A general generation device and method for city start-end matrix based on graph denoising diffusion

Through a general generation model based on graph denoising and diffusion, the problem of inaccurate generation of OD matrices in the existing technology is solved, and high-precision OD matrices are generated on different urban types, which improves the efficiency and quality of urban planning and traffic management. .

CN118761653BActive Publication Date: 2024-11-19TSINGHUA UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411245265.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-06
Publication Date
2024-11-19
Estimated Expiration
2044-09-06

AI Technical Summary

Technical Problem

The prior art cannot accurately generate the starting point to end point matrix (OD matrix) for any city type, resulting in insufficient urban planning and traffic management, and the inability to effectively simulate the population movement mode of any city around the world.

Method used

Using a general generation model based on graph denoising and diffusion, by obtaining the static attribute information of the target city and constructing a moving map of the city cluster, the forward diffusion sub-model and inverse denoising sub-model are used to generate an OD matrix, and combined with the Laplace eigenvector and distance information, denoising processing is performed to generate a high-precision OD matrix.

Benefits of technology

It realizes the generation of high-precision OD matrix on cities of different structures and types, improves urban infrastructure planning efficiency, optimizes traffic flow allocation, reduces congestion, and expands the application value of emergency management and environmental monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118761653B_ABST
    Figure CN118761653B_ABST
Patent Text Reader

Abstract

The present disclosure relates to the field of mobile data generation technology, and in particular to a general generation device and method for a city start-end matrix based on graph denoising diffusion. The method comprises: obtaining static attribute information of a target city; constructing a city group movement graph for the target city according to the static attribute information, the city group movement graph comprising a plurality of nodes and a plurality of directed edges connected between the nodes; inputting the city group movement graph into a trained target generation model to generate a start-end matrix for the target city; the target generation model is a general generation model based on graph denoising diffusion, the general generation model is used to determine a noisy multimodal graph with noise according to the city group movement graph, and then denoising the noisy multimodal graph to generate a start-end matrix for the target city. The excellent performance of high-precision generation is achieved on target cities of various structures and types. The generated start-end matrix has high accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of mobile data generation, and in particular to a general generation device and method for a city start-end matrix based on graph denoising diffusion. Background Art

[0002] The technology of generating mobility data for the entire city involves estimating the flow of people between different areas of the city, which is often expressed in the form of a matrix from the starting point to the end point, that is, the origin-destination matrix (OD matrix, also known as the origin-destination matrix) generation technology. This technology is widely used in urban planning, traffic management, and digital twin cities. By using the generated mobility data, city managers can plan transportation facilities and urban infrastructure more reasonably, thereby avoiding high urban transportation costs caused by unreasonable allocation of spatiotemporal resources.

[0003] However, in the related art, there is currently no method that can generate a reliable and accurate OD matrix for any city type, so as to scientifically and reasonably generate an OD matrix that represents the group flow between areas within all cities. Summary of the invention

[0004] In view of this, the present disclosure proposes a general generation device and method for a city start-end matrix based on graph denoising diffusion.

[0005] According to one aspect of the present disclosure, a general method for generating a city start-end matrix based on graph denoising diffusion is provided, the method comprising:

[0006] Obtaining static attribute information of a target city, wherein the static attribute information includes regional static characteristics of each area in the target city and distance information between different areas, wherein the regional static characteristics include population structure characteristics and / or quantity distribution characteristics of points of interest (POIs) of different categories;

[0007] According to the static attribute information, a city group mobility graph for the target city is constructed, wherein the city group mobility graph includes a plurality of nodes and a plurality of directed edges connected between the nodes, each of the nodes represents a different area of ​​the target city, the attribute information of each of the nodes includes a spatial feature of the area corresponding to the node, and the weight of each of the directed edges represents the OD flow between the areas corresponding to the node;

[0008] Input the city group movement graph into the trained target generation model to generate the start and end point matrix of the target city;

[0009] Among them, the target generation model is a general generation model based on graph denoising diffusion, and the general generation model is used to determine a noisy multimodal graph with noise according to the urban group movement graph, and then denoise the noisy multimodal graph to generate the start and end point matrix of the target city.

[0010] In this way, a start-end matrix for target cities of different structures and types can be generated with very high accuracy, which can provide strong data support for urban infrastructure construction, emergency management, environmental monitoring, etc., and provide guarantees for scientific planning and management.

[0011] In a possible implementation, the general generation model includes a forward diffusion sub-model and an inverse denoising sub-model, the inverse denoising sub-model includes a multi-layer perceptron, a transformer and a fully connected layer, the transformer includes a plurality of graph transformer layers,

[0012] The process of inputting the city group movement graph into the trained target generation model to generate the start and end point matrix of the target city includes:

[0013] Inputting the urban group movement graph into the forward diffusion sub-model to obtain a noisy multimodal graph with noise;

[0014] The noisy start-end flow matrix and the node features obtained by extracting the noisy multimodal graph are respectively input into the corresponding multilayer perceptron for processing to obtain the processed noisy start-end flow matrix and the processed node features;

[0015] Input the processed noisy start-end flow matrix and the processed node features into the transformer to obtain the target noisy start-end flow matrix, in which the node features and the noisy start-end flow matrix output by the transformer layer of the previous graph are used as inputs of the transformer layer of the next graph;

[0016] Inputting the target noise-added start-end traffic matrix into the fully connected layer to predict the noise to be removed;

[0017] The noisy multimodal graph is denoised based on the noise to be denoised to obtain the start-end point matrix of the target city.

[0018] In a possible implementation, in order to further improve the generation quality of the OD matrix, the method further includes:

[0019] Obtaining an inter-region distance matrix corresponding to the target city, and performing spectral decomposition on the inter-region distance matrix to obtain a plurality of Laplace eigenvectors and a plurality of distances, wherein each of the Laplace eigenvectors is used to indicate a position of a corresponding region in a plane space of the target city, and each of the distances is used to indicate a physical space distance between two regions;

[0020] The process of inputting the processed noisy start-end flow matrix and the processed node features into the converter to obtain the target noisy start-end flow matrix includes:

[0021] After combining the multiple Laplace eigenvectors with the processed node features and combining the multiple distances with the processed noisy start-end flow matrix, the resultant are input into the transformer to obtain a target noisy start-end flow matrix;

[0022] Among them, in the converter, the node features output by the previous graph converter layer are combined with the multiple Laplace feature vectors, and the noisy start-end flow matrix output by the previous graph converter layer is combined with the multiple distances as the input of the next graph converter layer.

