CN111489013A - Traffic station flow prediction method based on space-time multi-graph convolution network - Google Patents
Traffic station flow prediction method based on space-time multi-graph convolution network Download PDFInfo
- Publication number
- CN111489013A CN111489013A CN202010164538.1A CN202010164538A CN111489013A CN 111489013 A CN111489013 A CN 111489013A CN 202010164538 A CN202010164538 A CN 202010164538A CN 111489013 A CN111489013 A CN 111489013A
- Authority
- CN
- China
- Prior art keywords
- traffic
- time
- flow
- space
- 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
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000012360 testing method Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 20
- 238000013507 mapping Methods 0.000 claims abstract description 6
- 239000011159 matrix material Substances 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 230000004927 fusion Effects 0.000 claims description 6
- 230000000737 periodic effect Effects 0.000 claims description 6
- 239000000126 substance Substances 0.000 claims description 6
- 238000005065 mining Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 8
- 230000002354 daily effect Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 230000004913 activation Effects 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000000547 structure data Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000003442 weekly effect Effects 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- 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
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Abstract
The invention provides a traffic station flow prediction method based on space-time multi-graph convolution, which is used for solving the problems of low feature capture capability and low prediction precision of traffic station flow prediction in the prior art. The traffic station flow prediction method comprises the steps of firstly constructing a neighbor graph and a flow graph, respectively constructing a convolution component and capturing time-space characteristic output mapping of station flow to flow values with the same shape as a result to be predicted, and fusing the two components to obtain a space-time multi-graph convolution network model based on context gating; and then training and testing data are constructed according to the station flow in and out data to obtain a mature space-time multi-graph convolution network model, so that the station flow prediction is completed. The invention applies the multi-graph convolution to the deep mining of traffic station flow data, fully captures the space-time characteristics of traffic station flow from the space dimension and the time dimension, comprehensively considers various factors for predicting the traffic station flow entering and exiting, and improves the traffic station flow prediction precision.
Description
Technical Field
The invention belongs to the field of intelligent transportation, and particularly relates to a traffic station flow prediction method based on space-time multi-graph convolution.
Background
With the continuous development of cities, traffic is also more intelligent. Traffic flow prediction is an important component of intelligent transportation systems. In life, various traffic stations exist, such as urban subway stations, expressway toll stations, civil aviation airports and the like, the traffic jam condition of the stations is closely related to the normal operation of the whole traffic network, and the normal travel of passengers can be influenced. If the traffic flow of the traffic station can be effectively predicted, the normal operation of the whole traffic network is guaranteed, measures are taken in advance to avoid traffic jam, potential safety hazards are reduced, passengers can conveniently and reasonably arrange trips, trip routes are planned, and the development of an intelligent traffic system is assisted.
At present, three types of methods for predicting the traffic flow include a classic time series prediction method, a traditional machine learning method and a deep learning prediction method. Among the classical time series prediction methods, there are the Historical Average (HA), the Vector Autoregressive (VAR), the moving average autoregressive (ARIMA), and their variants. These methods mainly perform prediction by mining the rules in the time dimension from the time series of the traffic flow, and generally require that the time series have a certain periodicity or regularity, so the prediction effect is poor. The traditional machine learning method mainly comprises the steps of constructing an integrated model, for example, combining Empirical mode decomposition (Empirical mode decomposition) with a BP (back propagation) neural network to predict subway pedestrian flow, and for example, using a Wavelet support vector machine (Wavelet-SVM) to predict short-term flow of a subway, but the method only considers the time dimension attribute of a sequence, and does not consider the spatial correlation between traffic flow sequences. The deep learning prediction method overcomes the defects of the former two methods, and is an ideal traffic flow prediction method.
In the prior art, one of deep learning prediction methods calculates regional pedestrian volume by dividing an urban region into grids of equal size, and designs a model ST-ResNet based on a residual error unit for predicting the pedestrian volume of each region, thereby obtaining a good prediction effect. Although the method can effectively capture the space-time correlation among the flow rates of each region, the method can only process European-Margard structure data and is not suitable for non-Euclidean structure data. There is also a deep learning prediction method, which constructs a space-time map volume prediction model by capturing the space-time dependency of the vehicle speed of each section on the expressway through the use of map volume and gate control convolution, respectively, but neglects the periodicity of the flow. In addition, a space-time attention mechanism is adopted by modeling the relationship between the flow to be predicted and the recent flow, the daily periodic flow and the periodic flow, and is used for capturing the space-time correlation among the node traffic flows, but the model ignores the correlation among different periodic flows and has certain defects. In order to solve the limitation of the traditional convolution neural network on the graph data, a graph convolution technology is developed.
Disclosure of Invention
In order to improve the feature capture capability and the prediction accuracy of traffic station flow prediction, the embodiment of the invention provides a traffic station flow prediction method based on a space-time multi-graph convolution network.
