CN113935555B - Road network structure-based situation adaptive traffic prediction method and system - Google Patents
Road network structure-based situation adaptive traffic prediction method and system Download PDFInfo
- Publication number
- CN113935555B CN113935555B CN202111529992.3A CN202111529992A CN113935555B CN 113935555 B CN113935555 B CN 113935555B CN 202111529992 A CN202111529992 A CN 202111529992A CN 113935555 B CN113935555 B CN 113935555B
- Authority
- CN
- China
- Prior art keywords
- road section
- traffic
- road
- intersection
- feature map
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 230000003044 adaptive effect Effects 0.000 title claims description 12
- 238000010586 diagram Methods 0.000 claims abstract description 31
- 230000004927 fusion Effects 0.000 claims abstract description 9
- 230000002159 abnormal effect Effects 0.000 claims description 44
- 238000004364 calculation method Methods 0.000 claims description 18
- 230000007246 mechanism Effects 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 239000010410 layer Substances 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 239000002356 single layer Substances 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 230000008685 targeting Effects 0.000 abstract 1
- 238000005259 measurement Methods 0.000 description 9
- 230000000694 effects Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 206010039203 Road traffic accident Diseases 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 230000001932 seasonal effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011664 signaling Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 238000000547 structure data Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G06Q50/40—
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/12—Timing analysis or timing optimisation
Abstract
The invention discloses a situation self-adaptive traffic prediction method based on a road network structure and a model thereof, wherein the method comprises the following steps: targeting exceptional event datauTraffic data based on road segmentsxAnd traffic data based on intersectionsx’Calculating attention of multiple subgraphs to obtain a road section feature mapx ms Intersection characteristic diagramx’ ms (ii) a Will be provided withx ms Andx’ ms are respectively connected withxAndx’fusing and dynamically learning the importance of neighbor nodes in different neighborhood ranges to respectively obtain output road section feature mapsx mr Intersection characteristic diagramx’ mr (ii) a Will be provided withx mr Andx’ mr fusing the characteristics to obtain a road section characteristic diagramx mt (ii) a Based onx mt Learning the time dependence and outputting a traffic data prediction result; the model comprises the following steps: the system comprises a multi-sub-graph attention module, a multi-range attention module, a multi-task fusion module and a GRU module; the invention adaptively and simultaneously carries out traffic prediction on each road section, and has strong performance and high accuracy.
Description
Technical Field
The invention relates to the technical field of smart cities, in particular to a situation self-adaptive traffic prediction method and system based on a road network structure.
Background
Traffic prediction aims at predicting future traffic conditions (such as traffic speed, traffic flow, etc.) of roads from historical observations. As a basic function of an intelligent traffic system, a timely and accurate traffic prediction algorithm is important for urban traffic control and management.
Early research efforts for traffic prediction focused primarily on classical statistical models and shallow machine learning models, such as the Seasonal Arima (Seasonal autoregressive moving average model), Support Vector Regression (Support Vector Regression model), and K-Nearest Neighbor classification (K Nearest Neighbor) models. However, shallow features mined by the models cannot effectively represent complex nonlinear space-time correlation of the road network, and the traffic prediction performance is poor.
With the great success of deep learning in various fields, some studies have utilized a Convolutional Neural Network (Convolutional Neural Network) and a Recursive Neural Network (recurrent Neural Network) to obtain a spatio-temporal correlation from mesh-structured traffic data for accurate traffic prediction. Although the prediction performance of these models is improved compared to the shallow machine learning models, they are limited in that the input data must be standard mesh structure data in euclidean space.
In order to model non-euclidean correlations on irregular road networks, some studies integrate Graph conditional Network (Graph convolution Network) with RNN, CNN, and Attention Mechanism, which improves the performance of traffic prediction. However, since traffic measurement values under abnormal conditions (such as severe weather, traffic accidents, traffic congestion due to large activities such as road maintenance, concerts, and sports events, etc.) tend to deviate from daily traffic conditions, their performance is degraded without exception when various models are tested under abnormal conditions.
Only a few studies explicitly take into account traffic anomalies. However, they only use traffic data under exceptional conditions for performance testing, or activate different models for different traffic conditions, rather than explicitly modeling the spatiotemporal impact of local exceptional events on irregular road networks.
Therefore, it is a problem to be solved by those skilled in the art how to provide a road network structure-based situation adaptive traffic prediction method and system thereof, which can accurately predict the change trend of urban traffic measurement values under different traffic situations in real time.
