CN116166975A - Multi-graph hierarchical road surface characterization method based on graph neural network - Google Patents
Multi-graph hierarchical road surface characterization method based on graph neural network Download PDFInfo
- Publication number
- CN116166975A CN116166975A CN202310169056.9A CN202310169056A CN116166975A CN 116166975 A CN116166975 A CN 116166975A CN 202310169056 A CN202310169056 A CN 202310169056A CN 116166975 A CN116166975 A CN 116166975A
- Authority
- CN
- China
- Prior art keywords
- road
- structural
- matrix
- functional
- graph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012512 characterization method Methods 0.000 title claims abstract description 35
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 25
- 238000000034 method Methods 0.000 claims abstract description 21
- 230000003595 spectral effect Effects 0.000 claims abstract description 7
- 239000011159 matrix material Substances 0.000 claims description 98
- 230000006870 function Effects 0.000 claims description 44
- 239000013598 vector Substances 0.000 claims description 26
- 230000004913 activation Effects 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 8
- 230000002457 bidirectional effect Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000005096 rolling process Methods 0.000 claims description 6
- 238000012546 transfer Methods 0.000 claims description 6
- 230000008447 perception Effects 0.000 claims description 5
- 230000004931 aggregating effect Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000011144 upstream manufacturing Methods 0.000 claims description 3
- 230000003915 cell function Effects 0.000 claims 1
- 230000009286 beneficial effect Effects 0.000 abstract description 4
- 238000010276 construction Methods 0.000 abstract description 4
- 230000007246 mechanism Effects 0.000 abstract description 3
- 238000002347 injection Methods 0.000 abstract 1
- 239000007924 injection Substances 0.000 abstract 1
- 101100395452 Mus musculus Hrnr gene Proteins 0.000 description 5
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a multi-graph hierarchical road map characterization method based on a map neural network, wherein the map neural network is perceived through a structure composed of spectral clustering and map injection force networks, road map characterization of different layers is modeled, two virtual nodes, namely a structural area and a functional area, are introduced, a multi-graph mechanism is adopted to guide the virtual nodes to correspond to the structural area and the functional area of the real world, structural similarity is established by utilizing road type attribute, functional similarity among road sections is defined by utilizing urban POI information, message sharing is executed at a high layer, updated information is transmitted to a lower layer node, and the functional attribute of the road map is supplemented; the road network characterization comprising structural features and functional features can be obtained by using the method; the road network structure and the function roles are convenient to determine; the method is beneficial to revealing the functional areas of cities, route planning, arrival time estimation and position prediction, and construction of an intelligent traffic system.
Description
Technical Field
The invention relates to the technical field of intelligent transportation, in particular to a multi-graph hierarchical road surface condition method based on a graph neural network.
Background
Currently, intelligent transportation systems (Intelligent Transportation System, ITS) have become an integral part of people's daily lives, playing a vital role in different transportation applications, such as route planning, time of arrival estimation and next location prediction. Road networks serve as the infrastructure of ITS and play a very important role in various traffic-related systems and applications.
Because of its important role, it is highly necessary to develop a suitable, and in particular versatile, method to efficiently characterize and model a road network. Early researches mainly used road network topology as constraint and adopted standard graph data structure development model. With the development of deep learning, recent research is turned to the feature learning of road segments by using a graph neural network. In this way, the fundamental features of the road network can be efficiently exploited and utilized, which is expected to improve the performance of downstream applications.
However, due to the complexity of the road network, developing an efficient characterization learning model is not easy. In previous studies, at least three major problems have not been well addressed. First, the road network is not "flat". Road segments are naturally divided into different "clusters" of traffic units, including structural units (e.g., traffic junctions) or functional units (e.g., business areas). Furthermore, the roles of different road segments in a road network are not "equal". Some traffic units play an increasingly important role in various traffic tasks. In the prior art, the standard graph neural network is generally adopted to equally process different nodes, and the hierarchical characteristics of the nodes are ignored. Second, the road network may not be a "small world," which tends to have a longer average path. For example, the relationship of different road segments in a road network will far exceed the theory of "six degrees of separation". However, in standard graph neural networks, they only aggregate information from nearby nodes, and cannot effectively capture long-range dependencies between nodes. Third, road networks mainly exhibit structural features, while other aspects of information may not be available through the network structure. For example, it is often difficult to determine its functional role (e.g., shopping mall) based solely on the road connection of the traffic unit.