The embodiment of the invention adopts the following technical scheme:
a traffic station flow prediction method based on space-time multi-graph convolution comprises the following steps:
step S1, judging whether the sites are adjacent in physical position according to whether a direct connection line exists between the sites, thereby constructing a neighbor map in the traffic network;
step S2, taking all the traffic network access records generated by the traffic network in a preset period, calculating the flow rate between stations in the period, and constructing a flow rate graph;
step S3, constructing a neighbor graph-space-time graph convolution component and a flow graph-space-time graph convolution component on the neighbor graph and the flow graph respectively, capturing the space-time characteristics of the site flow respectively, then accessing the full connection layer respectively behind the two components, and mapping the space-time characteristic output captured by the two components into a flow value with the same shape as the result to be predicted;
step S4, the rear part of the full connection layers of the two components is accessed into a fusion module, the outputs of the full connection layers of the two components are fused in the fusion module to be used as a prediction result of seed making, and thus, a complete space-time multi-graph convolution network model based on context gating is obtained;
step S5, taking traffic network station ingress and egress flow data in a preset time period, constructing a training set and a test set, inputting the training set into the context gate-controlled space-time multi-graph convolution network model, training the model, and testing by adopting the test set; after training and testing are completed, a mature space-time multi-graph convolution network model based on context gating is obtained;
and step S6, predicting traffic station flow by using a mature space-time multi-graph convolution network model based on context gating.
As a preferred solution of the embodiment of the present invention, the traffic network G is a function of a station V and a relationship thereof, G ═ V, (E, a), E represents an edge set in the traffic network, an edge represents a direct link between stations, V represents all stations therein, V ∈ V, | V | ═ N represents the number of stations in the traffic network, and a is an adjacency matrix;
neighbor graph GNIndicating the positional adjacency between sites, GN＝(V，EN，AN)，ENRepresenting a set of edges in a neighbour graph, the edges representing neighbour relations between sites, ANIs a contiguous matrix; and is
As a preferable solution of the embodiment of the present invention, the records of entering and exiting the transportation network in step S2 include each record generated by the individual p entering and exiting the network from the station v and a record group generated by the individual p entering and exiting the network once; wherein, the quadruple Tk(p, v, tau, kappa) represents a record generated by an individual p from a station v to the network, wherein p is an individual identifier, v is a station identifier, tau is a time identifier, kappa is an access identifier, kappa ∈ { in, out }, k ∈ N represents a record serial number, and a binary T isio(Tk，Tk+1) Denotes a set of records generated by an individual entering and exiting a network in a row, where Tk·κ＝in，Tk+1·κ＝out，Tk·p＝Tk+1·p， Tk·τ＜Tk+1·τ。
As a preferred embodiment of the present invention, the throughput graph GFExpressing the relationship of flow rate among sites, and the expression is GF＝(V，EF，AF) Wherein E isFRepresenting the edge set in the flow chart, the edge represents the relationship of flow intensity, AFIs an adjacent matrix; and:
as a preferred scheme of the embodiment of the present invention, the method includes the steps of taking all incoming and outgoing traffic network records generated by a traffic network in a preset period, calculating a flow rate between stations in the period, and constructing a flow rate graph based on a spatio-temporal multi-graph convolution neural network, further including:
taking a period of historical time interval tf＝[tstart，tend]Traffic network entry and exit records, using Δ tf＝tstart-tendRepresents a time interval tfLength of (d), calculating tfSize of traffic between sites, with Fij fRepresenting a site viAnd vjAt a time interval tfThe flow rate value of the inner stream is as follows:
wherein the content of the first and second substances,
is shown in the time interval tfInternal slave sites viEntering into or from the network vjThe value of the flow leaving the network,
is shown in the time interval tfInternal slave sites vjEntering into or from the network viA traffic value leaving the network;
if the edge larger than the flow average value is reserved, the flow chart is as follows:
wherein the content of the first and second substances,
is shown in the time interval tfAverage traffic flow between sites within.
As a preferable aspect of the embodiment of the present invention, the expression of the station ingress/egress flow rate, the inflow flow rate generated at the station v in the time interval t, is:
during the time interval t, the outflow flow generated at the station v has the expression:
the ingress and egress traffic of all stations in the time interval t is expressed as tensor:
Xt∈RN×2wherein
Wherein t is [ t ]start，tend) Denotes a certain time interval, Δ t ═ tstart-tendIndicates the interval length.
As a preferred solution of the embodiment of the present invention, the method includes the steps of obtaining traffic network station ingress and egress traffic data within a period of time, and constructing a training set and a test set of the neighbor graph and the traffic flow graph, and further:
determination of TpThe time interval to be predicted is determined,
represents TpA value of interval to be predicted, using XhRecent segments representing the object to be predicted, denoted by XdSegments of the daily cycle representing the object to be predicted, by XwA periodic segment representing the target to be predicted, Xh、XdAnd XwSplicing in a time dimension to obtain X which is used as a training set or a test set of the model;
wherein the time period to be predicted cannot be crossed with the time period for constructing the flow chart.
As a preferred solution of the embodiment of the present invention, in the step S3, each of the neighborhood graph-space-time graph convolution components and the traffic graph-space-time graph convolution component includes a plurality of context-gating-based space-time convolution units.