Disclosure of Invention
In view of the above, the invention provides a road network structure-based situation adaptive traffic prediction method and system, aiming at accurately predicting the change trend of urban traffic measurement values under different traffic conditions in real time, helping users avoid congested road sections in advance and helping governments better perform traffic guidance.
In order to achieve the purpose, the invention adopts the following technical scheme:
a situation self-adaptive traffic prediction method based on a road network structure comprises the following steps:
s1, acquiring different event types according to abnormal event data u, classifying traffic data x based on road sections and traffic data x' based on intersections according to different event types, constructing a local sub-graph with dynamic spatial correlation for each type of event based on a multi-sub-graph attention machine system, and dividing the acquired local sub-graphs of different types of traffic data into b road section feature graphs xmsAnd b intersection feature maps x'ms；
S2, the road section characteristic diagram x is obtainedmsAnd the intersection feature map x'msRespectively fusing the data with initial traffic observation data x and x' according to corresponding event types, dynamically learning the importance of neighbor nodes in different neighborhood ranges according to fused characteristics, and respectively obtaining an output road section characteristic diagram xmrAnd intersection feature map x'mr；
S3, according to a parameter sharing strategy, the road section feature map x is obtainedmrAnd the intersection feature map x'mrThe features of the two feature maps are fused to obtain a common embedded x of the two feature mapsC ps；
S4, according to the multi-task learning module, the road section feature map x is obtainedmrAnd intersection feature map x'mrAnd common embedding xC psThe features of the road section are fused to obtain a road section feature map x with local short-term abnormal events and multi-range spatial informationmt；
S5, based on the road section characteristic diagram xmtAnd learning the time dependence, and outputting the traffic data prediction results after Q time slices.
Preferably, before performing S1, the method further includes encoding the initial traffic observation data in the first P time slices into the road-section-based traffic data x and the intersection-based traffic data x', which specifically includes the following:
respectively constructing a road network structure into a road section graph based on a road network and a cross graph based on an intersection according to the graph;
wherein the road segment map is a weighted graph G (V, E, A), and the traffic observation value of the road segment map is directly obtained from a traffic sensor; the corresponding intersection map is G ═ V ', E ', a ');
where V is a set of road segment sets, | V | ═ N, N is the total number of road segment nodes, E is a set of edge sets representing connectivity between nodes,
weighted adjacency matrix of G by Ai，jDenotes viAnd vjThe distance between two nodes;
the nodes in V 'are intersections between adjacent nodes in V, the edges in E' are nodes in the corresponding V, A 'is a weighted adjacency matrix, A'i，jRepresents node v'iAnd v'jAn abutting relationship therebetween;
the method comprises the following steps of giving a traffic observation value of each road section on a road section graph, and obtaining traffic data of a corresponding intersection node by adding traffic data of road sections connected to the intersection in order to obtain the traffic observation value of a specified intersection, wherein the specific calculation method comprises the following steps:
preferably, S1 specifically includes the following:
giving a historical traffic observation value x of a node set V and abnormal event data u in a corresponding time slice, and dividing the traffic data into B groups according to the characteristics of the abnormal event data u;
treating different nodes with the same type of exception event as K nearest neighbor node subsets of common virtual nodes v and v
And B attention weight calculations are carried out on the traffic data according to the following modes:
wherein B is more than or equal to 1 and less than or equal to B, and in the B-th calculation, the node viIs represented as
Is node v and its neighbor node viThe correlation between them; wbIs a hyper-parameter in the b-th calculation; | | denotes tandem operation; < - > represents an inner product operator; a (-) is a single layer feedforward neural network; the b-th calculated outputs are respectivelyIs composed of
And X'ms，b。
Preferably, S2 specifically includes the following:
for node viAll neighboring nodes of different neighborhood ranges within a hop distance of
Wherein L is more than or equal to 1 and less than or equal to L, and for adjacent nodes in different ranges, the multi-range attention mechanism calculation method comprises the following steps:
wherein the content of the first and second substances,
representing a node viThe features in the layer/are such that,
embedding neighborhood context, and initializing the neighborhood context into a random vector;
after obtaining the attention cross-correlation coefficient, calculating the node viThe linear combination of all adjacent nodes in the l-hop neighbor obtains the final node representation:
obtaining a road section characteristic map
And intersection feature map x'mr。
Preferably, the specific contents of S3 include:
the common non-linear transformation formula for embedding the shared parameters in both graphs is as follows:
XPs＝ReLU(XmrWPS+bPS)
X′PS＝ReLU(X′mrWPS+bPS)
obtaining a road section characteristic diagram xpsAnd intersection feature map x'psIs embedded in the public xC psThe method comprises the following steps:
wherein WpsAnd bpsAre shared weight parameters of the common non-linear transformation.
