CN115456238A - Urban trip demand prediction method based on dynamic multi-view coupling graph convolution - Google Patents

Urban trip demand prediction method based on dynamic multi-view coupling graph convolution Download PDF

Info

Publication number
CN115456238A
CN115456238A CN202210916072.5A CN202210916072A CN115456238A CN 115456238 A CN115456238 A CN 115456238A CN 202210916072 A CN202210916072 A CN 202210916072A CN 115456238 A CN115456238 A CN 115456238A
Authority
CN
China
Prior art keywords
graph
nodes
similarity
area
road
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210916072.5A
Other languages
Chinese (zh)
Inventor
刘志
卞纪新
张德举
陈洋
孔祥杰
沈国江
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202210916072.5A priority Critical patent/CN115456238A/en
Publication of CN115456238A publication Critical patent/CN115456238A/en
Pending legal-status Critical Current

Links

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
    • G06N3/08—Learning methods
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06315—Needs-based resource requirements planning or analysis

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Physics & Mathematics (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • General Business, Economics & Management (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Educational Administration (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a city travel demand prediction method based on convolution of a dynamic multi-view coupling graph, which comprises the steps of obtaining geographic position data, POI data and road data of each area in a city, and constructing a geographic similar graph, a functional similar graph and a road similar graph by taking each area as a node; taking unit time as an interval, acquiring travel demand data of each area in a city within each unit time, and constructing a dynamic demand similarity graph; and then, by combining a dynamic multi-view and coupling graph convolution module, modeling can be performed on complex space-time relations in the travel demand data from multiple angles, extracting time-related features from the fusion space features corresponding to each unit time by adopting a gate control cycle unit, extracting attention scores between the time-related features and external features, and performing weighted summation on the time-related features corresponding to all unit times by taking the attention scores as weights to obtain a final prediction result. The method and the device improve the accuracy of urban travel demand prediction.