Therefore, providing a multi-graph hierarchical road surface characterization method based on a graph neural network is a problem that needs to be solved by those skilled in the art.
Disclosure of Invention
In view of the above technical problems, the present invention provides a multi-graph hierarchical road map characterization method based on a graph neural network, which at least solves some of the above technical problems; by using the method, the road network representation comprising the structural characteristics and the functional characteristics can be effectively obtained; the road network structure and the function roles are convenient to determine; the method is beneficial to revealing the functional areas of cities, route planning, arrival time estimation and position prediction, and construction of an intelligent traffic system.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
a multi-graph hierarchical road surface characterization method based on a graph neural network comprises the following steps:
s1, giving a road section, and carrying out characteristic representation on the road section;
s2, establishing a relation between road sections, and constructing a natural connectivity adjacency matrix, a structural similarity matrix and a functional similarity matrix between the road sections;
s3, setting each road section to correspond to a single structural area, wherein different road sections have different importance degrees in one structural area, and obtaining a structural area distribution matrix for representing important relations between the road sections and the structural areas through a structural perception graph neural network formed by spectral clustering and graph annotation force networks; and associating the structural area with the road section;
s4, constructing a functional area based on the structural area with the same function, constructing a functional area distribution matrix, and expressing the functional area representation as a weighted linear combination of the structural area representation based on the functional area distribution matrix to obtain an adjacent matrix of the functional area nodes;
s5, updating the representation of the functional area, and transmitting updated information through the sequence of the functional area, the structural area and the road section.
Further, the step S1 includes:
giving a road section, taking the road section serial number ID, the number LN of lanes, the road section length SL and the longitude and latitude LL of all the time into consideration, performing context embedding, dividing the whole value range into a plurality of contact intervals for the contact characteristics, performing feature coding by using interval numbers, setting a unique embedded connection for each value, connecting the associated vectors in series to be used as context embedding, and initializing graph node embedding:
wherein ,vID Representing the feature vector, v, corresponding to the road segment serial number ID LN Feature vector v representing lane number LN SL Representing the corresponding feature vector, v, of the link length SL LL Representing the feature vector corresponding to the latitude and longitude LL,
representing road segment s i The feature V of all nodes is the initial road segment feature N S 。
Further, the step S2 includes:
s21, establishing an adjacency matrix of the road section based on the natural connectivity of the road network by using
Indicating that if one road section is upstream of another road section +.>
If the corresponding value in (a) is 1, otherwise, 0, adding two directional connections for the bidirectional road sections by reversing the starting point and the ending point of the bidirectional road sections;
s22, designing a structural code according to the road types, and setting the road types to be m types, wherein the road types are S for each road section i Generating an m-dimensional vector
At->
In which only the road section s is met i The category position is 1, the rest positions are 0, and each road section s is then divided into two sections i Structural code sum s i Splicing the sum of the structural codes of adjacent surrounding road sections, and performing normalization operation:
wherein ,
representing road segment s i Structural coding features of->
Representing road segment s j Is indicative of a concatenation operation, N (s i ) Representation and road segment s i A set of adjacent road segments, |x| 2 Representing the modulus of the vector x to obtain a structural matrix
wherein kS For the number of road segments, each row in F represents the structural feature of one road segment in the road network, and the cosine similarity is used for measuring the structural similarity of two road segments:
SS＝F·F T
generating a mask matrix to filter some edges with smaller similarity:
wherein ,SSmask Is a mask matrix, SS mask [i,j]Representing an element located in the ith row and jth column, θ 1 Is a similarity threshold, multiplies the structural similarity by a mask matrix, andgenerating final structure similarity matrix through an activation function
/>
wherein ,
the representation matrix is multiplied by elements, and sigma represents a sigmoid function;
s23, generating a single thermal code for each POI according to the type of the POI, wherein n types are shared by the POI, and p is the POI j Generating an n-dimensional vector
At->
In which only the POI p is met j The type of location is 1, the rest of locations are 0, consider the impact of each POI on its surrounding road segments:
wherein ,pj Representing points of interest, dis (p j ,s i ) Representing POI p j To road section s i Distance, theta 2 Is a distance threshold, all POIs are for road segment s i The influence of (a) forms a road segment s i Is characterized by the following functions:
obtaining a functional matrix
Each row in G represents a functional feature of a road segment in the road network, R represents linear space, k S N represents the dimension for the number of road segments; then the cosine similarity is used to measure the functional similarity of two road sections, and the two road sections are used to generate a final functional similarity matrix through an activation function>
Where σ represents the sigmoid function and the superscript T represents the transpose.