Preferably, the specific contents of S4 include:
obtaining the output road section characteristic diagram xmrAnd intersection feature map x'mrIs embedded in the public xC psThen, the road section characteristic map x is respectively inputmrAnd intersection feature map x'mrAnd learning the common embedded x using a multi-task learning moduleC psInputting a road section feature map xmrAnd intersection feature map x'mrThe corresponding importance of the three characteristics is obtained, and a road section characteristic diagram x with local short-term abnormal events and multi-range spatial information is obtainedmt。
A condition adaptive traffic prediction system based on a road network structure includes: the system comprises a multi-sub-graph attention module, a multi-range attention module, a multi-task fusion module and a GRU module;
the multi-sub-graph attention module is used for acquiring different event types according to the abnormal event data u, classifying the traffic data x based on the road section and the traffic data x' based on the intersection according to the different event types, constructing a local sub-graph with dynamic spatial correlation for each type of event based on a multi-sub-graph attention mechanism, and dividing the acquired local sub-graph into a road section feature graph xmsIntersection and crossingCharacteristic diagram x'ms；
The multi-range attention module is used for mapping the road section feature map xmsAnd the intersection feature map x'msRespectively fusing with the initial traffic observation data x and x', dynamically learning the importance of neighbor nodes in different neighborhood ranges according to the fused features, and respectively obtaining an output road section feature map xmrAnd intersection feature map x'mr；
The multi-task fusion module is used for sharing the road section feature map x according to a parameter sharing strategymrAnd the intersection feature map x'mrThe characteristics of the road section are fused to obtain a road section characteristic diagram xmt；
The GRU module is used for obtaining the road section characteristic diagram xmtAnd learning the time dependence, and outputting the traffic data prediction results after Q time slices.
Preferably, the system further comprises a traffic data coding module for coding the initial traffic observation data in the previous P time slices into the road section-based traffic data x and the intersection-based traffic data x'.
Compared with the prior art, the invention discloses a situation self-adaptive traffic prediction method and system based on a road network structure, and aims to accurately predict the change trend of urban traffic measurement values (such as traffic speed, traffic flow and the like) under different traffic conditions in real time, help users to avoid congested road sections in advance and help related departments to better guide traffic; the traffic condition self-adaptive traffic prediction is defined as that aiming at different traffic conditions, the traffic prediction is carried out on the road sections with or without abnormal events in a self-adaptive manner, a method different from the prior art is adopted, the area to be researched is divided into the road sections in the space dimension, the traffic data, the abnormal event data and the road characteristics are fused, the influence of different types of events on different road characteristics is modeled by utilizing multiple sub-graphs, and the accuracy of other road predictions is considered while the traffic prediction performance under the abnormal conditions is 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 embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a schematic flow chart of a situation adaptive traffic prediction method based on a road network structure according to the present invention;
FIG. 2 is a schematic diagram of a road network with traffic sensors in a situation adaptive traffic prediction method based on a road network structure according to the present invention;
FIG. 3 is a road segment diagram of a road network structure-based adaptive traffic prediction method according to the present invention;
fig. 4 is a cross-sectional view of a situation adaptive traffic prediction method based on a road network structure according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
To better illustrate this problem, the relevant definitions involved are as follows:
road network adjacency matrix: defining a road network as a directed weighted graph G ═ (V, E, a), where V is a set of road segments, | V | ═ N, N is the total number of road segment nodes, E is a set of edge sets representing connectivity between nodes,
weighted adjacency matrix of G by Ai，jDenotes viAnd vjThe distance between two nodes.
Traffic data: signaling urban traffic data collected from traffic sensors at time t as
Where C is the dimension of the collected traffic data (e.g., traffic speed, traffic flow, etc.).
Abnormal event data: defining exception event data as a D-dimensional vector
Wherein the D-dimensional characteristics comprise event occurrence time t and occurrence road section viEvent type (e.g., car accident), road structure characteristics (e.g., center divider), etc.