As a preferred solution of the embodiment of the present invention, the building of the neighbor graph-space-time graph convolution component and the traffic flow graph-space-time graph convolution component on two graph structures respectively includes the following steps:
step S301, building a graph convolution module based on context gating in two space-time graph convolution components respectively, wherein the two space-time graph convolution components have the same structure and are composed of a plurality of space-time graph convolution modules based on context gating in series;
step S302, performing graph convolution operation on each data frame in the time dimension in X based on a neighbor graph and a traffic flow graph respectively, and capturing the proximity correlation and the traffic flow dependency between site flows to enable the flow characteristic of each site to already contain characteristic information of adjacent sites on the corresponding graph;
step S303, capturing the space-time characteristics of the site flow by using two-dimensional convolution of a time dimension;
step S304, inputting the captured space-time characteristics of the site flow into a space-time map volume module with the same structure and based on the context, and further capturing the high-order space-time characteristics of the site flow;
and step S305, respectively mapping the output of the last context-gating-based space-time map convolution module in the two space-time map convolution components into a prediction result by using a full connection layer.
As a preferable scheme of the embodiment of the present invention, the step 4 further includes: are used separately
And
the prediction results obtained from the neighbor graph and the flow rate graph through the full connection layer are shown
And representing a final prediction result, fusing the prediction results on the two graphs by using a Hadamard product, wherein the expression is as follows:
wherein ⊙ denotes the Hadamard product, WNAnd WFIs the parameter tensor.
The invention has the following beneficial effects:
the traffic station flow prediction method based on the space-time multi-graph convolution disclosed by the embodiment of the invention applies the space-time multi-graph convolution to deep mining of traffic station flow data, constructs a neighbor graph-space-time graph convolution component and a traffic flow graph-space-time graph convolution component, captures space-time characteristics in a time period to be predicted through the two convolution components, and then fuses the captured space-time characteristics to obtain a traffic station flow prediction result. The embodiment of the invention starts from the space dimension and the time dimension, fully captures the space-time characteristics of the traffic station flow, comprehensively considers various factors for predicting the traffic station flow, improves the traffic station flow prediction precision, is favorable for passengers to carry out reasonable travel arrangement according to an accurate station flow prediction result, avoids traffic jam and reduces potential safety hazards.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a flow chart of a traffic station traffic prediction method based on a spatio-temporal multi-graph convolutional network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a two-dimensional convolution of the time dimension in an embodiment of the present invention;
FIG. 3 is a diagram illustrating multi-cycle segment splicing of an object to be predicted according to an embodiment of the present invention.
Detailed Description
The technical problems, aspects and advantages of the invention will be explained in detail below with reference to exemplary embodiments. The following exemplary embodiments are merely illustrative of the present invention and are not to be construed as limiting the invention. It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
Aiming at the intellectualization of traffic stations, the invention provides a traffic station flow prediction method based on space-time multi-graph convolution. The space-time multi-graph convolution neural network is one kind of deep learning neural network, and applies graph convolution to space-time data mining. There are two main types of graph convolution methods, the first is a space-based method, which rearranges nodes into a lattice form to satisfy the input of a conventional convolutional neural network; the second method is based on spectrogram theory, and the graph signal is convoluted in the spectrum domain. The method and the device fully capture the space-time characteristics of the traffic station flow from the space dimension and the time dimension, and solve the problems of incomplete consideration factors, insufficient characteristic capture, low prediction precision and the like in the prior art for predicting the traffic station flow entering and exiting.
The present invention will be described in further detail below with reference to specific embodiments thereof, with reference to the accompanying drawings.
The embodiment of the invention provides a traffic station flow prediction method based on space-time multi-graph convolution, and a flow chart of the traffic station flow prediction method based on the space-time multi-graph convolution is shown in figure 1. As shown in fig. 1, the traffic stop flow prediction method includes the following steps:
and step S1, judging whether the sites are adjacent in physical position according to whether a direct connection line exists between the sites, thereby constructing a neighbor map in the traffic network.
And step S2, taking all the in-and-out traffic network records generated by the traffic network in a preset period, calculating the flow rate between stations in the period, and constructing a flow rate graph.
Step S3, constructing a neighbor graph-space-time graph convolution component and a flow graph-space-time graph convolution component on the neighbor graph and the flow graph respectively, capturing the space-time characteristics of the site flow respectively, then accessing the full connection layer respectively behind the two components, and mapping the space-time characteristic output captured by each component to a flow value with the same shape as the result to be predicted;
step S4, the rear part of the full connection layers of the two components is accessed into a fusion module, the outputs of the full connection layers of the two components are fused in the fusion module to be used as a prediction result of seed making, and thus, a complete space-time multi-graph convolution network model based on context gating is obtained;
step S5, taking traffic network station ingress and egress flow data of a preset time period, constructing a training set and a test set, inputting the training set into a constructed space-time multi-graph convolution network model based on context gating, training the model, and testing by adopting the test set; after training and testing are completed, a mature space-time multi-graph convolution network model based on context gating is obtained;
and step S6, predicting traffic station flow by using a mature space-time multi-graph convolution network model based on context gating.