Further, the step S3 includes:
s31, dividing structural areas by adopting a spectral clustering algorithm, and giving an adjacency matrix of road sections
By subtracting the diagonal matrix D S Deriving its graph Laplace matrix L S Calculating the previous d' eigenvectors u 1 ,…u d' Obtaining a matrix consisting of d' eigenvectors>
R represents a linear space, < >>
Represented by k S Linear space formed by x d' dimensional matrix, by running standard K-means algorithm on matrix U, a hard map M from location to structural area is obtained 1 ；
Adjacency matrix
Representing natural connectivity between road segments, < >>
Is the structure of road sectionThe similarity matrix encodes the structural similarity into connection between road sections, models the two graphs by adopting two graph annotation force networks respectively, and aggregates the two graphs together to obtain soft mapping M 2 I.e.
Wherein GAT is the graph annotation force network, softmax (·) is a line normalization function, N S Representing a road section representation; finally, combining the hard mapping and the soft mapping to obtain a final constructed structural area allocation matrix A SR ：
A SR ＝softmax(αM 1 +(1-α)M 2 )
Wherein α is a hyper-parameter controlling fusion of the two mappings;
s32, through A SR Associating the structural zone representation with the road segment representation:
wherein ,NR Representing structural region features;
and meanwhile, the connection relation of the structural area and the connection relation of the road section are associated:
wherein ,
representing the structural region connection matrix.
Further, the step S4 includes:
s41, function adjacency matrix
Transfer from road segment to structure zone:
wherein ,
representing a structural region functional similarity matrix;
modeling the natural connectivity and the functional similarity of the structural areas by adopting two graph-annotation force networks respectively, and aggregating the two graph-annotation force networks to obtain a final functional area distribution matrix A RZ I.e.
S42, utilizing the function area allocation matrix A RZ The functional region representations are represented as weighted linear combinations of structural region representations:
by N Z Further deriving an adjacency matrix of the functional area nodes:
wherein ReLU is a linear rectification function and θ is an adjustable super-parameter.
Further, the step S5 includes:
s51, updating the functional area layer, updating the functional area representation, and preparing to transfer the information to the next level, and updating the functional area embedding by adopting a standard graph rolling network:
wherein GCN represents a graph rolling network,
representation of functional region characterization in t-round iterations, A Z Representing a functional area adjacency matrix->
Representing the functional region characterization in the t+1 round of iteration;
the functional area update information is then passed to the next layer to update the structural area characterization:
wherein ,
structural region characterization in iteration of t rounds, g ZR As gating unit function from functional region to structural region, +. RZ Representing allocation of structural zones to functional zonesMatrix (S)>
Representation of functional region characterization in the t+1 round of iteration, w ZR Is a learnable parameter;
sigmoid represents an activation function, the formula of which is:
wherein e represents a natural constant, and x represents the input of a sigmoid function;
s52, in the structural area, firstly updating the embedded representation of the layer by adopting a standard GCN:
then, the structure area is embedded and forwarded to the next layer of updated road section representation:
wherein ,wRS What learnable parameters are represented;
s53, combining the updated road segment representation with the information of the structural area and the functional area, and modeling the relationship between road segment nodes as a graph attention network
Repeating the operation to obtain the road network representation comprising the structural characteristics and the functional characteristics.
Compared with the prior art, the invention has at least the following technical effects:
the invention discloses a multi-graph hierarchical road map characterization method based on a graph neural network; the structure perception graph neural network is built, two virtual nodes, namely a structure area and a functional area, are introduced, the structure similarity and the functional similarity of road sections are built, the information sharing can be executed at a high level, and then updated information is transmitted to a low level; the road network characterization comprising the structural characteristics and the functional characteristics can be effectively obtained; the road network structure and the function roles are convenient to determine; the method is beneficial to revealing the functional areas of cities, route planning, arrival time estimation and position prediction, and construction of an intelligent traffic system.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and drawings.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a multi-graph hierarchical road surface characterization method based on a graph neural network.
Fig. 2 is a schematic structural diagram of a neural network model of a structural perception map provided by the invention.
