight matrix, the weight matrix,
Figure BDA0003446154150000095
is a bias vector.
And finally, updating node characteristics and learning a hidden state based on time sequence by circularly using the GAT and the GRU, and finally, predicting the traffic flow by using a final output hidden state to obtain a prediction result.
In the model training process, use
Figure BDA0003446154150000096
And representing the real result of the training sample, namely the real value of the traffic information in the future q time steps of all the nodes in the enhanced network.
Model-based predictive value
Figure BDA0003446154150000097
And true value
Figure BDA0003446154150000098
The training loss function of the model is defined as follows:
Figure BDA0003446154150000099
where Φ represents all parameters of the model, λ is the hyper-parameter, Lreg(Φ) is a regularization term used to prevent model overfitting.
Further as a preferred embodiment of the method, the gated loop unit comprises a reset gate and an update gate, the reset gate determining a degree of forgetting information of a past time step, the update gate determining a degree of passing the information of the past time step to a next hidden state.
As shown in fig. 6, a traffic prediction system based on a spatiotemporal hierarchical network includes:
the preprocessing module is used for acquiring traffic data, preprocessing the traffic data and constructing a hierarchical regional enhanced network and a traffic characteristic matrix;
and the prediction module is used for taking the hierarchical region enhancement network and the traffic characteristic matrix as the input of the prediction model, learning the spatial correlation and the temporal correlation and outputting the prediction result.
The contents in the above method embodiments are all applicable to the present system embodiment, the functions specifically implemented by the present system embodiment are the same as those in the above method embodiment, and the beneficial effects achieved by the present system embodiment are also the same as those achieved by the above method embodiment.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (7)

1. A traffic prediction method based on a space-time hierarchical network is characterized by comprising the following steps:
acquiring traffic data, preprocessing the traffic data, and constructing to obtain a hierarchical regional enhanced network and a traffic characteristic matrix;
taking the hierarchical regional enhancement network and the traffic characteristic matrix as the input of a prediction model, learning the spatial correlation and the temporal correlation, and outputting a prediction result;
the predictive models include a region-aware spatial correlation model and a region-aware temporal correlation model.
2. The traffic prediction method based on the spatio-temporal hierarchical network as claimed in claim 1, wherein the step of obtaining the traffic data and preprocessing the traffic data to construct the hierarchical regional enhanced network and the traffic feature matrix specifically comprises:
acquiring original traffic data;
from the original road network
Figure FDA0003446154140000011
And hierarchical region structure
Figure FDA0003446154140000012
Adding the region as a virtual node into the network, extracting the connection relation between the node and the region and between the regions, and establishing a hierarchical regional enhanced network
Figure FDA0003446154140000013
Modeling traffic time sequence information in original traffic data as node attributes, and constructing a traffic characteristic matrix
Figure FDA0003446154140000014
3. The traffic prediction method based on the spatio-temporal hierarchical network as claimed in claim 2, wherein the region-aware spatial correlation model adopts a graph attention network, the region-aware temporal correlation model adopts a gated cyclic unit, and the step of taking the hierarchical region enhancement network and the traffic feature matrix as input of the prediction model, learning spatial correlation and temporal correlation, and outputting the prediction result specifically comprises:
graph attention network enhancing network according to hierarchical regions
Figure FDA0003446154140000015
Spatial topology of, traffic characteristic matrix to be input
Figure FDA0003446154140000016
Conversion into a new feature matrix Xt；
Gated cyclic unit based on new feature matrix XtCalculating a corresponding hidden state matrix Ht；
Alternately passing through a graph attention network and a gate control circulation unit and recursively learning hidden states for multiple times, and then based on a hidden state matrix H of the last steptPerforming dimension transformation via a full link layer to obtain final prediction result
Figure FDA0003446154140000017
4. The traffic prediction method based on the spatiotemporal hierarchical network as claimed in claim 3, wherein the graph attention network enhances the network according to the hierarchical region
Figure FDA0003446154140000018
Spatial topology of, traffic characteristic matrix to be input
Figure FDA0003446154140000019
Conversion into a new feature matrix XtThis step, in particular, comprises:
for traffic characteristic matrix to be input
Figure FDA0003446154140000021
At each time step, modeling the spatial dependence relationship of the inter-node, the inter-node and inter-region perception based on the attention mechanism to obtain a corresponding new feature matrix Xt。
5. The traffic prediction method based on spatio-temporal hierarchical network of claim 4, wherein the gated cyclic unit is based on a new feature matrix XtCalculating a corresponding hidden state matrix HtThis step, in particular, comprises:
the gate control unit simultaneously models the time dependence relationship of region perception among nodes, between nodes and regions and between regions;
based on new feature matrix XtAnd the hidden state matrix H of the last time stept-1Obtaining the hidden state matrix H of the current time stept。
6. The traffic prediction method based on spatio-temporal hierarchical network of claim 5, characterized in that the gated cyclic unit comprises a reset gate and an update gate, the reset gate determines the degree of forgetting the information of the past time step, and the update gate determines the degree of transferring the information of the past time step to the next hidden state.
7. A traffic prediction system based on a spatio-temporal hierarchical network, comprising:
the preprocessing module is used for acquiring traffic data, preprocessing the traffic data and constructing a hierarchical regional enhanced network and a traffic characteristic matrix;
and the prediction module is used for taking the hierarchical region enhancement network and the traffic characteristic matrix as the input of the prediction model, learning the spatial correlation and the temporal correlation and outputting the prediction result.
CN202111656768.0A 2021-12-30 2021-12-30 Traffic prediction method and system based on space-time hierarchical network Pending CN114254214A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111656768.0A CN114254214A (en) 2021-12-30 2021-12-30 Traffic prediction method and system based on space-time hierarchical network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111656768.0A CN114254214A (en) 2021-12-30 2021-12-30 Traffic prediction method and system based on space-time hierarchical network