Based on the above, the situation-adaptive traffic prediction problem is modeled as a spatio-temporal prediction problem: the observations for the N nodes for the first p timeslices of a given history are:
and M abnormal events observed in p time intervals are:
the situation-adaptive traffic prediction problem will learn a function f that can predict the traffic data for the next Q time slices of all nodes
Based on the problem model, the designed traffic prediction method flow of the situation self-adaption mainly comprises two parts of traffic data coding and traffic prediction.
The embodiment of the invention discloses a condition self-adaptive traffic prediction method based on a road network structure, which comprises the following steps as shown in figure 1:
s1, acquiring different event types according to abnormal event data u, classifying traffic data x based on road sections and traffic data x' based on intersections according to different event types, constructing a local sub-graph with dynamic spatial correlation for each type of event based on a multi-sub-graph attention machine system, and dividing the acquired local sub-graphs of different types of traffic data into b road section feature graphs xmsAnd b intersection feature maps x'ms；
S2, road section feature map xmsAnd intersection feature map x'msRespectively fusing the data with initial traffic observation data x and x' according to corresponding event types, dynamically learning the importance of neighbor nodes in different neighborhood ranges according to fused characteristics, and respectively obtaining an output road section characteristic diagram xmrAnd intersection feature map x'mr；
S3, according to the parameter sharing strategy, the road section feature map x is obtainedmrAnd intersection feature map x'mrThe features of the two feature maps are fused to obtain a common embedded x of the two feature mapsC ps；
S4, according to the multi-task learning module, the road section feature map x is obtainedmrAnd intersection feature map x'mrAnd common embedding xC psThe features of the road section are fused to obtain a road section feature map x with local short-term abnormal events and multi-range spatial informationmt；
S5, based on road section characteristic diagram xmtAnd learning the time dependence, and outputting the traffic data prediction results after Q time slices.
In order to further implement the above technical solution, as shown in fig. 2 to 4, the road network passing graph is respectively constructed as a road segment graph based on a road network and an intersection graph based on an intersection.
Before proceeding to S1, the method further includes encoding the initial traffic observation data in the first P time slices into the road-section-based traffic data x and the intersection-based traffic data x', which specifically includes the following:
respectively constructing a road network structure into a road section graph based on a road network and a cross graph based on an intersection according to the graph;
the road section map is a weighted graph G (V, E, A), and the traffic observation value of the road section map is directly obtained from a traffic sensor; the corresponding intersection map is G ═ V ', E ', a ');
where V is a set of road segment sets, | V | ═ N, N is the total number of road segment nodes, E is a set of edge sets representing connectivity between nodes,
weighted adjacency matrix of G by Ai，jDenotes viAnd vjThe distance between two nodes;
the nodes in V 'are intersections between adjacent nodes in V, the edges in E' are nodes in the corresponding V, A 'is a weighted adjacency matrix, A'i，jRepresents node v'iAnd v'jAn abutting relationship therebetween;
the method comprises the following steps of giving a traffic observation value of each road section on a road section graph, and obtaining traffic data of a corresponding intersection node by adding traffic data of road sections connected to the intersection in order to obtain the traffic observation value of a specified intersection, wherein the specific calculation method comprises the following steps:
it should be noted that:
road networks with traffic sensors as shown in fig. 2, since the effects of abnormal events on surrounding road segments and intersections are always relevant, in order to fully exploit such effects from different perspectives, traffic observation data is encoded in preparation for subsequent enhanced prediction performance.
In order to further implement the above technical solution, S1 specifically includes the following contents:
giving a historical traffic observation value x of a node set V and abnormal event data u in a corresponding time slice, and dividing the traffic data into B groups according to the characteristics of the abnormal event data u;
treating different nodes with the same type of exception event as K nearest neighbor node subsets of common virtual nodes v and v
And B attention weight calculations are carried out on the traffic data according to the following modes:
wherein B is more than or equal to 1 and less than or equal to B, and in the B-th calculation, the node viIs represented as
Is node v and its neighbor node viThe correlation between them; wbIs a hyper-parameter in the b-th calculation; | | denotes tandem operation; < - > represents an inner product operator; a (-) is a single layer feedforward neural network; the output after the b-th calculation is respectively
And X'ms，b。
It should be noted that:
abnormal events around the road network often interrupt the periodic pattern of the conventional traffic data, and the characteristics of the abnormal events have different influences on the traffic conditions of adjacent road sections and intersections. However, the distribution of the abnormal events in the road network is not uniform, so that in real life, there is a problem that there is no historical abnormal event in part of the road sections, i.e. the event data is sparse. Thus, multi-subgraph attention is required to capture the features of historical traffic observations with different types of exceptional events as a common potential feature of traffic changes under exceptional events for all nodes. When an exception event occurs, the common underlying characteristics of the exception event are applied to nodes that do not have sufficient historical exception event data for traffic prediction.