FIG. 3 is a schematic diagram of the structural and functional regions generated by HRNR and HRNR+ according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, the embodiment of the invention discloses a multi-graph hierarchical road map characterization method based on a graph neural network, which comprises the following steps:
s1, giving a road section, and carrying out characteristic representation on the road section;
s2, establishing a relation between road sections, and constructing three adjacent matrixes among the road sections, wherein the three adjacent matrixes comprise a natural connectivity matrix, a structural similarity matrix and a functional similarity matrix;
s3, setting each road section to correspond to a single structural area, wherein different road sections have different importance degrees in one structural area, and obtaining a structural area distribution matrix A for representing important relation between the road sections and the structural areas through a structural perception graph neural network formed by spectral clustering and graph annotation force networks SR The method comprises the steps of carrying out a first treatment on the surface of the And associating the structural area with the road section;
s4, building a functional area based on the structural areas with the same functions, and building a functional area distribution matrix A RZ Based on the function area distribution matrix, representing the function area representation as a weighted linear combination of the structure area representation, and obtaining an adjacent matrix of the function area nodes;
s5, updating the representation of the functional area, and transmitting updated information through the sequence of the functional area, the structural area and the road section.
The following describes the above steps in detail:
the step S1 specifically comprises the following steps:
giving a road section, taking the road section serial number ID, the number LN of lanes, the road section length SL and the longitude and latitude LL of all the time into consideration, performing context embedding, dividing the whole value range into a plurality of contact intervals for the contact characteristics, performing feature coding by using interval numbers, setting a unique embedded connection for each value, connecting the associated vectors in series to be used as context embedding, and initializing graph node embedding:
wherein ,vID Representing the feature vector, v, corresponding to the road segment serial number ID LN Feature vector v representing lane number LN SL Representing the corresponding feature vector, v, of the link length SL LL Representing the feature vector corresponding to the latitude and longitude LL,
representing road segment s i The feature V of all nodes is the initial road segment feature N S ，N S ←V。
The step S2 specifically comprises the following steps:
s21, establishing an adjacency matrix of the road section based on the natural connectivity of the road network by using
Indicating that if one road section is upstream of another road section +.>
If the corresponding value in (a) is 1, otherwise, 0, for the bidirectional road sections, two directional connections are simply added by reversing the starting point and the ending point vertexes of the bidirectional road sections;
s22, designing a structural code according to the road types, and assuming that the road types share m types, S is the road segments i Generating an m-dimensional vector
At->
In which only the road section s is met i The category position is 1, the rest positions are 0, and each road section s is then divided into two sections i Structural code sum s i Structure of adjacent surrounding road sectionsSplicing the encoded sums, and performing normalization operation: />
wherein ,
representing road segment s i Structural coding features of->
Representing road segment s j Is indicative of a concatenation operation, N (s i ) Representation and road segment s i A set of adjacent road segments, |x| 2 Representing the modulus of the vector x to obtain a structural matrix
wherein kS For the number of road segments, each row in F represents the structural feature of one road segment in the road network, and the cosine similarity is used for measuring the structural similarity of two road segments:
SS＝F·F T
generating a mask matrix to filter some edges with smaller similarity:
wherein ,SSmask Is a mask matrix, SS mask [i,j]Representing an element located in the ith row and jth column, θ 1 Is a similarity threshold, the structural similarity is multiplied by a mask matrix and passed through an activation function to produce the final structural similarity matrix
wherein ,
the representation matrix is multiplied by elements, σ represents a sigmoid function, which is represented as
S23, generating a single thermal code for each POI according to the type of the POI, namely the point of interest, and assuming that the POIs share n classes, namely each POI p j Generating an n-dimensional vector
At->
Of them, only the POip is satisfied j The type of location is 1, the rest of locations are 0, consider the impact of each POI on its surrounding road segments:
wherein dis (p) j ,s i ) Representing POI p j To road section s i Distance, theta 2 Is a distance threshold, all POIs are for road segment s i The influence of (a) forms a road segment s i Is characterized by the following functions:
obtain a functional matrix
Each row in G represents the functional characteristics of one road section in the road network, and the cosine similarity is used for measuring the functional similarity of two road sections, and the two road sections are generated into a final result through an activation functionFunctional similarity matrix->
Referring to fig. 2, step S3 specifically includes:
s31, dividing structural areas by adopting a spectral clustering algorithm, and giving an adjacency matrix of road sections
By subtracting the diagonal matrix D S Deriving its graph Laplace matrix L S Calculating the previous d' eigenvectors u 1 ,…u d' Obtaining a matrix consisting of d' eigenvectors>
R represents a linear space, < >>
Represented by k S Linear space formed by x d' dimensional matrix, by running standard K-means algorithm on matrix U, a hard map M from location to structural area is obtained 1 ；
Adjacency matrix
Representing natural connectivity between road segments, < >>
For the structural similarity matrix of road segments, the structural similarity is encoded as a novel connection between road segments, two graphs are respectively modeled by adopting two graph annotation force networks, and the two graphs are aggregated together to obtain a soft map M 2 I.e.