[0023] In a possible implementation, in order to enable the target generation model to better capture the heterogeneity of OD flow distribution in different cities, the method further includes: a pre-processing step and a post-processing step;

[0024] The pre-processing step includes: before inputting the urban group movement map into the target generation model, performing a logarithmic transformation on the OD flows between different areas in the urban group movement map;

[0025] The post-processing step includes: after generating the start-end matrix of the target city, performing an inverse transformation on the OD flow in the start-end matrix.

[0026] In a possible implementation, the method further includes:

[0027] Acquire a training data set, the training data set comprising a plurality of sample data groups, each of the sample data groups comprising static attribute information of a sample city and a corresponding label start and end point matrix;

[0028] According to the training data set, the preset original model is trained to obtain the target generation model.

[0029] Through the general generation method of the city start-end matrix based on graph denoising diffusion provided by the embodiment of the present disclosure, the group flow of the entire city is regarded as a directed weighted mobile flow network, in which the network nodes represent the various regions of the city, the directed edges represent the flow of people between the regions, and the edge weights reflect the size of the flow of people. On this basis, by collecting the OD matrix data of as many cities as possible in the world to train the model, the trained target generation model can emerge with excellent performance that can achieve high-precision generation on target cities of various structures and types. By applying the optimal graph generation denoising diffusion model, the present disclosure aims to jointly model the urban spatial characteristics of a city with the OD matrix into a directed weighted graph sample, and use the massive samples of massive cities to train a target generation model so that the model obtains sufficiently strong generalization ability. By implementing the above scheme, it is possible to predict and understand the complex patterns of urban group movement, and on this basis, make more scientific suggestions for urban planning and traffic management. By utilizing the technical solution provided by the present disclosure, the planning efficiency and effect of urban infrastructure can be greatly improved, traffic flow distribution can be optimized, congestion can be reduced, and the quality of life of urban residents can be improved. In addition, the technical solution provided by the present disclosure can also be used in multiple fields such as emergency management and environmental monitoring in any city, further expanding its application value.

[0030] According to another aspect of the present disclosure, a general generating device for a city start-end matrix based on graph denoising diffusion is provided, the device comprising:

[0031] An information acquisition module, used to acquire static attribute information of a target city, wherein the static attribute information includes regional static characteristics of each area in the target city and distance information between different areas, wherein the regional static characteristics include population structure characteristics and / or quantity distribution characteristics of different categories of points of interest (POIs);

[0032] A graph construction module, for constructing a city group mobility graph for a target city according to the static attribute information, wherein the city group mobility graph includes a plurality of nodes and a plurality of directed edges connected between the nodes, each of the nodes represents a different area of ​​the target city, the attribute information of each of the nodes includes a spatial feature of the area corresponding to the node, and the weight of each of the directed edges represents the OD flow between the areas corresponding to the node;

[0033] A matrix construction module, used to input the city group movement graph into the trained target generation model to generate the start-end matrix of the target city;

[0034] Among them, the target generation model is a general generation model based on graph denoising diffusion, and the general generation model is used to determine a noisy multimodal graph with noise according to the urban group movement graph, and then denoise the noisy multimodal graph to generate the start and end point matrix of the target city.

[0035] In this way, a start-end matrix for target cities of different structures and types can be generated with very high accuracy, which can provide strong data support for urban infrastructure construction, emergency management, environmental monitoring, etc., and provide guarantees for scientific planning and management.

[0036] In a possible implementation, the general generation model includes a forward diffusion sub-model and a reverse denoising sub-model, the reverse denoising sub-model includes a multi-layer perceptron, a transformer and a fully connected layer, the transformer includes a plurality of graph transformer layers, the city group movement graph is input into a trained target generation model, and a start-end matrix of the target city is generated, including:

[0037] Inputting the urban group movement graph into the forward diffusion sub-model to obtain a noisy multimodal graph with noise;

[0038] The noisy start-end flow matrix and the node features obtained by extracting the noisy multimodal graph are respectively input into the corresponding multilayer perceptron for processing to obtain the processed noisy start-end flow matrix and the processed node features;

[0039] Input the processed noisy start-end flow matrix and the processed node features into the transformer to obtain the target noisy start-end flow matrix, in which the node features and the noisy start-end flow matrix output by the transformer layer of the previous graph are used as inputs of the transformer layer of the next graph;

[0040] Inputting the target noise-added start-end traffic matrix into the fully connected layer to predict the noise to be removed;

[0041] The noisy multimodal graph is denoised based on the noise to be denoised to obtain the start-end point matrix of the target city.

[0042] In a possible implementation manner, the device further includes:

[0043] a decomposition module, used to obtain an inter-region distance matrix corresponding to the target city, and perform spectral decomposition on the inter-region distance matrix to obtain a plurality of Laplace eigenvectors and a plurality of distances, wherein each of the Laplace eigenvectors is used to indicate a position of a corresponding region in a plane space of the target city, and each of the distances is used to indicate a physical space distance between two regions;

[0044] The process of inputting the processed noisy start-end flow matrix and the processed node features into the converter to obtain the target noisy start-end flow matrix includes:

[0045] After combining the multiple Laplace eigenvectors with the processed node features and combining the multiple distances with the processed noisy start-end flow matrix, the resultant are input into the transformer to obtain a target noisy start-end flow matrix;

[0046] Among them, in the converter, the node features output by the previous graph converter layer are combined with the multiple Laplace feature vectors, and the noisy start-end flow matrix output by the previous graph converter layer is combined with the multiple distances as the input of the next graph converter layer.

[0047] In a possible implementation, in order to enable the target generation model to better capture the heterogeneity of OD flow distribution in different cities, the device further includes:

[0048] A pre-processing step module, used for performing logarithmic transformation on the OD flows between different areas in the urban group movement map before inputting the urban group movement map into the target generation model;

[0049] The post-processing step module is used to generate the start-end matrix of the target city and then perform an inverse transformation on the OD flow in the start-end matrix.

[0050] In a possible implementation manner, the device further includes:

[0051] A data set acquisition module, used to acquire a training data set, wherein the training data set includes a plurality of sample data groups, each of which includes static attribute information of a sample city and a corresponding label start and end matrix;

[0052] The model training module is used to train the preset original model according to the training data set to obtain the target generation model.