In order to further implement the above technical solution, S2 specifically includes the following contents:
for node viAll neighboring nodes of different neighborhood ranges within a hop distance of
Wherein L is more than or equal to 1 and less than or equal to L, and for adjacent nodes in different ranges, the multi-range attention mechanism calculation method comprises the following steps:
wherein the content of the first and second substances,
representing a node viThe features in the layer/are such that,
embedding neighborhood context, and initializing the neighborhood context into a random vector;
after obtaining the attention cross-correlation coefficient, calculating the node viThe linear combination of all adjacent nodes in the l-hop neighbor obtains the final node representation:
obtaining a road segmentCharacteristic diagram
And intersection feature map x'mr。
It should be noted that:
the traffic conditions of a node are affected by other neighboring nodes, and this spatial correlation is highly dynamic and varies with time and different types of surrounding events. To model this dynamic effect, a multi-range attention mechanism is introduced to adaptively capture the correlation of nodes in different neighborhoods. The attention mechanism dynamically allocates different weights to different adjacent nodes in the same neighborhood range in different time steps, and node information from different neighborhoods can be effectively and dynamically aggregated.
The outputs of S1 are known to be
And x'mrA dynamic spatial representation is captured from multiple neighborhood ranges using a multi-range attention layer.
In order to further implement the above technical solution, the specific content of S3 includes:
the common non-linear transformation formula for embedding the shared parameters in both graphs is as follows:
xPS＝ReLU(XmrWPS+bPS)
X′PS＝ReLU(X′mrWPS+bPS)
obtaining a road section characteristic diagram xpsAnd intersection feature map x'psIs embedded in the public xC psThe method comprises the following steps:
wherein WpsAnd bpsAre shared weight parameters of the common non-linear transformation.
It should be noted that:
in order to learn the relevance of the road section and the intersection at the same time, a multitask fusion scheme is introduced, and the hidden relevance between the road section and the intersection is coupled so as to mutually enhance the prediction performance. In order to learn common information shared by a road section graph and a cross graph in a multi-range attention module, a parameter sharing strategy is introduced.
In order to further implement the above technical solution, the specific content of S4 includes:
obtaining the output road section characteristic diagram xmrAnd intersection feature map x'mrIs embedded in the public xC psThen, the road section characteristic map x is respectively inputmrAnd intersection feature map x'mrAnd learning the common embedded x using a multi-task learning moduleC psInputting a road section feature map xmrAnd intersection feature map x'mrThe corresponding importance of the three characteristics is obtained, and a road section characteristic diagram x with local short-term abnormal events and multi-range spatial information is obtainedmt。
A condition adaptive traffic prediction system based on a road network structure includes: the system comprises a multi-sub-graph attention module, a multi-range attention module, a multi-task fusion module and a GRU module;
a multi-sub-graph attention module for acquiring different event types according to the abnormal event data u, classifying the traffic data x based on the road section and the traffic data x' based on the intersection according to the different event types, constructing a local sub-graph with dynamic spatial correlation for each type of event based on a multi-sub-graph attention mechanism, and dividing the acquired local sub-graph into a road section feature graph xmsAnd intersection feature map x'ms；
A multi-range attention module for mapping a road segment feature map xmsAnd intersection feature map x'msRespectively fusing with the initial traffic observation data x and x', dynamically learning the importance of neighbor nodes in different neighborhood ranges according to the fused features, and respectively obtaining an output road section feature map xmrAnd intersection feature map x'mr；
A multi-task fusion module for sharing the road section feature map x according to the parameter sharing strategymrAnd intersection feature map x'mrThe characteristics of the road section are fused to obtain a road section characteristic diagram xmt；
GRU module for road section characteristic diagram xmtAnd learning the time dependence, and outputting the traffic data prediction results after Q time slices.
In order to further implement the technical scheme, the system further comprises a traffic data coding module, which is used for coding the initial traffic observation data in the previous P time slices into the traffic data x based on the road section and the traffic data x' based on the intersection.