Publications (1)

Publication Number Publication Date
CN114254214A true CN114254214A (en) 2022-03-29

Family

ID=80795855

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111656768.0A Pending CN114254214A (en) 2021-12-30 2021-12-30 Traffic prediction method and system based on space-time hierarchical network

Country Status (1)

Country Link
CN (1) CN114254214A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116910664A (en) * 2023-07-12 2023-10-20 南京航空航天大学 Cascade model-based flight ground guarantee dynamic prediction method

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116910664A (en) * 2023-07-12 2023-10-20 南京航空航天大学 Cascade model-based flight ground guarantee dynamic prediction method
CN116910664B (en) * 2023-07-12 2024-04-19 南京航空航天大学 Cascade model-based flight ground guarantee dynamic prediction method

Similar Documents

Publication Publication Date Title
CN113313947B (en) Road condition evaluation method of short-term traffic prediction graph convolution network
CN109887282B (en) Road network traffic flow prediction method based on hierarchical timing diagram convolutional network
CN110796110B (en) Human behavior identification method and system based on graph convolution network
CN111223301B (en) Traffic flow prediction method based on graph attention convolution network
CN112382082B (en) Method and system for predicting traffic running state in congested area
CN113053115B (en) Traffic prediction method based on multi-scale graph convolution network model
CN109508360B (en) Geographical multivariate stream data space-time autocorrelation analysis method based on cellular automaton
CN112183862A (en) Traffic flow prediction method and system for urban road network
CN112532439B (en) Network flow prediction method based on attention multi-component space-time cross-domain neural network model
CN112733656B (en) Skeleton action recognition method based on multiflow space attention diagram convolution SRU network
CN109344992B (en) Modeling method for user control behavior habits of smart home integrating time-space factors
CN110570035B (en) People flow prediction system for simultaneously modeling space-time dependency and daily flow dependency
CN114944053B (en) Traffic flow prediction method based on space-time hypergraph neural network
Alizadeh et al. Network traffic forecasting based on fixed telecommunication data using deep learning
CN113688253B (en) Hierarchical perception temporal knowledge graph representation learning method
CN115376318B (en) Traffic data compensation method based on multi-attribute fusion neural network
CN114005085A (en) Dense crowd distribution detection and counting method in video
Wan et al. Network traffic prediction based on LSTM and transfer learning
CN115346372A (en) Multi-component fusion traffic flow prediction method based on graph neural network
CN114780739A (en) Time sequence knowledge graph completion method and system based on time graph convolution network
CN116596109A (en) Traffic flow prediction model based on gating time convolution network
CN115376317A (en) Traffic flow prediction method based on dynamic graph convolution and time sequence convolution network
CN114254214A (en) Traffic prediction method and system based on space-time hierarchical network
CN115719036A (en) Space-time process simulation method and system based on stacked space-time memory unit
CN115862324A (en) Space-time synchronization graph convolution neural network for intelligent traffic and traffic prediction method

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