[0053] Through the general generation device of the city start-end matrix based on graph denoising diffusion provided by the embodiment of the present disclosure, the group flow of the whole city is regarded as a directed weighted mobile flow network, in which the network nodes represent the various regions of the city, the directed edges represent the flow of people between the regions, and the edge weights reflect the size of the flow of people. On this basis, by collecting the OD matrix data of as many cities as possible in the world to train the model, the trained target generation model can emerge with excellent performance that can achieve high-precision generation on target cities of various structures and types. By applying the optimal graph generation denoising diffusion model, the present disclosure aims to jointly model the urban spatial characteristics of a city with the OD matrix into a directed weighted graph sample, and train a target generation model using massive samples of massive cities, so that the model obtains sufficiently strong generalization ability. By implementing the above scheme, it is possible to predict and understand the complex pattern of urban group movement, and on this basis, make more scientific suggestions for urban planning and traffic management. By using the technical solution provided by the present disclosure, the planning efficiency and effect of urban infrastructure can be greatly improved, traffic flow distribution can be optimized, congestion can be reduced, and the quality of life of urban residents can be improved. In addition, the technical solution provided by the present disclosure can also be used in multiple fields such as emergency management and environmental monitoring in any city, further expanding its application value.

[0054] According to another aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0055] According to another aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions implement the above method when executed by a processor.

[0056] According to another aspect of the present disclosure, a computer program product is provided, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0057] Through the general generation device and method of the city start-end matrix based on graph denoising diffusion provided by the embodiment of the present disclosure, the group flow of the entire city is regarded as a directed weighted mobile flow network, in which the network nodes represent the various regions of the city, the directed edges represent the flow of people between the regions, and the edge weights reflect the size of the flow of people. On this basis, by collecting the OD matrix data of as many cities as possible in the world to train the model, the trained target generation model can emerge with excellent performance that can achieve high-precision generation on target cities of various structures and types. By applying the optimal graph generation denoising diffusion model, the solution provided by the embodiment of the present disclosure aims to jointly model the urban spatial characteristics of a city with the OD matrix into a directed weighted graph sample, and train a target generation model using massive samples of massive cities, so that the model obtains sufficiently strong generalization ability. By implementing the solution provided by the embodiment of the present disclosure, it is possible to predict and understand the complex pattern of urban group movement, and on this basis, make more scientific suggestions for urban planning and traffic management. By utilizing the technical solution provided by the present disclosure, the planning efficiency and effect of urban infrastructure can be greatly improved, traffic flow distribution can be optimized, congestion can be reduced, and the quality of life of urban residents can be improved. In addition, the technical solution provided by the present disclosure can also be used in multiple fields such as emergency management and environmental monitoring in any city, further expanding its application value.

[0058] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0060] Figure 1 A schematic diagram showing a problem of a related technical solution.

[0061] Figure 2 A schematic diagram of the application of a general method for generating a city start-end matrix based on graph denoising diffusion according to an embodiment of the present disclosure is shown.

[0062] Figure 3 A flowchart of a general method for generating a city start-end matrix based on graph denoising diffusion according to an embodiment of the present disclosure is shown.

[0063] Figure 4 A schematic diagram of constructing a city group movement graph in a general method for generating a city start-end matrix based on graph denoising diffusion according to an embodiment of the present disclosure is shown.

[0064] Figure 5A schematic diagram showing a forward diffusion process in a general method for generating a city start-end matrix based on graph denoising diffusion according to an embodiment of the present disclosure is shown.

[0065] Figure 6 A schematic diagram showing a reverse denoising process in a general method for generating a city start-end matrix based on graph denoising diffusion according to an embodiment of the present disclosure is shown.

[0066] Figure 7 It is a block diagram of a device 1900 for universal generation of city start-end matrix based on graph denoising diffusion according to an exemplary embodiment. DETAILED DESCRIPTION

[0067] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0068] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0069] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0070] The technology of generating urban group mobility data involves estimating the group flow between different areas of the city, which is often expressed as the OD matrix generation technology from the starting point to the end point. This technology has been widely used in urban planning, traffic management and digital twin cities. By using the generated mobility data, city managers can plan transportation facilities and urban infrastructure more reasonably, thereby avoiding high urban transportation costs caused by unreasonable allocation of spatiotemporal resources.

[0071] However, in the related art, there is currently no method that can generate a reliable and accurate OD matrix for any type of city, so as to scientifically and reasonably generate an OD matrix that represents the group flow of people between all urban areas. The methods in the related art usually only consider the characteristics of a single city independently and use its own or similar data to fit the model for estimation. It lacks the overall characterization of the group flow within the city and cannot effectively simulate the movement pattern of the population of any city in the world. When generating the overall group movement data of the city, the methods in the related art focus on regression prediction based on the characteristics of the starting point and the end point and the interaction characteristics between them (such as distance), and then estimate the flow of people between all areas in the city, thereby constructing the OD matrix of the entire city; or input the urban characteristics of all areas in the city at one time, the distance between all areas, and then output the entire OD matrix at one time. Due to the limitations of data collection, existing work only uses the data of a sample city to train the model, and then migrates to a similar city to generate data. Common models include gravity models, radiation models, random forests, and deep learning neural networks. Although these models can achieve good results when migrating between similar cities, such as Figure 1 As shown, when there is a large difference between the target city and the source city, it cannot be effectively applied.

[0072] The related technology includes the following technical solutions:

[0073] Technical Solution 1: This technology relies on a large amount of mobile signaling data to perform statistical analysis of group-level mobile data directly from intermittent point signaling data. This method faces several limitations, including high data cost requirements, data bias issues, and insufficient data reuse capabilities. These limitations are particularly significant in situations where it is challenging to obtain a large amount of continuous and high-cost mobile signaling data.

[0074] Technical Solution 2: This technology directly collects data through sensors (such as cameras) installed on the traffic network to obtain vehicle travel information on the road network. This method has strict requirements on equipment and requires the extensive deployment of advanced sensor networks in the target city, which is not only costly, but also has a long installation and deployment cycle.

[0075] In summary, the solutions in the related art have the following limitations:

[0076] (1) Existing urban pedestrian flow models are unable to model the entire urban OD matrix. They only focus on each element in the OD matrix independently, and fail to model it from the perspective of the group mobility flow network, thus ignoring the overall characteristics of the group mobility flow network of the entire city. In other words, existing urban pedestrian flow models fail to comprehensively model the OD matrix of cities of any urban type, structure, and size. They only focus on migration between similar cities, while ignoring the differences between the OD matrices of different cities.