Description

Urban trip demand prediction method based on convolution of dynamic multi-view coupling graph
Technical Field
The application belongs to the technical field of data prediction, and particularly relates to a city travel demand prediction method based on dynamic multi-view coupling graph convolution.
Background
With the rapid development of the intelligent traffic system, vehicles such as taxis, network taxi appointments, buses and subways become main tools for people to go out daily, and the problems of traffic jam prediction, traffic accident prediction, abnormal taxi path detection and the like become important challenges for the construction of intelligent cities. How to improve the utilization rate of the vehicle and reduce the waiting time of passengers is a very urgent challenge. The accurate urban travel demand prediction can reduce waiting time of passengers, improve travel efficiency, help vehicle operators to reasonably pre-schedule vehicles and help traffic departments to relieve traffic jam.
Urban travel demand prediction is an important problem in an intelligent transportation system, and has great influence on traffic management, urban planning and the like. For this reason, a large number of traffic demand prediction methods have been proposed, which mainly focus on how to effectively extract temporal and spatial correlations. Early traffic demand predictions mostly used machine learning and statistical analysis methods such as autoregressive moving average ARIMA model and its variants, least squares support vector machine (LS-SVM), K-nearest neighbor, etc., which mainly studied the temporal sequence variation while ignoring the effect of spatial correlation between different regions.
Some recent deep learning efforts have shown excellent performance in processing complex spatiotemporal data, and many scholars combine Convolutional Neural Networks (CNNs) with other networks such as (recurrent neural networks RNNs, long-short term memory neural networks LSTM, gated cyclic units GRU) to capture spatiotemporal correlations. But most of them model complex traffic networks as static graphs, neglecting the importance of building a map of the area dynamically over the entire time axis.
Disclosure of Invention
The urban travel demand prediction method based on the convolution of the dynamic multi-view coupling graph overcomes the defects of the prior art and improves the accuracy of urban travel demand prediction.
In order to achieve the purpose, the technical scheme of the application is as follows:
a city travel demand prediction method based on dynamic multi-view coupling graph convolution comprises the following steps:
acquiring geographic position data, POI data and road data of each area in a city, and constructing a geographic similar graph, a functional similar graph and a road similar graph by taking each area as a node;
taking unit time as an interval, acquiring travel demand data of each region in a city within each unit time, obtaining a characteristic matrix corresponding to each unit time, and constructing a dynamic demand similarity graph;
inputting the characteristic matrix and the adjacent matrixes of the dynamic demand similar graph, the geographical similar graph, the functional similar graph and the road similar graph into corresponding coupling graph convolution modules to obtain the spatial characteristics corresponding to the similar graphs, and fusing the spatial characteristics corresponding to the similar graphs to obtain fused spatial characteristics;
and respectively extracting time correlation characteristics from the fusion space characteristics corresponding to each unit time by adopting a gate control cycle unit, extracting attention scores between the time correlation characteristics and external characteristics through an attention layer, taking the attention scores as weights, and performing weighted summation on the time correlation characteristics corresponding to all the unit times to obtain a final prediction result.
Further, the geographical similarity is represented as
Figure BDA0003775604880000021
Where V represents the set of nodes in the graph, E g Representing sets of edges between nodes, A g Representation diagram
Figure BDA0003775604880000022
Of a contiguous matrix of g Each element in the group represents whether two nodes are adjacent or not;
A g any one of the elements A g (i, j) is expressed as:
Figure BDA0003775604880000023
the functional similarity is shown as
Figure BDA0003775604880000024
Where V represents the set of nodes in the graph, E p Representing a set of edges between nodes, A p Representation diagram
Figure BDA0003775604880000025
Of a neighboring matrix of p Each element in the set represents POI distribution similarity between two nodes;
A p any one of the elements A p (i, j) is expressed as:
A p (i,j)＝1-JS(P i ||P j )；
Figure BDA0003775604880000026
wherein, P i ,
Figure BDA0003775604880000027
POI distribution representing region i and region j, K represents number of POI types, P i (k),P j (k) Respectively representing the number of POI with the type k in the area i and the area j;
the road similarity is shown as
Figure BDA0003775604880000028
Where V represents the set of nodes in the graph, E r Representing sets of edges between nodes, A r Representation diagram
Figure BDA0003775604880000029
Of a contiguous matrix of r Each element in the set represents the similarity of road characteristics between two nodes;
A r any one of the elements A r (i, j) is expressed as:
A r (i,j)＝1-JS(R i ||R j )；
Figure BDA0003775604880000031
wherein R is i ,
Figure BDA0003775604880000032
Indicating the road distribution of area i and area j, L indicating the number of road features, R i (l),R j (l) Respectively representing the number of roads with the characteristic l in the area i and the area j;
the dynamic demand similarity is represented as