Wherein GAT is a graph-annotation force network, softmax (·) is a line normalization function, and the hard map and the soft map are combined to obtain the final constructed structure region allocation matrix A SR ：
A SR ＝softmax(αM 1 +(1-α)M 2 )
Where alpha is a hyper-parameter that controls the fusion of the two mappings,
s32, through A SR Associating the structural zone representation with the road segment representation:
wherein ,NR The features of the structural region are represented,
and meanwhile, the connection relation of the structural area and the connection relation of the road section are associated:
wherein ,
representing the structural region connection matrix.
The step S4 specifically comprises the following steps:
s41, function adjacency matrix
Transfer from road segment to structure zone:
wherein ,
representing a structural region functional similarity matrix;
modeling the natural connectivity and the functional similarity of the structural areas by adopting two other graph-annotation force networks respectively, and aggregating the two to obtain a final functional area distribution matrix A RZ I.e.
S42, utilizing the function area allocation matrix A RZ The functional region representations are represented as weighted linear combinations of structural region representations:
by N Z Further deriving an adjacency matrix of the functional area nodes:
wherein ReLU is a linear rectification function and θ is an adjustable super-parameter.
The step S5 specifically comprises the following steps:
s51, updating the functional area layer, updating the functional area representation, and preparing to transfer the information to the next level, and updating the functional area embedding by adopting a standard graph rolling network:
wherein GCN represents a graph rolling network,
representation of functional region characterization in t-round iterations, A Z Representing a functional area adjacency matrix->
Representation of functional region characterization in the t+1 round of iteration
The functional area update information is then passed to the next layer to update the structural area characterization:
wherein ,
structural region characterization in iteration of t rounds, g ZR As gating unit function from functional region to structural region, +. RZ Representing the allocation matrix of structural areas to functional areas, < >>
Representing the functional region characterization in the t +1 round of iteration,
sigmoid represents an activation function, the formula of which is:
wherein ,wZR Is a learnable parameter.
S52, in the structural area, firstly updating the embedded representation of the layer by adopting a standard GCN:
then, the structure area is embedded and forwarded to the next layer of updated road section representation:
s53, combining the updated road segment representation with the information of the structural area and the functional area, and modeling the relationship between road segment nodes as a graph attention network
Repeating the steps to obtain the road network characterization comprising the structural characteristics and the functional characteristics.
After training is completed, a road section to structure area and structure area to function area distribution matrix A can be obtained SR and ARZ . The distribution matrix can be utilized to obtain the probability distribution of each road section belonging to each structural area and the probability distribution of each structural area belonging to each functional area, so that the road sections can be clustered, the road sections belonging to the same structural area are marked as the same color as the structural area, and further, the road sections belonging to the same functional area are marked as the same color as the functional area, thus the work of performing advanced semantic clustering on the urban road network can be completed, and the dividing method can be further dividedAnd (5) separating.
The method is verified in one specific example as follows:
referring to fig. 3, fig. 3 (a) and 3 (b) show the structural regions around the beijing left house bridge constructed by the existing optimal method (HRNR) and the present method (hrnr+). The left house bridge is the portal of the city center to the capital airport. It can be seen that in hrnr+, the bridge section is significantly separated from the residential section. Whereas in HRNR, the residential area and the bridge section are mixed together. From the structural similarity, the road types of bridge sections and residential areas are mostly highways and residential roads, respectively, and the difference facilitates the hrnr+ differentiation of different structural areas.