[0077] (2) Existing graph generation models cannot cope with the modeling of city-level OD matrices, that is, they cannot model such high-dimensional continuous weighted networks and can only cope with small-scale discrete graph structures. Therefore, there is no way to solve the problem of generating city-level OD matrices.

[0078] (3) Existing methods all use massive mobile phone positioning data or road traffic data on the road network to obtain the OD matrix, which requires extremely high data costs and is difficult to universalize, limiting its widespread application.

[0079] (4) Existing method designs are not suitable for modeling a variety of different types of cities. They often only consider large-scale cities and ignore developing cities. However, data related to the OD matrix of these cities are needed to support scientific and reasonable policy making.

[0080] To solve the above problems, the embodiments of the present disclosure provide a general generation device and method for the city origin and destination matrix based on graph denoising diffusion, which regards the entire city's group flow as a directed weighted mobile flow network (i.e., the city mobile flow network), where the network nodes represent the various areas of the city, the directed edges represent the flow of people between areas, and the edge weights reflect the size of the flow of people. On this basis, Figure 2As shown, by collecting OD matrix data of as many cities as possible in the world to train the model, the trained target generation model can emerge with excellent performance that can achieve high-precision generation on target cities of various structures and types. By applying the optimal graph generation denoising diffusion model, the solution provided by the embodiment of the present disclosure aims to jointly model the urban spatial characteristics of a city with the OD matrix into a directed weighted graph sample, and use massive samples of massive cities to train a target generation model so that the model obtains sufficiently strong generalization ability. By implementing the solution provided by the embodiment of the present disclosure, it is possible to predict and understand the complex patterns of urban group movement, and on this basis, make more scientific suggestions for urban planning and traffic management. By utilizing the technical solution provided by the present disclosure, the planning efficiency and effect of urban infrastructure can be greatly improved, traffic flow distribution can be optimized, congestion can be reduced, and the quality of life of urban residents can be improved. In addition, the technical solution provided by the present disclosure can also be used in multiple fields such as emergency management and environmental monitoring in any city, further expanding its application value.

[0081] like Figure 3 As shown, the general method for generating a city start-end matrix based on graph denoising diffusion provided by the embodiment of the present disclosure includes steps S101 to S103. The method can be applied to a server or an electronic device.

[0082] In step S101, static attribute information of a target city is obtained, wherein the static attribute information includes regional static characteristics of each area in the target city and distance information between different areas, wherein the regional static characteristics include population structure characteristics and / or quantity distribution characteristics of different categories of points of interest (POIs).

[0083] In this embodiment, the distance information between different areas can be the physical space distance between different areas. The population structure characteristics may include: population density (the number of people per unit area); the distribution of demographic attributes such as age, gender, and education level, such as the number of people divided by age group and gender, the total population, the proportion of people of different genders in the total population, etc.; the distribution of the population in cities, villages or different geographical areas; the trend of population growth or decrease, and migration patterns; socioeconomic characteristics such as income level, employment status, and education level; housing type, residential density, house ownership, etc. Among them, the age group division method can be set according to actual needs, and the present disclosure does not limit this. The quantity distribution characteristics of points of interest (POI) may include the number of points of interest of different categories. The categories of points of interest may be catering, entertainment, life services, etc. The categories of points of interest may be set according to actual needs, and the present disclosure does not limit this.

[0084] In this embodiment, the data structure of regional static features can be set in advance to support accurate urban group mobile flow network modeling.

[0085] For example, the data structure of the regional static features of a certain area in the target city can be as follows:

[0086] {"total_population", / / Total population

[0087] "male_fraction", / / The proportion of males in the total population, a floating point number between 0 and 1

[0088] "female_fraction", / / The proportion of women in the total population, a floating point number between 0 and 1

[0089] "0~15", / / The population aged between 0 and 15 in the total population

[0090] "16~25", / / The population aged between 16 and 25 in the total population

[0091] "26~35", / / The population aged between 26 and 35 in the total population

[0092] "36~45", / / The population aged between 36 and 45 in the total population

[0093] "46~55", / / The population aged between 46 and 55 in the total population

[0094] "56~65", / / The population aged between 56 and 65 in the total population

[0095] "65~", / / Population aged 65 and above in the total population

[0096] "food", / / The number of POIs belonging to the restaurant category

[0097] "entertainment", / / the number of POIs that belong to the entertainment category

[0098] "life_service", / / The number of POIs that belong to the life service category

[0099] …

[0100] }

[0101] In step S102, based on the static attribute information, an urban group mobility graph for the target city is constructed, the urban group mobility graph includes multiple nodes and multiple directed edges connected between the nodes, each of the nodes represents a different area of ​​the target city, the attribute information of each node includes the spatial characteristics of the area corresponding to the node, and the weight of each directed edge represents the OD flow between the areas corresponding to the node. Among them, the OD flow (origin and destination flow) is the traffic volume between the origin area and the destination area in the two areas connected by the directed edge.

[0102] In this embodiment, if Figure 4 As shown in Figure 2, the urban group mobility graph can be constructed based on the static attribute information. , . Each node on the graph Represents a region within the target city , V represents the node set corresponding to the target city. And, the directed edge represents the OD flow from area i to area j , E represents the set of all regions in the target city. is the number of nodes in graph G, representing the number of regions in the target city. Represents the cardinality of a set. Each directed edge Weight is the OD flow from region i to region j .

[0103] Figure 4 The graph G shown has attribute information of each region, which can be the spatial characteristic attribute of the region (the spatial characteristic attribute can be determined based on the population structure characteristics, the number distribution characteristics of the points of interest POI, etc. in the regional static characteristics of each region), and these spatial characteristic attributes are used as the spatial characteristics of each node v Therefore, the spatial characteristics (also called spatial features) of the target city The feature matrix To represent this feature matrix From the attribute information of all nodes v on the graph G and all region pairs (a region pair includes a starting region and an end zone ) R represents the set of all regions of the target city.

[0104] At the same time, the OD matrix (origin and destination matrix) of the target city to be generated Equivalent to Figure All directed edges of and its weight A collection of . Fitting the conditional distribution by building a conditional generative model , that is, given all nodes And the spatial characteristics of each node v In the case of and its corresponding weight on each directed edge e , can build an OD matrix generation model The conditional distribution Reflects , where C is the condition of the OD matrix F, that is .