In step S1, the traffic network G is a function of the station V and its relationship, G ═ V, (E, a), E represents an edge set in the traffic network, an edge represents a direct link between stations, V represents all stations therein, V ∈ V, | V | N represents the number of stations in the traffic network, and a is an adjacency matrix.
Obtaining the adjacent relation between the stations according to the traffic network diagram, thereby constructing a neighbor graph GN，GN＝(V，EN，AN)，ENRepresenting a set of edges in a neighbor graph, the edges representing the adjacency between sites, ANIs a moment of abutmentThe array, expressed as:
in the step S2, the in-out traffic network record refers to each record generated by the individual p entering and exiting the network from the station v and a record group generated by the individual p continuously entering and exiting the network once; wherein, the quadruple Tk(p, v, τ, κ) representing a record generated by an individual p entering and exiting the network from the site v, where p is an individual identifier, v is a site identifier, τ is a time identifier, κ is an entry and exit identifier, κ ∈ { in, out }. k ∈ N represents a record sequence number, and if a certain individual enters and exits the network multiple times, there is a record chain T0→T1→…Tk→ …; binary Tio(Tk，Tk+1) Denotes a set of records generated by an individual entering and exiting a network in a row, where Tk·κ＝in，Tk+1·κ＝out， Tk·p＝Tk+1·p，Tk·τ＜Tk+1·τ。
The flow rate chart GFExpressing the relationship of flow quantity among stations, and the expression is GF＝(V，EF，AF) Wherein E isFRepresenting the edge set in the flow chart, the edge represents the relationship of flow intensity, AFIs a contiguous matrix. The traffic flow here includes traffic in all directions between stations.
The method comprises the following steps of taking all incoming and outgoing traffic network records generated by a traffic network in a preset period, calculating the flow rate between stations in the period, and constructing a flow rate graph based on a spatio-temporal multi-graph convolution neural network, and further comprising the following steps of:
taking a period of historical time interval tf＝[tstart，tend]Traffic network entry and exit records, using Δ tf＝tstart-tendRepresents a time interval tfLength of (d), calculating tfSize of traffic between sites, using
Representing a site viAnd vjAt a time interval tfThe flow rate value of the inner stream is as follows:
wherein the content of the first and second substances,
is shown in the time interval tfInternal slave sites viEntering into or from the network vjThe value of the flow leaving the network,
is shown in the time interval tfInternal slave sites vjEntering into or from the network viThe value of the traffic leaving the network.
In equations (2-1) to (2-3), in order to control the sparsity of the map and reduce the amount of calculation, only the side larger than the flow average value is reserved, and then the flow rate map is:
wherein the content of the first and second substances,
is shown in the time interval tfBetween domestic sitesAverage flow through.
In step S3, each of the neighbor graph-space-time graph convolution components and the traffic graph-space-time graph convolution components includes a plurality of context-gating-based space-time convolution units.
The method for constructing the neighbor graph-space-time graph convolution component and the traffic graph-space-time graph convolution component on two graph structures respectively comprises the following steps:
step S301, a graph convolution module based on context gating is respectively constructed in two space-time graph convolution components, the two space-time graph convolution components are identical in structure and are composed of a plurality of space-time graph convolution modules based on context gating in series.
The method specifically comprises the following steps:
step S3011, in each context-gated space-time diagram convolution, using context-gated gating unit
A plurality of time slices in (a) are modeled for importance. By using
To represent
For a data frame of XtPerforming Chebyshev diagram convolution operation
To obtain
K represents the size of the graph convolution kernel, at this time
Each node in (a) contains information of neighboring nodes, which will then
Is spliced to obtain
Namely, it is
In this way it is possible to obtain,
wherein each node characteristic information contains the context information thereof.
Step S3012 based on
Constructing a context gate control unit, and aggregating the information of each data frame by using a global averaging method to obtain a vector
Namely, it is
Step S3013, obtaining a gating vector by applying an attention mechanism based on the context information vector z
I.e. let s be σ (W)2(W1z)) in which W is1And W2Is a parameter matrix, sigma represents an activation function Re L U and represents an activation function sigmoid, and finally, X is scaled by a gating vector s to obtain the matrix
Wherein
Step S302, based on the neighbor graph and the flow chart pair
Each data frame in the medium time dimension is subjected to a graph convolution operation for capturing the adjacent correlation between site traffic and the traffic dependency, respectively.Re L U is used as an activation function of the graph convolution layer.
Step S303, capturing the space-time characteristics of the site flow by using two-dimensional convolution of a time dimension.
Fig. 2 is a schematic diagram of a two-dimensional convolution in the time dimension. As shown in fig. 2, the features of different time slices at each station are convolved along the time direction by using two-dimensional convolution, wherein the convolution kernel parameters are shared among the stations in order to reduce the parameters.
Step S304, the captured space-time characteristics of the site flow are input into a space-time map volume module based on the context with the same structure, and the high-order space-time characteristics of the site flow are further captured.
And step S305, respectively mapping the output of the last context-gating-based space-time map convolution module in the two space-time map convolution components into a prediction result by using a full connection layer. In order to reduce parameters and avoid overfitting, all sites share the same full-link layer in the same time-space diagram convolution component.