Figure BDA0003775604880000033
Where V represents the set of nodes in the graph, E d Representing a set of edges between nodes, A d Representation diagram
Figure BDA0003775604880000034
Of a contiguous matrix of d Each element in the tree represents the similarity of travel demands between two nodes;
A d any one of the elements A d (i, j) is expressed as:
A d (i,j)＝1-JS(X i ||X j )；
Figure BDA0003775604880000035
wherein X i (m),X j (M) denotes the mth features of the region i and the region j, respectively, and M denotes the number of required features.
Further, the feature matrix includes departure requirements and arrival requirements of each node in unit time.
Further, the extracting of the attention score between the time-related feature and the external feature through the attention layer is formulated as follows:
α＝soft max(Re L U(UW U +EW E +b α ))；
wherein U represents time-related features corresponding to all unit times, E represents external features, and W U 、W E And b α Denotes a weight parameter, and α denotes an attention score vector.
Further, the urban trip demand prediction method based on the convolution of the dynamic multi-view coupling diagram further includes:
and converting the prediction result into a grid form through a mapping matrix of the areas and the urban grid to obtain the prediction result of each area.
According to the urban travel demand prediction method based on the dynamic multi-view coupling graph convolution, the dynamic multi-view coupling graph convolution module and the coupling graph convolution module are combined, the complex space-time relation in travel demand data can be modeled from multiple angles, and the accuracy of urban travel demand prediction can be improved.
Drawings
FIG. 1 is a flow chart of a city travel demand prediction method based on dynamic multi-view coupling diagram convolution according to the application;
FIG. 2 is a schematic diagram of a city demand prediction model structure according to the present application;
fig. 3 is a schematic diagram of a multi-view coupling diagram convolutional network structure according to the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of and not restrictive on the broad application.
According to the urban trip demand prediction method based on the convolution of the dynamic multi-view coupling graph, in order to capture deep spatiotemporal correlation, a demand similarity graph is dynamically constructed at each moment, and the weight of an edge is obtained from demand data. Spatio-temporal correlations are then captured in conjunction with the GRU by fusing predefined geographical, functional and road similarity maps using graph convolution. And finally, combining external environmental factors to obtain the output of urban travel demand prediction through the attention layer and the mapping matrix.
In one embodiment, as shown in fig. 1, a city travel demand prediction method based on convolution of dynamic multi-view coupling graph includes:
s1, acquiring geographic position data, POI data and road data of each area in a city, and constructing a geographic similar graph, a functional similar graph and a road similar graph by taking each area as a node.
In the embodiment, the city is divided into a plurality of areas according to the grid, the area with rare data is eliminated, and other areas after data cleaning are selected for research. The method is based on the electronic map of the city, and the geographical position data, the POI data and the road data of each area are easily acquired, so that a geographical similar graph, a functional similar graph and a road similar graph are constructed.
The geographical similarity is shown as
Figure BDA0003775604880000041
Indicating the geographic relevance of the areas, i.e. whether they are geographically adjacent. Wherein V represents a set of nodes, i.e., regions, in the graph; e g Representing a set of edges between nodes; a. The g Representation diagram
Figure BDA0003775604880000042
Of a neighboring matrix of g Each element in (1) indicates whether two nodes are adjacent to each other.
Specifically, travel demands of two geographically adjacent regions present a certain correlation, and a graph is constructed by connecting the two geographically adjacent regions, where an edge in the graph defines a formula as follows:
Figure BDA0003775604880000051
the functional similarity is shown as
Figure BDA0003775604880000052
Representing POI distribution similarity among areas, wherein V represents a set of nodes in the graph, and the nodes are areas; e p Representing a set of edges between nodes; a. The p Representation diagram
Figure BDA0003775604880000053
Of a contiguous matrix of p Each element in (a) represents the POI distribution similarity between two nodes.
The travel demand of passengers is closely related to the functional area distribution of cities, and two areas with similar functions have similar demand patterns even though they are geographically distant. Since POI data may reflect the functionality of an area, the present embodiment uses POI similarity to describe the functional similarity of areas. POIs are classified into 7 categories (residential, school, entertainment, social services, cultural, transportation, and business) with similarity calculated as follows:
A p (i,j)＝1-JS(P i ||P j )；
Figure BDA0003775604880000054
wherein, P i ,
Figure BDA0003775604880000055
The POI distributions of the area i and the area j are indicated, and K indicates the number of POI types. P i (k),P j (k) Indicating the number of POIs of type k in region i and region j, respectively.
The road similarity is shown as
Figure BDA0003775604880000056
Representing the similarity of road characteristics between areas, such as the number of roads, the type of roads, etc. Wherein V represents a collection of nodes in the graph, a nodeNamely, the area; e r Representing a set of edges between nodes; a. The r Representation diagram