[0105] In step S103, the city group movement graph is input into the trained target generation model to generate the start-end matrix of the target city. The target generation model is a general generation model based on graph denoising diffusion, and the general generation model is used to determine a noisy multimodal graph with noise according to the city group movement graph, and then denoise the noisy multimodal graph to generate the start-end matrix of the target city. In this way, the start-end matrix of target cities of different structures and types can be generated, and the start-end matrix has a very high accuracy, which can provide strong data support for urban infrastructure construction, emergency management, environmental monitoring, etc., and provide guarantee for scientific planning and management.

[0106] In this embodiment, the target generation model is a general generation model based on graph denoising diffusion, and the general generation model includes a method for implementing Figure 5 The forward diffusion submodel of the forward diffusion process q shown and used to implement Figure 6 The reverse denoising process shown The inverse denoising sub-model simulates the conditional distribution Both processes are based on the directed edge e and its corresponding weight The urban cluster movement diagram Since the OD flow is numerical data, continuous Gaussian noise can be used for modeling in the forward diffusion process and the reverse denoising process.

[0107] Among them, the forward diffusion process q can be described by the following calculation formula at each OD flow level.

[0108]

[0109] Among them, q is a conditional probability distribution, For the given conditions Next The probability distribution of is the OD flow from area i to area j at diffusion step t, is the OD flow from area i to area j at diffusion step t-1. represents a Gaussian distribution, Controls the noise size of diffusion step t, which is the preset value. It is a diagonal matrix. For the given conditions Next The probability distribution of is the OD flow from area i to area j at diffusion step 0, T is the preset value, and t, i, and j are all positive integers.

[0110] The reverse denoising process is the reverse process of the forward diffusion process. The reverse denoising process consists of the denoising neural network To promote, the denoising neural network θ is based on The potential state of the noise space at a certain moment predicts the tiny noise to be removed , the goal is to achieve The noise state of the step. Different from the forward diffusion process, to ensure that the simulated OD matrix The joint distribution of all elements in , the noise that needs to be removed for each directed edge must be based on the corresponding noised data In addition, in order to ensure that the OD matrix can be generated for the target city with given spatial characteristics, a denoising process of the OD matrix guided by the urban spatial characteristics is designed, which is guided by the spatial characteristics of the corresponding city, i.e., the nodes and their characteristics. Therefore, the denoising step in the inverse process can finally be expressed by the following formula.

[0111]

[0112] in, .

[0113] in, and . is the default value, is the control parameter of the noise level of diffusion step i in the diffusion process, is the control parameter for the accumulated noise level. It is the intermediate state of diffusion step t-1 in the reverse denoising process. It is the intermediate state of diffusion step t in the reverse denoising process. It is to predict the pure noise map (such as Figure 5 ). Represents the spatial characteristics of the target city and is used to control the output of the diffusion process. Given a denoising neural network , intermediate state and the spatial characteristics of the target city In the case of The probability distribution of . is the noise predicted by the denoising neural network θ, which is based on the intermediate state , diffusion step and the spatial characteristics of the corresponding target cities conduct.

[0114] Then in step S103, Figure 5 As shown, the urban group movement graph is input into the forward diffusion sub-model to obtain a noisy multimodal graph with noise. Figure 5 Original data graph This is the urban cluster movement map.

[0115] In each step of the inverse denoising process, the denoising neural network θ predicts the tiny Gaussian noise that needs to be removed based on the current noisy state. . To this end, the inverse denoising sub-model adopts a transformer-based neural network architecture with excellent learning ability and wide generalization performance. This neural network architecture can not only capture the complex relationship between data, but also adapt to a variety of different data distributions, thereby enhancing the model's ability to handle various noise states. In this way, the noise components embedded in the OD traffic can be accurately located and eliminated, thereby gradually restoring the true state of the data.

[0116] like Figure 6 As shown, the inverse denoising sub-model may include a multilayer perceptron (MLP), a transformer and a fully connected layer, and the transformer includes multiple graph transformer layers. After obtaining the noisy multimodal graph with noise in step S103, it should also be: input the noisy start-end flow matrix and node features obtained by extracting the noisy multimodal graph into the corresponding multilayer perceptron for processing, respectively, to obtain the processed noisy start-end flow matrix and the processed node features; input the processed noisy start-end flow matrix and the processed node features into the transformer, to obtain the target noisy start-end flow matrix, in which the node features and the noisy start-end flow matrix output by the previous graph transformer layer are used as the input of the next graph transformer layer; input the target noisy start-end flow matrix into the fully connected layer, and predict the noise to be removed. Based on the noise to be removed The noisy multimodal graph is denoised to obtain the origin and destination matrix (ie, OD matrix) of the target city.

[0117] Each graph transformer layer in the transformer can simultaneously accept inputs from nodes and edges, capture graph features, and then output noise predictions at the edge level. The features of each region (i.e., node features) serve as node inputs, while the noisy OD matrix (i.e., the noisy start-end flow matrix) in the current state provides edge inputs. These inputs are processed by their respective multi-layer perceptrons and then sent to the transformer. The graph transformer contains multiple graph transformer layers. In each graph transformer layer, each node calculates attention weights with all other nodes through the self-attention mechanism, and aggregates information from all other nodes based on these weights to form new node feature outputs. In order to simulate the dependency between nodes and edges, the weights calculated by self-attention are combined with the noisy start-end flow matrix using feature linear modulation (FiLM) to obtain the final attention weights. At the same time, the calculated attention information is also used to combine with the original noisy start-end flow matrix as the new noisy start-end flow matrix in the calculation of the next layer of denoising neural network. In addition, after the node and edge information is aggregated, the data is passed through the feedforward network. The calculations within each graph transformer layer can be described by the following formula.

[0118]

[0119] ,

[0120] ,

[0121]

[0122] in, and They respectively represent the node features and the noisy start-end traffic matrix (that is, the edge features, with the start point being node i and the end point being node j) of the node i in the l-th graph transformer layer in the transformer. and They represent the node features and the noisy start-end traffic matrix in the l+1th graph transformer layer in the transformer respectively. , and It is Tier The matrix of query, key, and value for each attention head. It is Layer The weight matrix of the attention head. and It is The output of the multilayer perceptron with layer nodes and the noisy start and end traffic matrix. are the dimensions of the query and key vectors. is the attention coefficient between node i and node j of the kth attention head in the lth graph transformer layer. is the number of attention heads, which is the default value. Is with the node A set of connected neighbor nodes.

[0123] After multiple layers of processing by the transformer, the final output target noisy start-end traffic matrix is ​​fed into the fully connected layer to predict the noise.

[0124] In a possible implementation, the method may further include: obtaining an inter-region distance matrix corresponding to the target city, and performing spectral decomposition on the inter-region distance matrix to obtain a plurality of Laplace eigenvectors and a plurality of distances, wherein each Laplace eigenvector is used to indicate a position of a corresponding region in a plane space of the target city, and each distance is used to indicate a distance between two regions. The inter-region distance matrix may be constructed based on the distance information between different regions in the static attribute information of the target city.

[0125] The step of inputting the processed noisy start-end flow matrix and the processed node features into the converter to obtain the target noisy start-end flow matrix may include: combining the multiple Laplace eigenvectors with the processed node features, and combining the multiple distances with the processed noisy start-end flow matrix, and then inputting the combined data into the converter to obtain the target noisy start-end flow matrix;

[0126] Among them, in the converter, the node features output by the previous graph converter layer are combined with the multiple Laplace feature vectors, and the noisy start-end flow matrix output by the previous graph converter layer is combined with the multiple distances as the input of the next graph converter layer.

[0127] To fully utilize the spatial interaction With OD matrix The relationship between them is designed based on the distance condition guidance at the node and edge level to guide the denoising generation. Figure 6 As shown, the features of the distance matrix between regions are spectrally decomposed before being input into the neural network to obtain The Laplace feature vectors and multiple distances obtained based on the region index are used to implement distance-based Laplace position encoding (d-LaPEs) to encode the specific location of each region in the planar urban space. The calculation process is as follows:

[0128]

[0129]

[0130]

[0131] Where L is the Laplace matrix based on the distance between regions. is the degree matrix of the target city, and W is the weight matrix of the target city. The degree matrix D contains the degree of each node (region) (i.e., the number of edges connected to the node), and the weight matrix W contains the weights between nodes (such as distance). is a diagonal matrix including the eigenvalues ​​of L. The result of spectral decomposition. is a set of Laplace eigenvectors, yes is the transpose. The columns of are the eigenvectors of L.

[0132] Subsequently, the node features and the noisy origin-destination traffic matrix are combined with the corresponding d-LaPEs and distances before being input into each graph transformer layer.

[0133] In a possible implementation, in order to enable the target generation model to better capture the heterogeneity of OD flow distribution in different cities, logarithmic transformation can be used to pre-process and post-process the OD flow. The method can also include: a pre-processing step and a post-processing step.

[0134] The pre-processing step may include: before inputting the urban group movement map into the target generation model, performing a pre-processing on the OD flow between different areas in the urban group movement map. Perform logarithmic transformation, as shown in the formula , and the logarithmic transformed OD flow is obtained . is the logarithmically transformed OD flow, which is used to train the denoising neural network θ.

[0135] The post-processing step may include: after generating the start-end matrix of the target city, processing the OD flow in the start-end matrix Perform the inverse transformation, as shown in the formula , get the actual size of OD flow .

[0136] In a possible implementation, the method may further include:

[0137] A training data set is obtained, wherein the training data set includes a plurality of sample data groups, each of which includes static attribute information of a sample city and a corresponding label start and end point matrix; and a preset original model is trained according to the training data set to obtain the target generation model.

[0138] Among them, the static attribute information of different sample cities and the corresponding label start and end matrices can be obtained according to actual needs, so that the generated target generation model has better performance. A target generation model can be trained using a large number of sample data sets from a large number of cities to enable the model to have sufficiently strong generalization capabilities.

[0139] For example, in a specific experimental scenario, the attributes of all cities in the United States, group mobility flow data and OD matrix are used as the basis, and the training set, validation set and test set are divided into 8:1:1 to verify the effectiveness of the general generation method of the city origin and destination matrix based on graph denoising diffusion provided by the embodiment of the present disclosure. The data set includes regional demographic characteristic data corresponding to the spatial division of the census area, and POI distribution data from a public map service. The solution provided by the embodiment of the present disclosure is compared with the existing method to verify the advancement and effectiveness of the solution provided by the embodiment of the present disclosure. Evaluation indicators include root mean square error (RMSE) and commuting common part (CPC).

[0140] The calculation method is as follows:

[0141]

[0142]

[0143] Among them, RMSE is the square root of the average of the squares of the differences between the predicted values ​​and the true values. RMSE is used to measure the starting and ending matrix of the prediction The actual start and end matrix The difference between. is the number of elements in the origin and destination matrix. R is the set of all regions in the city. They represent the i-th region and the j-th region in the set R respectively.

[0144] Is a ratio indicator used to measure the start and end point matrix of the forecast The actual start and end matrix The degree of overlap between them is calculated by the proportion of the common parts between the two, which reflects the degree of match between the predicted value and the actual true value. The closer the CPC value is to 1, the more accurate the prediction is.

[0145] Finally, the solution provided by the embodiment of the present disclosure achieved superior performance compared with the gravity model, random forest, and graph representation learning machine, as shown in Table 1. Obviously, the solution provided by the embodiment of the present disclosure can achieve excellent performance of high-precision generation on cities of various structures and types, and the model has sufficiently strong generalization ability.

[0146] Table 1 Test results

[0147] Model RMSE CPC Gravity Model 174.0 0.321 Random Forest 100.4 0.458 Graph Representation Learning Machine 94.3 0.440 Embodiments of the present disclosure 68.6 0.593

[0148] The present disclosure also provides a general generation device for a city start-end matrix based on graph denoising diffusion, the device comprising:

[0149] An information acquisition module, used to acquire static attribute information of a target city, wherein the static attribute information includes regional static characteristics of each area in the target city and distance information between different areas, wherein the regional static characteristics include population structure characteristics and / or quantity distribution characteristics of different categories of points of interest (POIs);

[0150] A graph construction module, for constructing a city group mobility graph for a target city according to the static attribute information, wherein the city group mobility graph includes a plurality of nodes and a plurality of directed edges connected between the nodes, each of the nodes represents a different area of ​​the target city, the attribute information of each of the nodes includes a spatial feature of the area corresponding to the node, and the weight of each of the directed edges represents the OD flow between the areas corresponding to the node;

[0151] A matrix construction module, used to input the city group movement graph into the trained target generation model to generate the start-end matrix of the target city;

[0152] Among them, the target generation model is a general generation model based on graph denoising diffusion, and the general generation model is used to determine a noisy multimodal graph with noise according to the urban group movement graph, and then denoise the noisy multimodal graph to generate the start and end point matrix of the target city.

[0153] In this way, a start-end matrix for target cities of different structures and types can be generated with very high accuracy, which can provide strong data support for urban infrastructure construction, emergency management, environmental monitoring, etc., and provide guarantees for scientific planning and management.

[0154] In a possible implementation, the general generation model includes a forward diffusion sub-model and an inverse denoising sub-model, the inverse denoising sub-model includes a multi-layer perceptron, a transformer and a fully connected layer, and the transformer includes multiple graph transformer layers.

[0155] The step of inputting the city group movement graph into a trained target generation model to generate a start-end matrix of the target city may include:

[0156] Inputting the urban group movement graph into the forward diffusion sub-model to obtain a noisy multimodal graph with noise;

[0157] The noisy start-end flow matrix and the node features obtained by extracting the noisy multimodal graph are respectively input into the corresponding multilayer perceptron for processing to obtain the processed noisy start-end flow matrix and the processed node features;

[0158] Input the processed noisy start-end flow matrix and the processed node features into the transformer to obtain the target noisy start-end flow matrix, in which the node features and the noisy start-end flow matrix output by the transformer layer of the previous graph are used as inputs of the transformer layer of the next graph;

[0159] Inputting the target noise-added start-end traffic matrix into the fully connected layer to predict the noise to be removed;

[0160] The noisy multimodal graph is denoised based on the noise to be denoised to obtain the start-end point matrix of the target city.

[0161] In a possible implementation, the device may further include:

[0162] a decomposition module, used to obtain an inter-region distance matrix corresponding to the target city, and perform spectral decomposition on the inter-region distance matrix to obtain a plurality of Laplace eigenvectors and a plurality of distances, wherein each of the Laplace eigenvectors is used to indicate a position of a corresponding region in a plane space of the target city, and each of the distances is used to indicate a physical space distance between two regions;

[0163] The step of inputting the processed noisy start-end flow matrix and the processed node features into the converter to obtain the target noisy start-end flow matrix may include:

[0164] After combining the multiple Laplace eigenvectors with the processed node features and combining the multiple distances with the processed noisy start-end flow matrix, the resultant are input into the transformer to obtain a target noisy start-end flow matrix;

[0165] Among them, in the converter, the node features output by the previous graph converter layer are combined with the multiple Laplace feature vectors, and the noisy start-end flow matrix output by the previous graph converter layer is combined with the multiple distances as the input of the next graph converter layer.

[0166] In a possible implementation, the device may further include:

[0167] A pre-processing step module, used for performing logarithmic transformation on the OD flows between different areas in the urban group movement map before inputting the urban group movement map into the target generation model;

[0168] The post-processing step module is used to generate the start-end matrix of the target city and then perform an inverse transformation on the OD flow in the start-end matrix.

[0169] In a possible implementation, the device may further include:

[0170] A data set acquisition module, used to acquire a training data set, wherein the training data set includes a plurality of sample data groups, each of which includes static attribute information of a sample city and a corresponding label start and end matrix;

[0171] The model training module is used to train the preset original model according to the training data set to obtain the target generation model.

[0172] The general generation device of the city start-end matrix based on graph denoising diffusion provided by the embodiment of the present disclosure regards the group flow of the whole city as a directed weighted mobile flow network (i.e., the urban mobile flow network), in which the network nodes represent the various regions of the city, the directed edges represent the flow of people between the regions, and the edge weights reflect the size of the flow of people. On this basis, by collecting the OD matrix data of as many cities as possible in the world to train the model, the trained target generation model can emerge with excellent performance that can achieve high-precision generation on target cities of various structures and types. By applying the optimal graph generation denoising diffusion model, the solution provided by the embodiment of the present disclosure aims to jointly model the urban spatial characteristics of a city with the OD matrix into a directed weighted graph sample, and train a target generation model using massive samples of massive cities, so that the model obtains sufficiently strong generalization ability. By implementing the solution provided by the embodiment of the present disclosure, it is possible to predict and understand the complex pattern of urban group movement, and on this basis, make more scientific suggestions for urban planning and traffic management. By utilizing the technical solution provided by the present disclosure, the planning efficiency and effect of urban infrastructure can be greatly improved, traffic flow distribution can be optimized, congestion can be reduced, and the quality of life of urban residents can be improved. In addition, the technical solution provided by the present disclosure can also be used in multiple fields such as emergency management and environmental monitoring in any city, further expanding its application value.

[0173] It should be noted that, although the above embodiment is used as an example to introduce the general generation method and device of the city start-end matrix based on graph denoising diffusion, those skilled in the art can understand that the present disclosure should not be limited thereto. In fact, the user can flexibly set each module and step according to personal preferences and / or actual application scenarios, as long as it conforms to the idea of ​​the technical solution of the present disclosure.

[0174] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0175] The embodiment of the present disclosure also provides a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the above method when executed by a processor. The computer-readable storage medium can be a volatile or non-volatile computer-readable storage medium.

[0176] An embodiment of the present disclosure further proposes an electronic device, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0177] The embodiments of the present disclosure also provide a computer program product, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0178] Figure 7 1 is a block diagram of a device 1900 for universal generation of a city start-end matrix based on graph denoising diffusion according to an exemplary embodiment. For example, the device 1900 may be provided as a server or an electronic device. Figure 7 , the apparatus 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions, such as an application, that can be executed by the processing component 1922. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0179] The device 1900 may also include a power supply component 1926 configured to perform power management of the device 1900, a wired or wireless network interface 1950 configured to connect the device 1900 to a network, and an input / output interface 1958 (I / O interface). The device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server 2000. TM , MacOS X TM , Unix TM , Linux TM , FreeBSD TM or similar.

[0180] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the device 1900 to perform the above method.

[0181] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0182] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0183] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0184] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0185] Various aspects of the present disclosure are described herein with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of boxes in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0186] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0187] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0188] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0189] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A general method for generating city start-end matrix based on graph denoising diffusion, characterized in that: The method comprises: Obtaining static attribute information of a target city, wherein the static attribute information includes regional static characteristics of each area in the target city and distance information between different areas, wherein the regional static characteristics include population structure characteristics and / or quantity distribution characteristics of points of interest (POIs) of different categories; According to the static attribute information, a city group mobility graph for the target city is constructed, wherein the city group mobility graph includes a plurality of nodes and a plurality of directed edges connected between the nodes, each of the nodes represents a different area of ​​the target city, the attribute information of each of the nodes includes a spatial feature of the area corresponding to the node, and the weight of each of the directed edges represents the start-end OD flow between the areas corresponding to the node; Input the city group movement graph into the trained target generation model to generate the start and end point matrix of the target city; The target generation model is a general generation model based on graph denoising diffusion, and the general generation model is used to determine a noisy multimodal graph with noise according to the urban group movement graph, and then denoise the noisy multimodal graph to generate the start-end matrix of the target city; The general generation model includes a forward diffusion sub-model and a reverse denoising sub-model, the reverse denoising sub-model includes a multi-layer perceptron, a transformer and a fully connected layer, the transformer includes multiple graph transformer layers, The process of inputting the city group movement graph into the trained target generation model to generate the start and end point matrix of the target city includes: Inputting the urban group movement graph into the forward diffusion sub-model to obtain a noisy multimodal graph with noise; The noisy start-end flow matrix and the node features obtained by extracting the noisy multimodal graph are respectively input into the corresponding multilayer perceptron for processing to obtain the processed noisy start-end flow matrix and the processed node features; Input the processed noisy start-end flow matrix and the processed node features into the transformer to obtain the target noisy start-end flow matrix, in which the node features and the noisy start-end flow matrix output by the transformer layer of the previous graph are used as inputs of the transformer layer of the next graph; Inputting the target noise-added start-end traffic matrix into the fully connected layer to predict the noise to be removed; De-noising the noisy multimodal graph based on the noise to be removed to obtain the start-end matrix of the target city; Wherein, the method further comprises: Obtaining an inter-region distance matrix corresponding to the target city, and performing spectral decomposition on the inter-region distance matrix to obtain a plurality of Laplace eigenvectors and a plurality of distances, wherein each of the Laplace eigenvectors is used to indicate a position of a corresponding region in a plane space of the target city, and each of the distances is used to indicate a physical space distance between two regions; The process of inputting the processed noisy start-end flow matrix and the processed node features into the converter to obtain the target noisy start-end flow matrix includes: After combining the multiple Laplace eigenvectors with the processed node features and combining the multiple distances with the processed noisy start-end flow matrix, the resultant are input into the transformer to obtain a target noisy start-end flow matrix; Among them, in the converter, the node features output by the previous graph converter layer are combined with the multiple Laplace feature vectors, and the noisy start-end flow matrix output by the previous graph converter layer is combined with the multiple distances as the input of the next graph converter layer.

2. The method according to claim 1, characterized in that The method further comprises: a pre-processing step and a post-processing step; The pre-processing step includes: before inputting the urban group movement map into the target generation model, performing a logarithmic transformation on the OD flows between different areas in the urban group movement map; The post-processing step includes: after generating the start-end matrix of the target city, performing an inverse transformation on the OD flow in the start-end matrix.

3. The method according to claim 1, characterized in that The method further comprises: Acquire a training data set, the training data set comprising a plurality of sample data groups, each of the sample data groups comprising static attribute information of a sample city and a corresponding label start and end point matrix; According to the training data set, the preset original model is trained to obtain the target generation model.

4. A general generation device for city start-end matrix based on graph denoising diffusion, characterized in that: The device comprises: An information acquisition module, used to acquire static attribute information of a target city, wherein the static attribute information includes regional static characteristics of each area in the target city and distance information between different areas, wherein the regional static characteristics include population structure characteristics and / or quantity distribution characteristics of different categories of points of interest (POIs); A graph construction module, for constructing a city group mobility graph for a target city according to the static attribute information, wherein the city group mobility graph includes a plurality of nodes and a plurality of directed edges connected between the nodes, each of the nodes represents a different area of ​​the target city, the attribute information of each of the nodes includes a spatial feature of the area corresponding to the node, and the weight of each of the directed edges represents the OD flow between the areas corresponding to the node; A matrix construction module, used to input the city group movement graph into the trained target generation model to generate the start-end matrix of the target city; The target generation model is a general generation model based on graph denoising diffusion, and the general generation model is used to determine a noisy multimodal graph with noise according to the urban group movement graph, and then denoise the noisy multimodal graph to generate the start-end matrix of the target city; The general generation model includes a forward diffusion sub-model and a reverse denoising sub-model, the reverse denoising sub-model includes a multi-layer perceptron, a transformer and a fully connected layer, the transformer includes multiple graph transformer layers, The process of inputting the city group movement graph into the trained target generation model to generate the start and end point matrix of the target city includes: Inputting the urban group movement graph into the forward diffusion sub-model to obtain a noisy multimodal graph with noise; The noisy start-end flow matrix and the node features obtained by extracting the noisy multimodal graph are respectively input into the corresponding multilayer perceptron for processing to obtain the processed noisy start-end flow matrix and the processed node features; Input the processed noisy start-end flow matrix and the processed node features into the transformer to obtain the target noisy start-end flow matrix, in which the node features and the noisy start-end flow matrix output by the transformer layer of the previous graph are used as inputs of the transformer layer of the next graph; Inputting the target noise-added start-end traffic matrix into the fully connected layer to predict the noise to be removed; De-noising the noisy multimodal graph based on the noise to be removed to obtain the start-end matrix of the target city; Wherein, the device further comprises: a decomposition module, used to obtain an inter-region distance matrix corresponding to the target city, and perform spectral decomposition on the inter-region distance matrix to obtain a plurality of Laplace eigenvectors and a plurality of distances, wherein each of the Laplace eigenvectors is used to indicate a position of a corresponding region in a plane space of the target city, and each of the distances is used to indicate a physical space distance between two regions; The process of inputting the processed noisy start-end flow matrix and the processed node features into the converter to obtain the target noisy start-end flow matrix includes: After combining the multiple Laplace eigenvectors with the processed node features and combining the multiple distances with the processed noisy start-end flow matrix, the resultant are input into the transformer to obtain a target noisy start-end flow matrix; Among them, in the converter, the node features output by the previous graph converter layer are combined with the multiple Laplace feature vectors, and the noisy start-end flow matrix output by the previous graph converter layer is combined with the multiple distances as the input of the next graph converter layer.

5. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to implement the method described in any one of claims 1 to 3 when executing the instructions stored in the memory.

6. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 3 is implemented.