The step S4 further includes: are used separately
And
the prediction results obtained from the neighbor graph and the flow rate graph through the full connection layer are shown
To depict the final prediction results
And
for different influence degrees of the predicted target, the prediction results on the two graphs are fused by using the Hadamard product, namely
Wherein ⊙ denotes the Hadamard product, WNAnd WFIs the parameter tensor.
In step S5, the expression of the station ingress/egress flow rate, the inflow flow rate occurring at the station v in the time interval t, is:
xt in，v＝|{Tk(p，v，τ，κ)|Tk·κ＝in∧Tk·v＝v∧Tk·τ∈t}|； (4-1)
during the time interval t, the outflow flow generated at the station v has the expression:
xt out，v＝|{Tk(p，v，τ，κ)|Tk·κ＝out∧Tk·v＝v∧Tk·τ∈t}| (4-2)
wherein t is [ t ]start，tend) Denotes a certain time interval, Δ t ═ tstart-tendIndicates the interval length.
The incoming and outgoing flows of all stations in the time interval t are expressed as tensor:
Xt∈RN×2wherein
The traffic network station ingress and egress flow data in a period of time is taken, a training set and a test set of the neighbor graph and the traffic flow graph are constructed, and the method further comprises the following steps:
determination of TpThe time interval to be predicted is determined,
represents TpA value of interval to be predicted, using XhRecent segments representing the object to be predicted, denoted by XdSegments of the daily cycle representing the object to be predicted, by XwIndicating the target cycle to be predictedFragment of A, Xh、XdAnd XwAnd splicing in the time dimension to be used as the input of the model training set. FIG. 3 is a diagram illustrating multi-cycle segment splicing of an object to be predicted. As shown in fig. 3, the week cycle segment, the day cycle segment, and the recent segment of the target to be predicted are spliced in the time dimension according to the time sequence.
The same method is used to construct the test set.
The process of constructing a data set is described in further detail below, taking a test set as an example.
Assuming that each time interval takes M minutes, 24 × 60/M time interval values are totally obtained every day, and if q is 24 × 60/M, the recent segment is obtained
Wherein T isrIs TpInteger multiple of the number of the last segment sequence. Segments of the daily cycle
Wherein T isdIs TpMultiple of (c), controlling the number of daily cycle segments. Periodic segment
Wherein T iswIs TpMultiple of (c), controlling the number of weekly cycle segments. The input of the whole model is X ═ Xw，Xd，Xh]。
And when a training set is constructed, the time period to be predicted cannot be crossed with the time period for constructing the flow chart.
According to the technical scheme, the traffic station traffic flow prediction method based on the space-time multi-graph convolution applies the multi-graph convolution to deep mining of traffic station flow data, a neighbor graph-space-time graph convolution component and a flow graph-space-time graph convolution component are constructed, space-time characteristics in a time period to be predicted are captured through the two convolution components, and the captured space-time characteristics are fused to obtain a traffic station traffic prediction result. The embodiment of the invention starts from the space dimension and the time dimension, fully captures the space-time characteristics of the traffic station flow, comprehensively considers various factors for predicting the traffic station flow, and improves the traffic station flow prediction precision.
While the foregoing is directed to the preferred embodiment of the present invention, it is understood that the invention is not limited to the exemplary embodiments disclosed, but is made merely for the purpose of providing those skilled in the relevant art with a comprehensive understanding of the specific details of the invention. It will be apparent to those skilled in the art that various modifications and adaptations can be made within the scope of the present invention without departing from the principles of the invention.
Claims (10)
1. A traffic station flow prediction method based on space-time multi-graph convolution is characterized by comprising the following steps:
step S1, judging whether the sites are adjacent in physical position according to whether a direct connection line exists between the sites, thereby constructing a neighbor map in the traffic network;
step S2, taking all the traffic network access records generated by the traffic network in a preset period, calculating the flow rate between stations in the period, and constructing a flow rate graph;
step S3, constructing a neighbor graph-space-time graph convolution component and a flow graph-space-time graph convolution component on the neighbor graph and the flow graph respectively, capturing the space-time characteristics of the site flow, then respectively accessing a full connection layer behind the two components, and mapping the space-time characteristic output captured by the two components into a flow value with the same shape as the result to be predicted;
step S4, the rear part of the full connection layers of the two components is accessed into a fusion module, the outputs of the full connection layers of the two components are fused in the fusion module to be used as a prediction result of seed making, and thus, a complete space-time multi-graph convolution network model based on context gating is obtained;
step S5, taking traffic network station traffic flow data in a period of time, constructing a training set and a test set, inputting the training set into a constructed space-time multi-graph convolution network model based on context gating, training the model, and testing by adopting the test set; after training and testing are completed, a mature space-time multi-graph convolution network model based on context gating is obtained;
and step S6, predicting traffic station flow by using a mature space-time multi-graph convolution network model based on context gating.
2. The traffic stop traffic prediction method according to claim 1, wherein the traffic network G is a function of a stop V and a relationship thereof, G ═ V (E, a), E represents an edge set in the traffic network, edges in the edge set represent lines directly connected between stops, V represents all stops therein, V ∈ V, | V | ═ N represents the number of stops in the traffic network, and a is an adjacency matrix;
neighbor graph GNIndicating the positional adjacency between sites, GN＝(V，EN，AN)，ENRepresenting a set of edges in a neighbor graph, the edges representing the adjacency between sites, ANIs a contiguous matrix; and is
3. The traffic site traffic prediction method of claim 2, wherein the traffic site entry and exit records of step S2 include each record generated by an individual p entering and exiting the network from the site v and a set of records generated by an individual p entering and exiting the network once in a row; wherein, the quadruple Tk(p, v, tau, kappa) represents a record generated by the individual p from the station v to the network, wherein p is the individual identifier, v is the station identifier, tau is the time identifier, kappa is the access identifier, kappa ∈ { in, out }, k ∈ N represents the record serial number, and the binary T isio(Tk，Tk+1) Denotes a set of records generated by an individual entering and exiting a network in a row, where Tk·κ＝in，Tk+1·κ＝out，Tk·p＝Tk+1·p，Tk.τ＜Tk+1·τ。
4. The traffic stop flow prediction method according to claim 3, wherein the flow chart G is a flow chart of the traffic flowFExpressing the relationship of flow quantity among stations, and the expression is GF＝(V，EF，AF) Wherein E isFRepresenting the edge set in the flow chart, the edge represents the relationship of flow intensity, AFIs a contiguous matrix; and:
5. the traffic station flow prediction method according to claim 4, wherein the method comprises the steps of taking all incoming and outgoing traffic network records generated by a traffic network in a preset period, calculating the flow rate of the traffic between stations in the period, and constructing a flow rate graph based on a spatiotemporal manifold convolution neural network, and further comprises the following steps:
taking a period of historical time interval tf＝[tstart，tend]Traffic network entry and exit records, using Δ tf＝tstart-tendRepresents a time interval tfLength of (d), calculating tfSize of traffic between sites, using
Representing a site viAnd vjAt a time interval tfThe flow rate value of the internal flow rate is as follows:
wherein the content of the first and second substances,
is shown in the time interval tfInternal slave sites viEntering into or from the network vjThe value of the flow leaving the network,
is shown in the time interval tfInternal slave sites vjEntering into or from the network viA traffic value leaving the network;
if the edge larger than the flow average value is reserved, the flow chart is as follows:
wherein the content of the first and second substances,
is shown in the time interval tfAverage traffic flow between sites within.
6. The traffic stop flow prediction method according to claim 5, wherein the stop incoming and outgoing flow, the time interval t, and the incoming flow expression generated at the stop v are:
xt in，v＝|{Tk(p，v，τ，κ)|Tk.κ＝in∧Tk.v＝v∧Tk.τ∈t}|； (4-1)
during the time interval t, the outflow flow generated at the station v has the expression:
xt out，v＝|{Tk(p，v，τ，κ)|Tk.κ＝out∧Tk.v＝v∧Tk.τ∈t}| (4-2)
the ingress and egress traffic of all stations in the time interval t is expressed as tensor:
Xt∈RN×2wherein (X)t)v，0＝xt in，v，(Xt)v，1＝xt out，v(5)
Wherein t is [ t ]start，tend) Denotes a certain time interval, Δ t ═ tstart-tendIndicates the interval length.
7. The traffic stop traffic prediction method of claim 6, wherein the method of taking traffic network stop ingress and egress traffic data over a period of time to construct a training set and a testing set of the neighborhood graph and the traffic flow graph further comprises:
determination of TpThe time interval to be predicted is determined,
represents TpA value of interval to be predicted, using XhRecent segments representing the object to be predicted, denoted by XdSegments of the daily cycle representing the object to be predicted, by XwA periodic segment representing the object to be predicted, Xh、XdAnd XwSplicing in a time dimension to be used as a training set or a test set of the model;
wherein the time period to be predicted cannot be crossed with the time period for constructing the flow chart.
8. The traffic stop traffic flow prediction method according to claim 7, wherein in step S4, each spatiotemporal graph convolution component includes a plurality of context-gating-based spatiotemporal convolution units.
9. The traffic stop flow prediction method according to claim 8, wherein the constructing a neighbor graph-space-time graph convolution component and a traffic flow graph-space-time graph convolution component comprises the steps of:
step S301, graph convolution modules based on context gating are respectively constructed in two space-time graph convolution components, the two space-time graph convolution components are identical in structure and are composed of a plurality of space-time graph convolution modules based on context gating in series;
step S302, based on the neighbor graph and the flow chart pair
Performing graph convolution operation on each data frame in the medium time dimension, capturing the adjacent correlation and the traffic dependency between the site traffic, and enabling the traffic characteristics of each site to already contain the characteristic information of the adjacent sites on the corresponding graph;
step S303, capturing the space-time characteristics of the site flow by using two-dimensional convolution of a time dimension;
step S304, inputting the captured space-time characteristics of the site flow into a context-based space-time graph convolution module with the same structure, and further capturing high-order space-time characteristics of the site flow;
step S305, respectively using the full connection layer to map the output of the last context-gating-based space-time graph convolution module in the two space-time graph convolution components as a prediction result.
10. The traffic point traffic prediction method according to claim 9, wherein the step 4 is further: are used separately
And
representing traffic in a neighbor graphPrediction results obtained through the full link layer on the graph, using
And representing a final prediction result, fusing the prediction results on the two graphs by using a Hadamard product, wherein the expression is as follows:
wherein ⊙ denotes the Hadamard product, WNAnd WFIs the parameter tensor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010164538.1A CN111489013A (en) | 2020-03-11 | 2020-03-11 | Traffic station flow prediction method based on space-time multi-graph convolution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010164538.1A CN111489013A (en) | 2020-03-11 | 2020-03-11 | Traffic station flow prediction method based on space-time multi-graph convolution network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111489013A true CN111489013A (en) | 2020-08-04 |
Family
ID=71794292
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010164538.1A Pending CN111489013A (en) | 2020-03-11 | 2020-03-11 | Traffic station flow prediction method based on space-time multi-graph convolution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111489013A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111815956A (en) * | 2020-09-11 | 2020-10-23 | 浙江高速信息工程技术有限公司 | Expressway traffic flow prediction method |
| CN111913946A (en) * | 2020-08-11 | 2020-11-10 | 上海海事大学 | Public bicycle taking and returning demand quantity prediction system, method and device |
| CN111968375A (en) * | 2020-08-27 | 2020-11-20 | 北京嘀嘀无限科技发展有限公司 | Traffic flow prediction method and device, readable storage medium and electronic equipment |
| CN112257920A (en) * | 2020-10-20 | 2021-01-22 | 北京工商大学 | Space-time data prediction method capable of directly capturing space-time correlation |
| CN112365708A (en) * | 2020-09-29 | 2021-02-12 | 西北大学 | Scenic spot traffic volume prediction model establishing and predicting method based on multi-graph convolution network |
| CN112488185A (en) * | 2020-11-27 | 2021-03-12 | 湖南大学 | Method, system, electronic device and readable storage medium for predicting vehicle operating parameters including spatiotemporal characteristics |
| CN112990555A (en) * | 2021-02-24 | 2021-06-18 | 上海交通大学 | Subway pedestrian flow prediction method and system based on space-time parallel grid neural network |
| CN114117064A (en) * | 2021-11-09 | 2022-03-01 | 西南交通大学 | Knowledge dynamic evolution method based on multi-time granularity and application |
| CN114785703A (en) * | 2022-03-09 | 2022-07-22 | 桂林航天工业学院 | Internet of things safety detection method and system based on graph convolution |
| CN114997543A (en) * | 2022-08-03 | 2022-09-02 | 通号通信信息集团有限公司 | People flow prediction method, model training method, electronic device and readable medium |
| CN115081717A (en) * | 2022-06-27 | 2022-09-20 | 北京建筑大学 | Rail transit passenger flow prediction method integrating attention mechanism and graph neural network |
| CN115565369A (en) * | 2022-09-20 | 2023-01-03 | 北京工商大学 | Hypergraph-based time-space hypergraph convolution traffic flow prediction method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110264709A (en) * | 2019-05-06 | 2019-09-20 | 北京交通大学 | The prediction technique of the magnitude of traffic flow of road based on figure convolutional network |
| CN110674987A (en) * | 2019-09-23 | 2020-01-10 | 北京顺智信科技有限公司 | Traffic flow prediction system and method and model training method |
| WO2020010717A1 (en) * | 2018-07-13 | 2020-01-16 | 南京理工大学 | Short-term traffic flow prediction method based on spatio-temporal correlation |
-
2020
- 2020-03-11 CN CN202010164538.1A patent/CN111489013A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020010717A1 (en) * | 2018-07-13 | 2020-01-16 | 南京理工大学 | Short-term traffic flow prediction method based on spatio-temporal correlation |
| CN110264709A (en) * | 2019-05-06 | 2019-09-20 | 北京交通大学 | The prediction technique of the magnitude of traffic flow of road based on figure convolutional network |
| CN110674987A (en) * | 2019-09-23 | 2020-01-10 | 北京顺智信科技有限公司 | Traffic flow prediction system and method and model training method |
Non-Patent Citations (1)
| Title |
|---|
| 荣斌 等: ""基于时空多图卷积网络的交通站点流量预测"", 《HTTP://KNS.CNKI.NET/KCMS/DETAIL/31.1289.TP.20191205.1708.007.HTML》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111913946A (en) * | 2020-08-11 | 2020-11-10 | 上海海事大学 | Public bicycle taking and returning demand quantity prediction system, method and device |
| CN111968375A (en) * | 2020-08-27 | 2020-11-20 | 北京嘀嘀无限科技发展有限公司 | Traffic flow prediction method and device, readable storage medium and electronic equipment |
| CN111815956A (en) * | 2020-09-11 | 2020-10-23 | 浙江高速信息工程技术有限公司 | Expressway traffic flow prediction method |
| CN112365708B (en) * | 2020-09-29 | 2022-04-19 | 西北大学 | Scenic spot traffic volume prediction model establishing and predicting method based on multi-graph convolution network |
| CN112365708A (en) * | 2020-09-29 | 2021-02-12 | 西北大学 | Scenic spot traffic volume prediction model establishing and predicting method based on multi-graph convolution network |
| CN112257920A (en) * | 2020-10-20 | 2021-01-22 | 北京工商大学 | Space-time data prediction method capable of directly capturing space-time correlation |
| CN112488185A (en) * | 2020-11-27 | 2021-03-12 | 湖南大学 | Method, system, electronic device and readable storage medium for predicting vehicle operating parameters including spatiotemporal characteristics |
| CN112488185B (en) * | 2020-11-27 | 2024-04-26 | 湖南大学 | Method and system for predicting vehicle operating parameters including spatiotemporal characteristics |
| CN112990555A (en) * | 2021-02-24 | 2021-06-18 | 上海交通大学 | Subway pedestrian flow prediction method and system based on space-time parallel grid neural network |
| CN114117064B (en) * | 2021-11-09 | 2023-05-26 | 西南交通大学 | Urban subway flow prediction method based on knowledge dynamic evolution of multi-time granularity |
| CN114117064A (en) * | 2021-11-09 | 2022-03-01 | 西南交通大学 | Knowledge dynamic evolution method based on multi-time granularity and application |
| CN114785703A (en) * | 2022-03-09 | 2022-07-22 | 桂林航天工业学院 | Internet of things safety detection method and system based on graph convolution |
| CN114785703B (en) * | 2022-03-09 | 2022-12-13 | 桂林航天工业学院 | Internet of things safety detection method and system based on graph convolution |
| CN115081717A (en) * | 2022-06-27 | 2022-09-20 | 北京建筑大学 | Rail transit passenger flow prediction method integrating attention mechanism and graph neural network |
| CN114997543A (en) * | 2022-08-03 | 2022-09-02 | 通号通信信息集团有限公司 | People flow prediction method, model training method, electronic device and readable medium |
| CN114997543B (en) * | 2022-08-03 | 2023-01-13 | 通号通信信息集团有限公司 | People flow prediction method, electronic device and readable medium |
| WO2024027362A1 (en) * | 2022-08-03 | 2024-02-08 | 通号通信信息集团有限公司 | Passenger flow prediction method, model training method, electronic device and readable medium |
| CN115565369B (en) * | 2022-09-20 | 2024-04-16 | 北京工商大学 | Space-time hypergraph convolution traffic flow prediction method and system based on hypergraph |
| CN115565369A (en) * | 2022-09-20 | 2023-01-03 | 北京工商大学 | Hypergraph-based time-space hypergraph convolution traffic flow prediction method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111489013A (en) | Traffic station flow prediction method based on space-time multi-graph convolution network | |
| CN112216108B (en) | Traffic prediction method based on attribute-enhanced space-time graph convolution model | |
| CN109887282B (en) | Road network traffic flow prediction method based on hierarchical timing diagram convolutional network | |
| CN110633871B (en) | Regional traffic demand prediction method based on convolution long-term and short-term memory network | |
| CN108197739B (en) | Urban rail transit passenger flow prediction method | |
| CN110866649A (en) | Method and system for predicting short-term subway passenger flow and electronic equipment | |
| CN112257934A (en) | Urban people flow prediction method based on space-time dynamic neural network | |
| CN113326974B (en) | Multi-source traffic flow prediction method based on multi-task hypergraph convolutional network | |
| CN112466117A (en) | Road network short-term traffic flow prediction method based on deep space-time residual error network | |
| CN113808396B (en) | Traffic speed prediction method and system based on traffic flow data fusion | |
| CN111047078B (en) | Traffic characteristic prediction method, system and storage medium | |
| CN115512545B (en) | Traffic speed prediction method based on space-time dynamic graph convolution network | |
| CN113283581B (en) | Multi-fusion graph network collaborative multi-channel attention model and application method thereof | |
| CN113112791A (en) | Traffic flow prediction method based on sliding window long-and-short term memory network | |
| CN114529081A (en) | Space-time combined traffic flow prediction method and device | |
| Ni et al. | STGMN: A gated multi-graph convolutional network framework for traffic flow prediction | |
| CN113112793A (en) | Traffic flow prediction method based on dynamic space-time correlation | |
| Dai et al. | Spatio-temporal deep learning framework for traffic speed forecasting in IoT | |
| Wei et al. | Dual graph for traffic forecasting | |
| CN116311939A (en) | OD demand prediction method and system based on dynamic space-time correlation | |
| CN115456314B (en) | Atmospheric pollutant space-time distribution prediction system and method | |
| CN114566048B (en) | Traffic control method based on multi-view self-adaptive space-time diagram network | |
| CN116258253A (en) | Vehicle OD prediction method based on Bayesian neural network | |
| CN114169632A (en) | Passenger flow distribution prediction method and device based on multi-view passenger flow change trend modeling | |
| Iram et al. | Controlling Smart-City Traffic using Machine Learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200804 |