Figure BDA0003775604880000057
Of a neighboring matrix of r Each element in (a) represents the similarity of road characteristics between two nodes.
Road characteristics (e.g., total length of road, road type, etc.) are also highly correlated to traffic conditions within the area, and similar to the POI similarity calculation, the present embodiment uses JS divergence to calculate the road similarity between areas, and the calculation formula is as follows:
A r (i,j)＝1-JS(R i ||R j )；
Figure BDA0003775604880000058
wherein R is i ,
Figure BDA0003775604880000059
Indicating the road distribution of the area i and the area j, and L indicating the number of road features. R i (l),R j (l) Indicating the number of roads with feature i in region i and region j, respectively.
And S2, taking unit time as an interval, acquiring travel demand data of each area in the city in each unit time, obtaining a characteristic matrix corresponding to each unit time, and constructing a dynamic demand similarity graph.
In a specific embodiment, the feature matrix includes departure demands and arrival demands of each node in a unit time, that is, the travel demand data includes departure demands and arrival demands. The method and the device for calculating the travel demand data are not limited to specific contents included in the travel demand data, and statistics can be carried out according to vehicles, such as taxies or buses.
Taking a taxi as an example, screening and counting original taxi demand data to obtain departure demand and arrival demand of each region in each unit time by taking 1 hour as a unit, and taking the departure demand and the arrival demand as trip demand data. For example, there are N regions in a city, that is, there are N nodes, and in each unit time interval, each node has 2 features, which are the departure demand and arrival demand of the node at the current unit time, respectively, to form a feature matrix.
The constructed dynamic demand similarity graph is represented as
Figure BDA0003775604880000061
Representing the similarity of travel needs between different regions. Wherein V represents a set of nodes, i.e., regions, in the graph; e d Representing a set of edges between nodes; a. The d Representation diagram
Figure BDA0003775604880000062
Of the adjacent matrix.
The demand similarity degree of the two regions is evaluated based on Jensen-Shannon (JS) divergence, the JS divergence is widely applied to measure the similarity of two probability distributions, the value range is [0,1], and the smaller the JS divergence is, the higher the similarity of the two is shown. The calculation formula is as follows:
A d (i,j)＝1-JS(X i ||X j )；
Figure BDA0003775604880000063
wherein X i (m),X j (M) represents the mth features of zone i and zone j, respectively, and M represents the number of required features, M being equal to 2 in this embodiment, being the departure requirement and the arrival requirement, respectively.
This example E G 、E p 、E R 、E D All represent a set of edges between nodes, and the similarity between nodes is taken as the weight of the edges, thereby forming a graph structure. Each element in the adjacency matrix represents the similarity between nodes.
It should be noted that, in order to avoid that the difference between the data and the neural network parameter is too large, so that the learning rate difference between different layers is obvious, the embodiment performs normalization processing on various types of data, where a normalization formula is as follows:
Figure BDA0003775604880000064
where min (x) is the minimum value in the history data and max (x) is the maximum value in the history data. The data can be distributed more evenly between [0,1] by linear normalization.
And S3, inputting the feature matrix and the adjacency matrixes of the dynamic demand similar graph, the geographical similar graph, the function similar graph and the road similar graph into corresponding coupling graph convolution modules to obtain the spatial features corresponding to the similar graphs, and fusing the spatial features corresponding to the similar graphs to obtain fused spatial features.
In the embodiment, the travel demand is predicted by constructing an urban demand prediction model, and the constructed urban demand prediction model is shown in fig. 2 and comprises a multi-view coupling graph convolution network, a gating cycle unit GRU and an attention layer.
In the step, a multi-view coupling graph convolution network is adopted, and the complex space-time dependency relationship in the travel demand data is extracted by combining the dynamic demand similarity graph of each unit time with the predefined geographic similarity graph, the predefined functional similarity graph and the predefined road similarity graph.
The multi-view coupling graph convolution network comprises coupling graph convolution modules corresponding to four similar graphs respectively, as shown in FIG. 3. The propagation law of the coupling graph convolution module can be expressed as:
Figure BDA0003775604880000071
wherein Z (l) Denotes the input of l +1, Z (l+1) The output of the l +1 layer and the input of the l +2 layer in the network are represented, K represents that a diffusion process model is built by K steps under an undirected graph structure, A represents an adjacency matrix, and W represents trainable parameters.
The multilevel map signal obtained by CGCN is represented as:
Figure BDA0003775604880000072
where M represents the total number of map convolution layers, and the attention mechanism score is calculated by linear variation:
Figure BDA0003775604880000073
Figure BDA0003775604880000074
wherein W α And b α Representing weights and deviations in linear transformation, Z (m) Is Z (m) In a flattened form. Alpha is alpha (m) Is Z (m) H is the final output of CGCN.