Likewise, fig. 3 (c) and 3 (d) are functional regions produced by HRNR and hrnr+, respectively. Three prominent business areas in Beijing, wangfujing, national trade and Sanrill, are marked by icons of shops. It can be seen that in hrnr+, the business areas are separated from surrounding residential areas, whereas in HRNR these are intermixed. The functional differences between business and residential areas contribute to a more rational division of functional areas. In addition, the three business areas belong to the same cluster. This may be interpreted as that with the functional similarity matrix, the regional clustering may break through the limitation of geographical connections. This is significant for downstream tasks, because similar functional areas may share certain properties, which makes it easier to distinguish different areas in a high-dimensional space.
From the description of the above embodiments, those skilled in the art will recognize that the core idea of the present application is to build a hierarchical neural network to model road map features of different levels. In particular, by means of fine-grained road segment aggregation, high-level traffic units are obtained, such as the aforementioned structural and functional clusters. For this purpose, two virtual nodes are introduced, namely a structural region and a functional region. The structural area is mainly used for representing the road sections with space communication and plays a certain role in traffic, such as overpasses, crossroads and the like. In addition, a functional area is formed above the structural area to provide a certain function for the travel population, such as a shopping area. In addition, a multi-graph mechanism is employed to direct virtual nodes to correspond to real-world structural and functional areas. In addition to the natural connectivity of the road network, structural similarity is established by using the road type attribute, and functional similarity between road segments is defined by using urban POI information. With such a three-tier organization, problems associated with remote node dependencies may be alleviated by first performing message sharing at a high level and then propagating updated information to lower level nodes. The multi-graph mechanism complements the functional attributes of the road network. In addition, the functional areas of the cities are revealed by combining with real track data, and the track data of the users can be used for finding potential functions or life style related modes; the method is beneficial to route planning, arrival time estimation and position prediction, construction of an intelligent traffic system and provision of certain functions for travel people, such as shopping areas.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
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 (6)
1. The multi-graph hierarchical road surface characterization method based on the graph neural network is characterized by comprising the following steps of:
s1, giving a road section, and carrying out characteristic representation on the road section;
s2, establishing a relation between road sections, and constructing a natural connectivity adjacency matrix, a structural similarity matrix and a functional similarity matrix between the road sections;
s3, setting each road section to correspond to a single structural area, wherein different road sections have different importance degrees in one structural area, and obtaining a structural area distribution matrix for representing important relations between the road sections and the structural areas through a structural perception graph neural network formed by spectral clustering and graph annotation force networks; and associating the structural area with the road section;
s4, constructing a functional area based on the structural area with the same function, constructing a functional area distribution matrix, and expressing the functional area representation as a weighted linear combination of the structural area representation based on the functional area distribution matrix to obtain an adjacent matrix of the functional area nodes;
s5, updating the representation of the functional area, and transmitting updated information through the sequence of the functional area, the structural area and the road section.
2. The method for multi-graph hierarchical road surface characterization based on the graph neural network according to claim 1, wherein the step S1 specifically comprises:
giving a road section, taking the road section serial number ID, the number LN of lanes, the road section length SL and the longitude and latitude LL of all the time into consideration, performing context embedding, dividing the whole value range into a plurality of contact intervals for the contact characteristics, performing feature coding by using interval numbers, setting a unique embedded connection for each value, connecting the associated vectors in series to be used as context embedding, and initializing graph node embedding:
wherein ,vID Representing the feature vector, v, corresponding to the road segment serial number ID LN Feature vector v representing lane number LN SL Representing the corresponding feature vector, v, of the link length SL LL Representing the feature vector corresponding to the latitude and longitude LL,
representing road segment s i Is the feature vector of (i) representing the vector concatenation operationThe characteristic V of all nodes is the initial road section characteristic N S 。
3. The method for multi-graph hierarchical road surface characterization based on the graph neural network according to claim 2, wherein the step S2 specifically comprises:
s21, establishing an adjacency matrix of the road section based on the natural connectivity of the road network by using
Indicating that if one road section is upstream of another road section +.>
If the corresponding value in (a) is 1, otherwise, 0, adding two directional connections for the bidirectional road sections by reversing the starting point and the ending point of the bidirectional road sections;
s22, designing a structural code according to the road types, and setting the road types to be m types, wherein the road types are S for each road section i Generating an m-dimensional vector
At->
In which only the road section s is met i The category position is 1, the rest positions are 0, and each road section s is then divided into two sections i Structural code sum s i Splicing the sum of the structural codes of adjacent surrounding road sections, and performing normalization operation:
wherein ,
representing road segment s i Structural coding features of->
Representing road segment s j Is indicative of a concatenation operation, N (s i ) Representation and road segment s i A set of adjacent road segments, |x| 2 Representing the modulus of the vector x to obtain a structural matrix
wherein kS For the number of road segments, each row in F represents the structural feature of one road segment in the road network, and the cosine similarity is used for measuring the structural similarity of two road segments:
SS＝F·F T
generating a mask matrix to filter some edges with smaller similarity:
wherein ,SSmask Is a mask matrix, SS mask [i,j]Representing an element located in the ith row and jth column, θ 1 Is a similarity threshold, the structural similarity is multiplied by a mask matrix and passed through an activation function to produce the final structural similarity matrix
wherein ,
the representation matrix is multiplied by elements, and sigma represents a sigmoid function;
s23, generating a single thermal code for each POI according to the type of the POI, wherein n types are shared by the POI, and p is the POI j Generating an n-dimensional vector
At->
In which only the POI p is met j The type of location is 1, the rest of locations are 0, consider the impact of each POI on its surrounding road segments:
wherein ,pj Representing points of interest, dis (p j ,s i ) Representing POI p j To road section s i Distance, theta 2 Is a distance threshold, all POIs are for road segment s i The influence of (a) forms a road segment s i Is characterized by the following functions:
obtaining a functional matrix
Each row in G represents a functional feature of a road segment in the road network, R represents linear space, k S N represents the dimension for the number of road segments; then the cosine similarity is used to measure the functional similarity of two road sections, and the two road sections are used to generate a final functional similarity matrix through an activation function>
Where σ represents the sigmoid function and the superscript T represents the transpose.
4. The method for multi-graph hierarchical road surface characterization based on the graph neural network according to claim 3, wherein the step S3 specifically comprises:
s31, dividing structural areas by adopting a spectral clustering algorithm, and giving an adjacency matrix of road sections
By subtracting the diagonal matrix D S Deriving its graph Laplace matrix L S Calculating the previous d' eigenvectors u 1 ,…u d' Obtaining a matrix consisting of d' eigenvectors>
R represents a linear space, < >>
Represented by k S Linear space formed by x d' dimensional matrix, by running standard K-means algorithm on matrix U, a hard map M from location to structural area is obtained 1 ；
Adjacency matrix
Representing natural connectivity between road segments, < >>
For the structural similarity matrix of road segments, the structural similarity is encoded as the connection between road segments, two graphs are respectively modeled by adopting two graph-meaning force networks, and the two graphs are aggregated together to obtain a soft map M 2 I.e. +.>
Wherein GAT is the graph annotation force network, softmax (·) is a line normalization function, N S Representing a road section representation; finally, combining the hard mapping and the soft mapping to obtain a final constructed structural area allocation matrix A SR ：
A SR ＝softmax(αM 1 +(1-α)M 2 )
Wherein α is a hyper-parameter controlling fusion of the two mappings;
s32, through A SR Associating the structural zone representation with the road segment representation:
wherein ,NR Representing structural region features;
and meanwhile, the connection relation of the structural area and the connection relation of the road section are associated:
wherein ,
representing the structural region connection matrix.
5. The method for multi-graph hierarchical road surface characterization based on the graph neural network according to claim 4, wherein the step S4 specifically comprises:
s41, function adjacency matrix
Transfer from road segment to structure zone:
wherein ,
representing a structural region functional similarity matrix;
modeling the natural connectivity and the functional similarity of the structural areas by adopting two graph-annotation force networks respectively, and aggregating the two graph-annotation force networks to obtain a final functional area distribution matrix A RZ I.e.
S42, utilizing the function area allocation matrix A RZ The functional region representations are represented as weighted linear combinations of structural region representations:
by N Z Further deriving an adjacency matrix of the functional area nodes:
wherein ReLU is a linear rectification function and θ is an adjustable super-parameter.