The prior art mainly aims at periodic traffic prediction under a normal condition, although dynamic space-time correlation among different nodes of a road network is considered, traffic measurement values under an abnormal condition are usually seriously deviated from the normal condition, most of the prior art methods do not clearly distinguish normal traffic conditions from abnormal traffic conditions, and the prediction performance is greatly reduced under the abnormal condition. The method improves the accuracy of overall prediction while considering the prediction performance under normal conditions by constructing a two-way network structure and utilizing potential influence characteristics extracted from different abnormal conditions and road structure information, and respectively compares the prior art with the applied art under the same conditions as follows:
(1) the method specifically predicts the traffic measurement values under normal and abnormal conditions by using the events and the road structure information, and excavates the difference and identity between the traffic measurement values under the abnormal conditions and the traffic measurement values under the normal conditions by analyzing the time-space relationship among the abnormal conditions, the road structure information and the traffic measurement values, thereby further improving the traffic prediction performance under different abnormal events;
(2) in the method, multi-task multi-range multi-sub-graph attention calculation is carried out, and abnormal flow and event data (such as traffic accidents) are included in a fitting process so as to clearly model the influence of the abnormal events, and the abnormal events are not only used for performance test. Firstly, a multi-subgraph attention mechanism is utilized, different subgraphs are generated in a self-adaptive mode according to the characteristics of abnormal events to model the node characteristics, and the problem of sparse data of the abnormal events can be effectively solved; meanwhile, an attention mechanism is used on each subgraph to automatically capture the dynamic spatial correlation of adjacent nodes; secondly, a multi-range attention mechanism is introduced to aggregate information of different neighborhoods, and the importance of different neighborhood ranges is dynamically adjusted so as to automatically capture the influence of abnormal events; in addition, in order to further capture fine-grained spatial correlation, the non-Euclidean correlation of the road network is coded into a road section graph and a cross graph, and on the basis, a multitask fusion module with a parameter sharing strategy is used for learning the hidden correlation between the road section and the intersection.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (7)
1. A situation self-adaptive traffic prediction method based on a road network structure is characterized by comprising the following steps:
s1, acquiring different event types according to abnormal event data u, classifying traffic data x based on road sections and traffic data x' based on intersections according to different event types, constructing a local sub-graph with dynamic spatial correlation for each type of event based on a multi-sub-graph attention machine system, and dividing the acquired local sub-graphs of different types of traffic data into b road section feature graphs xmsAnd b intersection feature maps x'ms；
S2. will instituteRoad section characteristic diagram xmsAnd the intersection feature map x'msRespectively fusing the data with initial traffic observation data x and x' according to corresponding event types, dynamically learning the importance of neighbor nodes in different neighborhood ranges according to fused characteristics, and respectively obtaining an output road section characteristic diagram xmrAnd intersection feature map x'mr；
S3, according to a parameter sharing strategy, the road section feature map x is obtainedmrAnd the intersection feature map x'mrThe features of the two feature maps are fused to obtain the common embedding of the two feature maps
S4, according to the multi-task learning module, the road section feature map x is obtainedmrAnd intersection feature map x'mrAnd common embedding
The features of the road section are fused to obtain a road section feature map x with local short-term abnormal events and multi-range spatial informationmt；
S5, based on the road section characteristic diagram xmtLearning the time dependency and outputting traffic data prediction results after Q time slices;
s1 specifically includes the following:
giving a historical traffic observation value x of a node set V and abnormal event data u in a corresponding time slice, and dividing the traffic data into B groups according to the characteristics of the abnormal event data u;
treating different nodes with the same type of exception event as K nearest neighbor node subsets of common virtual nodes v and v
And B attention weight calculations are carried out on the traffic data according to the following modes:
wherein B is more than or equal to 1 and less than or equal to B, and in the B-th calculation, the node viIs represented as
Is node v and its neighbor node viThe correlation between them; wbIs a hyper-parameter in the b-th calculation; | | denotes tandem operation; < - > represents an inner product operator; a (-) is a single layer feedforward neural network; the output after the b-th calculation is respectively