According to the difference of the similar graphs of the input, the output h of each coupling graph convolution module is h d ,h g ,h p ,h r The dynamic demand similarity graph, the geographic similarity graph, the function similarity graph and the road similarity graph are respectively and correspondingly input. The formula is as follows:
Figure BDA0003775604880000075
Figure BDA0003775604880000081
Figure BDA0003775604880000082
Figure BDA0003775604880000083
wherein A is d ,A g ,A p ,A r Representing adjacency matrix, X representing input feature matrix, h d ,h g ,h p ,h r Representing the spatial characteristics of the output.
In this embodiment, the adjacency matrix and the feature matrix are input into the multi-view coupling graph convolution network to perform convolution operation, so as to capture the spatial correlation between the regions, and then the output of the multiple views is fused to obtain the fusion spatial feature.
Fusing the spatial features, updating the feature vectors of corresponding nodes in each graph into new vectors with corresponding multiple sizes, and fusing the edge sets in each graph in a union mode to obtain:
H＝(h d ,h g ,h p ,h r )
Figure BDA0003775604880000084
G d ＝(A d ,h d )
Figure BDA0003775604880000085
G r ＝(A r ,h r )
where H represents the new feature matrix in the fused graph and A represents the new adjacency matrix.
And S4, extracting time correlation characteristics from the fusion space characteristics corresponding to each unit time by adopting a gate control cycle unit, extracting attention scores between the time correlation characteristics and external characteristics through an attention layer, taking the attention scores as weights, and performing weighted summation on the time correlation characteristics corresponding to all the unit times to obtain a final prediction result.
The embodiment uses a Gated Round Unit (GRU) to model the dynamic time dependence of the travel demand, then combines external factors (weather events), and finally obtains the output of the urban travel demand prediction through an attention layer and a mapping matrix.
The input of each GRU unit is information of the current moment and the hidden state of the previous moment, and the output is the hidden state of the current moment and is fused with external features. The attention layer is used to capture the degree of influence of the historical time step on the target time step. And predicting the departure flow and the arrival flow respectively, wherein X is a prediction result of the departure flow or a prediction result of the arrival flow.
When using gated round robin units to model dynamic temporal correlations of travel demands, the temporal correlations are extracted using GRUs:
Figure BDA0003775604880000091
wherein H t Represents the output of the multi-view coupling graph convolution network at unit time t,
Figure BDA0003775604880000092
representing a hidden state in unit time t, d H Representing the number of hidden units in the GRU, is typically set to 64 sizes.
All GRU units share the same parameters, and the output is the hidden state of all GRU units
Figure BDA0003775604880000093
Represents the time-related characteristics corresponding to all the unit times, wherein,
Figure BDA0003775604880000094
Figure BDA0003775604880000095
where N represents the number of nodes.
The weather condition has a great influence on the traffic demand, the embodiment divides the weather into five grades of sunny days, cloudy days, rainy days, snowy days and foggy days, the temperature is scaled to the range of [0,1] through the minimum and maximum linear normalization, and the data of all external factors are connected into tensors which are input into the model as external features.
Attention tier by computing U and appearance E = { E = { E } 1 ,e 2 ,…,e T Attention score between (a) }, wherein
Figure BDA0003775604880000096
d e Time dependency is captured dynamically with temporal attention, expressed as the number of extrinsic features:
α＝soft max(Re L U(UW U +EW E +b α ))；
wherein the content of the first and second substances,
Figure BDA0003775604880000097
is a weight parameter that can be learned by the user,
Figure BDA0003775604880000098
Figure BDA0003775604880000099
is a temporal attention score vector representing the distribution of importance of different historical time intervals over the target interval. Re L U is an activation function, and softmax is a classification function, which are well-established technologies in the field and are not described herein.
All the time-related features and the corresponding attention scores are obtained, and all the time-related features are subjected to weighted summation, so that the final prediction result obtained by time-correlation extraction can be represented as:
Figure BDA00037756048800000910
in a specific embodiment, the urban travel demand prediction method based on the convolution of the dynamic multi-view coupling graph further includes:
and converting the prediction result into a grid form through a mapping matrix of the areas and the urban grid to obtain the prediction result of each area.
The mapping matrix of this embodiment represents the correspondence between the area nodes and the grid, the grid is a set of all areas of the whole city, and the output is converted into a grid form:
Figure BDA00037756048800000911
obtaining a prediction result:
Figure BDA0003775604880000101
according to the method and the system, the characteristic matrix, the dynamic demand similar graph, the geographical similar graph, the function similar graph and the road similar graph in historical unit time are input into the network, and the travel demand of the next unit time is predicted, so that the scheduling of vehicles and the command of traffic can be accurately guided, and great convenience is brought to urban travel.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (5)

1. The urban trip demand prediction method based on the convolution of the dynamic multi-view coupling graph is characterized by comprising the following steps of:
acquiring geographic position data, POI data and road data of each area in a city, and constructing a geographic similarity graph, a functional similarity graph and a road similarity graph by taking each area as a node;
taking unit time as an interval, acquiring travel demand data of each region in a city within each unit time, obtaining a characteristic matrix corresponding to each unit time, and constructing a dynamic demand similarity graph;
inputting the characteristic matrix and the adjacent matrixes of the dynamic demand similar graph, the geographical similar graph, the functional similar graph and the road similar graph into corresponding coupling graph convolution modules to obtain the spatial characteristics corresponding to the similar graphs, and fusing the spatial characteristics corresponding to the similar graphs to obtain fused spatial characteristics;
and extracting time correlation characteristics from the fusion space characteristics corresponding to each unit time by adopting a gate control cycle unit, extracting attention scores between the time correlation characteristics and external characteristics through an attention layer, taking the attention scores as weights, and performing weighted summation on the time correlation characteristics corresponding to all the unit times to obtain a final prediction result.
2. The dynamic multi-view coupled graph convolution-based urban travel demand prediction method according to claim 1, wherein the geographical similarity graph is represented as
Figure FDA0003775604870000011
Where V represents the set of nodes in the graph, E g Representing sets of edges between nodes, A g Representation diagram
Figure FDA0003775604870000012
Of a neighboring matrix of g Each element in the group represents whether two nodes are adjacent or not;
A g any one of the elements A g (i, j) is expressed as:
Figure FDA0003775604870000013
the functional similarity is shown as
Figure FDA0003775604870000014
Where V represents the set of nodes in the graph, E p Representing sets of edges between nodes, A p Representation diagram
Figure FDA0003775604870000015
Of a neighboring matrix of p Each element in the set represents POI distribution similarity between two nodes;
A p any one of the elements A p (i, j) is expressed as:
A p (i,j)＝1-JS(P i ||P j )；
Figure FDA0003775604870000016
wherein the content of the first and second substances,
Figure FDA0003775604870000017
POI distribution representing area i and area j, K representing number of POI types, P i (k),P j (k) Respectively representing the number of POIs with the type k in the area i and the area j;
the road similarity is shown as
Figure FDA0003775604870000021
Where V represents the set of nodes in the graph, E r Representing sets of edges between nodes, A r Representation diagram
Figure FDA0003775604870000022
Of a neighboring matrix of r Each element in the set represents the similarity of road characteristics between two nodes;
A r any one of the elements A r (i, j) is expressed as:
A r (i,j)＝1-JS(R i ||R j )；
Figure FDA0003775604870000023
wherein the content of the first and second substances,
Figure FDA0003775604870000024
indicating the road distribution of area i and area j, L indicating the number of road features, R i (l),R j (l) Respectively representing the number of roads with the characteristic l in the area i and the area j;
the dynamic demand similarity is represented as
Figure FDA0003775604870000025
Where V represents the set of nodes in the graph, E d Representing sets of edges between nodes, A d Representation diagram
Figure FDA0003775604870000026
Of a neighboring matrix of d Each element in the tree represents the similarity of travel demands between two nodes;
A d any one of the elements A d (i, j) is expressed as:
A d (i,j)＝1-JS(X i ||X j )；
Figure FDA0003775604870000027
wherein X i (m),X j (M) represents the mth features of the region i and the region j, respectively, and M represents the number of required features.
3. The urban travel demand prediction method based on dynamic multi-view coupling graph convolution according to claim 1, characterized in that the feature matrix comprises departure demands and arrival demands of each node in unit time.
4. The method for forecasting urban travel demand based on convolution of dynamic multi-view coupling graph according to claim 1, characterized in that the attention score between the time-related feature and the external feature is extracted through the attention layer, and is expressed by the following formula:
α＝softmax(ReLU(UW U +EW E +b α ))；
wherein U represents time-related features corresponding to all unit times, E represents external features, and W U 、W E And b α Representing the weight parameter and alpha the attention score vector.
5. The urban travel demand prediction method based on dynamic multi-view coupling graph convolution according to claim 1, wherein the urban travel demand prediction method based on dynamic multi-view coupling graph convolution further comprises:
and converting the prediction result into a grid form through a mapping matrix of the region and the city grid to obtain the prediction result of each region.
CN202210916072.5A 2022-08-01 2022-08-01 Urban trip demand prediction method based on dynamic multi-view coupling graph convolution Pending CN115456238A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210916072.5A CN115456238A (en) 2022-08-01 2022-08-01 Urban trip demand prediction method based on dynamic multi-view coupling graph convolution

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210916072.5A CN115456238A (en) 2022-08-01 2022-08-01 Urban trip demand prediction method based on dynamic multi-view coupling graph convolution

Publications (1)

Publication Number Publication Date
CN115456238A true CN115456238A (en) 2022-12-09

Family

ID=84296076

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210916072.5A Pending CN115456238A (en) 2022-08-01 2022-08-01 Urban trip demand prediction method based on dynamic multi-view coupling graph convolution

Country Status (1)

Country Link
CN (1) CN115456238A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117829375A (en) * 2024-02-29 2024-04-05 华侨大学 Method, device, equipment and medium for predicting multi-region demand of inter-city shuttle passenger transport

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117829375A (en) * 2024-02-29 2024-04-05 华侨大学 Method, device, equipment and medium for predicting multi-region demand of inter-city shuttle passenger transport
CN117829375B (en) * 2024-02-29 2024-05-28 华侨大学 Method, device, equipment and medium for predicting multi-region demand of inter-city shuttle passenger transport

Similar Documents

Publication Publication Date Title
CN112328914B (en) Task allocation method based on space-time crowdsourcing worker behavior prediction
CN110503104B (en) Short-time remaining parking space quantity prediction method based on convolutional neural network
CN108986453A (en) A kind of traffic movement prediction method based on contextual information, system and device
CN111242395B (en) Method and device for constructing prediction model for OD (origin-destination) data
CN113159364A (en) Passenger flow prediction method and system for large-scale traffic station
CN113077090A (en) Passenger flow prediction method, system and computer readable storage medium
Liang et al. A deep spatio-temporal fuzzy neural network for passenger demand prediction
CN114692984B (en) Traffic prediction method based on multi-step coupling graph convolution network
CN112488185B (en) Method and system for predicting vehicle operating parameters including spatiotemporal characteristics
CN113283581B (en) Multi-fusion graph network collaborative multi-channel attention model and application method thereof
Jiang et al. CrowdPatrol: A mobile crowdsensing framework for traffic violation hotspot patrolling
CN115565369B (en) Space-time hypergraph convolution traffic flow prediction method and system based on hypergraph
CN111242292A (en) OD data prediction method and system based on deep space-time network
CN113298314A (en) Rail transit passenger flow prediction method considering dynamic space-time correlation
CN115204477A (en) Bicycle flow prediction method of context awareness graph recursive network
CN114066503A (en) Method for forecasting taxi demand based on virtual service area construction
CN115456238A (en) Urban trip demand prediction method based on dynamic multi-view coupling graph convolution
Rahman et al. Attention based deep hybrid networks for traffic flow prediction using google maps data
Thu et al. Multi-source data analysis for bike sharing systems
CN114372830A (en) Network taxi booking demand prediction method based on space-time multi-graph neural network
CN114418606A (en) Network taxi appointment order demand prediction method based on space-time convolutional network
CN112101132B (en) Traffic condition prediction method based on graph embedding model and metric learning
CN113159371A (en) Unknown target feature modeling and demand prediction method based on cross-modal data fusion
CN117494034A (en) Air quality prediction method based on traffic congestion index and multi-source data fusion
Faghih et al. Predicting short-term Uber demand using spatio-temporal modeling: A New York City Case Study

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