6. The method for multi-graph hierarchical road surface characterization based on the graph neural network according to claim 5, wherein the step S5 specifically comprises:
s51, updating the functional area layer, updating the functional area representation, and preparing to transfer the information to the next level, and updating the functional area embedding by adopting a standard graph rolling network:
wherein GCN represents a graph rolling network,
representation of functional region characterization in t-round iterations, A Z Representing a functional area adjacency matrix->
Representing the functional region characterization in the t+1 round of iteration;
the functional area update information is then passed to the next layer to update the structural area characterization:
wherein ,
structural region characterization in iteration of t rounds, g ZR As a gating cell function from functional region to structural region,as indicated by Hadamard product, i.e., the matrix elements are multiplied one by one, A RZ Representing the allocation matrix of structural areas to functional areas, < >>
Representation of functional region characterization in the t+1 round of iteration, w ZR Is a learnable parameter;
sigmoid represents an activation function, the formula of which is:
wherein e represents a natural constant, and x represents the input of a sigmoid function;
s52, in the structural area, firstly updating the embedded representation of the layer by adopting a standard GCN:
then, the structure area is embedded and forwarded to the next layer of updated road section representation:
wherein ,wRS What learnable parameters are represented;
s53, combining the updated road segment representation with the information of the structural area and the functional area, and modeling the relationship between road segment nodes as a graph attention network
Repeating the operation to obtain the road network representation comprising the structural characteristics and the functional characteristics.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310169056.9A CN116166975A (en) | 2023-02-27 | 2023-02-27 | Multi-graph hierarchical road surface characterization method based on graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310169056.9A CN116166975A (en) | 2023-02-27 | 2023-02-27 | Multi-graph hierarchical road surface characterization method based on graph neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116166975A true CN116166975A (en) | 2023-05-26 |
Family
ID=86421711
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310169056.9A Pending CN116166975A (en) | 2023-02-27 | 2023-02-27 | Multi-graph hierarchical road surface characterization method based on graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116166975A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117635218A (en) * | 2024-01-25 | 2024-03-01 | 北京交通大学 | Business district flow prediction method based on six-degree separation theory and graph annotation network |
-
2023
- 2023-02-27 CN CN202310169056.9A patent/CN116166975A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117635218A (en) * | 2024-01-25 | 2024-03-01 | 北京交通大学 | Business district flow prediction method based on six-degree separation theory and graph annotation network |
| CN117635218B (en) * | 2024-01-25 | 2024-04-09 | 北京交通大学 | Business district flow prediction method based on six-degree separation theory and graph annotation network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108470444B (en) | A kind of city area-traffic big data analysis System and method for based on genetic algorithm optimization | |
| WO2023029234A1 (en) | Method for bus arrival time prediction when lacking data | |
| Liu et al. | Contextual fuzzy cognitive map for decision support in geographic information systems | |
| CN113378891B (en) | Urban area relation visual analysis method based on track distribution representation | |
| Li et al. | Prediction of human activity intensity using the interactions in physical and social spaces through graph convolutional networks | |
| Yu et al. | Taxi-based mobility demand formulation and prediction using conditional generative adversarial network-driven learning approaches | |
| Iwanski et al. | Analyzing an offender's journey to crime: A criminal movement model (CriMM) | |
| CN116166975A (en) | Multi-graph hierarchical road surface characterization method based on graph neural network | |
| Chen et al. | Personalized travel route recommendation algorithm based on improved genetic algorithm | |
| Song et al. | Graph kernel based clustering algorithm in MANETs | |
| Huang et al. | Bayonet-corpus: a trajectory prediction method based on bayonet context and bidirectional GRU | |
| Li et al. | HMGCL: Heterogeneous multigraph contrastive learning for LBSN friend recommendation | |
| CN113793500B (en) | Method and system for constructing computable road network of urban traffic road | |
| CN104821086A (en) | Method for positioning low-efficient road section combination in large-scale traffic network | |
| Huang et al. | A time-dependent attention convolutional LSTM method for traffic flow prediction | |
| Ke et al. | AutoSTG+: An automatic framework to discover the optimal network for spatio-temporal graph prediction | |
| Pei et al. | Self-supervised spatiotemporal clustering of vehicle emissions with graph convolutional network | |
| Sun et al. | TZVis: Visual analysis of bicycle data for traffic zone division | |
| Jiang et al. | Partitioning of urban networks with polycentric congestion pattern for traffic management policies: Identifying protected networks | |
| Hou et al. | Urban region profiling with spatio-temporal graph neural networks | |
| CN116665448A (en) | Traffic speed real-time prediction method and system based on graph convolution network | |
| CN111783262A (en) | Road network representation learning method based on hierarchical graph neural network | |
| Ramadan et al. | Traffic forecasting using temporal line graph convolutional network: Case study | |
| Huang et al. | Multistep coupled graph convolution with temporal-attention for traffic flow prediction | |
| Liu et al. | Co-prediction of multimodal transportation demands with self-learned spatial dependence |
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 |