And
2. the method according to claim 1, further comprising encoding initial traffic observation data in the first P time slices into road segment-based traffic data x and intersection-based traffic data x' before S1, wherein the method specifically comprises the following steps:
respectively constructing a road network structure into a road section graph based on a road network and a cross graph based on an intersection according to the graph;
wherein the road segment map is a weighted graph G (V, E, A), and the traffic observation value of the road segment map is directly obtained from a traffic sensor; the corresponding intersection map is G ═ V ', E ', a ');
where V is a set of road segment sets, | V | ═ N, N is the total number of road segment nodes, E is a set of edge sets representing connectivity between nodes,
weighted adjacency matrix of G by Ai,jDenotes viAnd vjThe distance between two nodes;
the nodes in V 'are intersections between adjacent nodes in V, the edges in E' are nodes in the corresponding V, A 'is a weighted adjacency matrix, A'i,jRepresents node v'iAnd v'jAn abutting relationship therebetween;
the method comprises the following steps of giving a traffic observation value of each road section on a road section graph, and obtaining traffic data of a corresponding intersection node by adding traffic data of road sections connected to the intersection in order to obtain the traffic observation value of a specified intersection, wherein the specific calculation method comprises the following steps:
3. the method for predicting traffic adaptively according to the situation of the road network structure as claimed in claim 1, wherein S2 specifically includes the following contents:
for node viAll neighboring nodes of different neighborhood ranges within a hop distance of
Wherein L is more than or equal to 1 and less than or equal to L, and for adjacent nodes in different ranges, the multi-range attention mechanism calculation method comprises the following steps:
wherein the content of the first and second substances,
representing a node viThe features in the layer/are such that,
embedding neighborhood context, and initializing the neighborhood context into a random vector;
after obtaining the attention cross-correlation coefficient, calculating the node viThe linear combination of all adjacent nodes in the l-hop neighbor obtains the final node representation:
obtaining a road section characteristic map
And intersection feature map x'mr。
4. The method according to claim 1, wherein the details of S3 include:
the common non-linear transformation formula for embedding the shared parameters in both graphs is as follows:
obtaining a road section characteristic diagram xpsAnd intersection feature map x'psIn public
The method comprises the following steps:
wherein WpsAnd bpsAre shared weight parameters of the common non-linear transformation.
5. The method according to claim 1, wherein the details of S4 include:
obtaining the output road section characteristic diagram xmrAnd intersection feature map x'mrIn public
Then, the road section characteristic map x is respectively inputmrAnd intersection feature map x'mrAnd learning common embedding using a multitask learning module
Inputting road section characteristic diagram xmrAnd intersection feature map x'mrThe corresponding importance of the three characteristics is obtained, and a road section characteristic diagram x with local short-term abnormal events and multi-range spatial information is obtainedmt。
6. A condition adaptive traffic prediction system based on a road network structure, a condition adaptive traffic prediction method based on a road network structure according to any one of claims 1 to 5, comprising: the system comprises a multi-sub-graph attention module, a multi-range attention module, a multi-task fusion module and a GRU module;
the multi-subgraph attention module is used for acquiring abnormal event datau, acquiring different event types, classifying the traffic data x based on the road section and the traffic data x' based on the intersection according to the different event types, constructing a local sub-graph with dynamic spatial correlation for each type of event based on a multi-sub-graph attention machine system, and dividing the acquired local sub-graph into a road section feature graph xmsAnd intersection feature map x'ms；
The multi-range attention module is used for mapping the road section feature map xmsAnd the intersection feature map x'msRespectively fusing with the initial traffic observation data x and x', dynamically learning the importance of neighbor nodes in different neighborhood ranges according to the fused features, and respectively obtaining an output road section feature map xmrAnd intersection feature map x'mr；
The multi-task fusion module is used for sharing the road section feature map x according to a parameter sharing strategymrAnd the intersection feature map x'mrThe characteristics of the road section are fused to obtain a road section characteristic diagram xmt；
The GRU module is used for obtaining the road section characteristic diagram xmtAnd learning the time dependence, and outputting the traffic data prediction results after Q time slices.
7. The system according to claim 6, further comprising a traffic data encoding module for encoding the initial traffic observation data in the previous P time slices into the road section-based traffic data x and the intersection-based traffic data x'.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111529992.3A CN113935555B (en) | 2021-12-15 | 2021-12-15 | Road network structure-based situation adaptive traffic prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111529992.3A CN113935555B (en) | 2021-12-15 | 2021-12-15 | Road network structure-based situation adaptive traffic prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113935555A CN113935555A (en) | 2022-01-14 |
| CN113935555B true CN113935555B (en) | 2022-03-18 |
Family
ID=79289113
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111529992.3A Active CN113935555B (en) | 2021-12-15 | 2021-12-15 | Road network structure-based situation adaptive traffic prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113935555B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111161535A (en) * | 2019-12-23 | 2020-05-15 | 山东大学 | Attention mechanism-based graph neural network traffic flow prediction method and system |
| CN112435462A (en) * | 2020-10-16 | 2021-03-02 | 同盾控股有限公司 | Method, system, electronic device and storage medium for short-time traffic flow prediction |
| CN112927507A (en) * | 2021-02-04 | 2021-06-08 | 南京航空航天大学 | Traffic flow prediction method based on LSTM-Attention |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113487856B (en) * | 2021-06-04 | 2022-10-14 | 兰州理工大学 | Traffic flow combination prediction model based on graph convolution network and attention mechanism |
| CN113627676B (en) * | 2021-08-18 | 2023-09-01 | 湘潭大学 | Traffic prediction method and system based on multi-attention causal relationship |
-
2021
- 2021-12-15 CN CN202111529992.3A patent/CN113935555B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111161535A (en) * | 2019-12-23 | 2020-05-15 | 山东大学 | Attention mechanism-based graph neural network traffic flow prediction method and system |
| CN112435462A (en) * | 2020-10-16 | 2021-03-02 | 同盾控股有限公司 | Method, system, electronic device and storage medium for short-time traffic flow prediction |
| CN112927507A (en) * | 2021-02-04 | 2021-06-08 | 南京航空航天大学 | Traffic flow prediction method based on LSTM-Attention |
Non-Patent Citations (1)
| Title |
|---|
| 图卷积神经网络综述;徐冰冰等;《计算机学报》;20200531;第43卷(第5期);第755-776页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113935555A (en) | 2022-01-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cai et al. | PSO-ELM: A hybrid learning model for short-term traffic flow forecasting | |
| CN111612206B (en) | Neighborhood people stream prediction method and system based on space-time diagram convolution neural network | |
| Iranitalab et al. | Comparison of four statistical and machine learning methods for crash severity prediction | |
| Li et al. | Estimation of missing values in heterogeneous traffic data: Application of multimodal deep learning model | |
| Koesdwiady et al. | Improving traffic flow prediction with weather information in connected cars: A deep learning approach | |
| Elhenawy et al. | Dynamic travel time prediction using data clustering and genetic programming | |
| Kim et al. | Urban traffic flow prediction system using a multifactor pattern recognition model | |
| CN111160753B (en) | Road network node importance assessment method based on knowledge graph | |
| CN101216998B (en) | An urban traffic flow information amalgamation method of evidence theory based on fuzzy rough sets | |
| George et al. | Traffic prediction using multifaceted techniques: A survey | |
| EP2590151A1 (en) | A framework for the systematic study of vehicular mobility and the analysis of city dynamics using public web cameras | |
| Chen et al. | Multi-step prediction of experienced travel times using agent-based modeling | |
| Chen et al. | A multiscale-grid-based stacked bidirectional GRU neural network model for predicting traffic speeds of urban expressways | |
| CN114944053A (en) | Traffic flow prediction method based on spatio-temporal hypergraph neural network | |
| Xu et al. | Urban short-term traffic speed prediction with complicated information fusion on accidents | |
| Cheong et al. | Missing traffic data imputation for artificial intelligence in intelligent transportation systems: review of methods, limitations, and challenges | |
| Zhang et al. | Off-deployment traffic estimation—a traffic generative adversarial networks approach | |
| Menguc et al. | A Data Driven Approach to Forecasting Traffic Speed Classes Using Extreme Gradient Boosting Algorithm and Graph Theory | |
| Zheng et al. | Rebuilding city-wide traffic origin destination from road speed data | |
| CN113935555B (en) | Road network structure-based situation adaptive traffic prediction method and system | |
| CN116486624A (en) | Traffic flow prediction method and system based on space-time diagram convolutional neural network | |
| Hu et al. | Spatio-temporal dynamic change mechanism analysis of traffic conflict risk based on trajectory data | |
| CN116386020A (en) | Method and system for predicting exit flow of highway toll station by multi-source data fusion | |
| Kamal et al. | Ordinal-ResLogit: Interpretable deep residual neural networks for ordered choices | |
| Al Mamlook et al. | Smart Traffic Control System for Dubai: A Simulation Study Using YOLO Algorithms |
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 | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 211500 third floor, Beidou building, No. 6, Huida Road, Jiangbei new area, Nanjing, Jiangsu Patentee after: Hualui Cloud Technology Co.,Ltd. Address before: 211500 third floor, Beidou building, No. 6, Huida Road, Jiangbei new area, Nanjing, Jiangsu Patentee before: Hua Lu Yun Technology Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder |