WO2021174876A1 - 基于智能决策的人口流动预测方法、装置及计算机设备 - Google Patents

基于智能决策的人口流动预测方法、装置及计算机设备 Download PDF

Info

Publication number
WO2021174876A1
WO2021174876A1 PCT/CN2020/124428 CN2020124428W WO2021174876A1 WO 2021174876 A1 WO2021174876 A1 WO 2021174876A1 CN 2020124428 W CN2020124428 W CN 2020124428W WO 2021174876 A1 WO2021174876 A1 WO 2021174876A1
Authority
WO
WIPO (PCT)
Prior art keywords
node
feature
spatial
city
information
Prior art date
Application number
PCT/CN2020/124428
Other languages
English (en)
French (fr)
Inventor
贾雪丽
王健宗
张之勇
程宁
Original Assignee
平安科技（深圳）有限公司
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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021174876A1 publication Critical patent/WO2021174876A1/zh

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A30/00—Adapting or protecting infrastructure or their operation
    • Y02A30/60—Planning or developing urban green infrastructure

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment, and storage medium for predicting population flow based on intelligent decision-making.
  • the purpose of the embodiments of the present application is to propose a population flow prediction method, device, computer equipment, and storage medium based on intelligent decision-making, so as to solve the problem of low accuracy of population flow prediction.
  • the embodiments of the present application provide a population flow prediction method based on intelligent decision-making, which adopts the following technical solutions:
  • an embodiment of the present application also provides a population flow prediction device based on intelligent decision-making, which adopts the following technical solutions:
  • Map acquisition module for acquiring city maps
  • the map division module is used to divide the city map, and use the city areas in the city map as nodes to generate a city node network;
  • An information acquisition module for acquiring historical population information of each node in the city node network
  • An information calculation module configured to calculate the historical population information of each node through a graph neural network to obtain the spatial characteristics and time series characteristics of each node;
  • a vector generating module configured to generate the point embedding vector of each node according to the spatial feature and the time series feature
  • the information generating module is configured to generate population flow information based on the point embedding vector; wherein the population flow information is used as a node connection line to connect the nodes.
  • an embodiment of the present application further provides a computer device, including a memory and a processor, the memory stores computer-readable instructions, and the processor implements the following steps when executing the computer-readable instructions:
  • embodiments of the present application also provide a computer-readable storage medium, the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions implement the following steps when executed by a processor:
  • the embodiments of the present application mainly have the following beneficial effects: after obtaining the city map, the city map is divided by the city area as a node, and the city area can be flexibly selected, thereby flexibly generating a city node network according to actual applications;
  • the graph neural network can integrate the characteristics of nodes and the interaction between nodes, and input the historical population information of each node into the graph neural network for calculation, and can accurately obtain the spatial characteristics and time series characteristics of each node; the spatial characteristics and time series characteristics are used
  • the point embedding vector is generated, and the point embedding vector is used to generate population flow information, thereby improving the accuracy of the generated population flow prediction.
  • Figure 1 is an exemplary system architecture diagram to which the present application can be applied;
  • Figure 2 is a flowchart of an embodiment of a population flow prediction method based on intelligent decision-making according to the present application
  • FIG. 3 is a flowchart of a specific implementation of step S204 in FIG. 2;
  • FIG. 4 is a flowchart of a specific implementation of step S2041 in FIG. 2;
  • FIG. 5 is a flowchart of a specific implementation manner of step S2042 in FIG. 2;
  • Fig. 6 is a schematic structural diagram of an embodiment of a population flow prediction device based on intelligent decision-making according to the present application
  • Fig. 7 is a schematic structural diagram of an embodiment of a computer device according to the present application.
  • the system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105.
  • the network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105.
  • the network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, and so on.
  • the user can use the terminal devices 101, 102, and 103 to interact with the server 105 through the network 104 to receive or send messages and so on.
  • Various communication client applications such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social platform software, can be installed on the terminal devices 101, 102, and 103.
  • the terminal devices 101, 102, and 103 may be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III, dynamic Video experts compress standard audio layer 3), MP4 (Moving Picture Experts Group Audio Layer IV, dynamic image experts compress standard audio layer 4) players, laptop portable computers and desktop computers, etc.
  • MP3 players Moving Picture Experts Group Audio Layer III, dynamic Video experts compress standard audio layer 3
  • MP4 Moving Picture Experts Group Audio Layer IV, dynamic image experts compress standard audio layer 4
  • laptop portable computers and desktop computers etc.
  • the server 105 may be a server that provides various services, for example, a background server that provides support for pages displayed on the terminal devices 101, 102, and 103.
  • the population flow prediction method based on intelligent decision-making provided by the embodiments of the present application is generally executed by a server.
  • the population flow prediction device based on intelligent decision-making is generally set in the server.
  • terminal devices, networks, and servers in FIG. 1 are merely illustrative. There can be any number of terminal devices, networks, and servers according to implementation needs.
  • FIG. 2 there is shown a flowchart of an embodiment of a method for predicting population flow based on intelligent decision-making according to the present application.
  • the population flow prediction method based on intelligent decision-making includes the following steps:
  • Step S201 Obtain a city map.
  • the electronic device (such as the server shown in FIG. 1) on which the population flow prediction method based on intelligent decision runs can communicate with the terminal or the information storage server through a wired connection or a wireless connection.
  • the above-mentioned wireless connection methods can include, but are not limited to, 3G/4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or developed in the future .
  • the server obtains a city map, and performs population flow prediction on the city corresponding to the city map.
  • the server may receive a map selection instruction from the terminal, and query the city map from the information storage server according to the map identifier in the map selection instruction.
  • the information storage server and the server mentioned in this application may be the same server or different servers.
  • step S202 the city map is divided, and the city area in the city map is used as a node to generate a city node network.
  • the city map may be accompanied by a map description, which records the entities corresponding to each part of the city map.
  • the map description records that a certain part of the city map is an administrative district of a certain city, or a certain building on the city map is a high-speed rail station.
  • Each part recorded in the map description can be regarded as a city area.
  • the server divides the city map according to the map description, and uses the city areas in the city map as nodes to construct a city node network.
  • the city map can be arbitrarily divided according to geographical reality, and the city area can have any shape, not necessarily a regular geometric shape.
  • the server will predict the population flow information, and the population flow information will be used as the right node connection line to connect each node.
  • the server screens urban areas according to map descriptions, and removes urban areas that have little relevance to population flow prediction. For example, if a city area corresponds to several mountains in the city, the flow of people in the city area is relatively small and it has little effect on population flow prediction. Therefore, the city area can be deleted without participating in the construction of the city node network.
  • the map description has recorded which urban areas need to be used for population flow prediction, and the server can directly construct a city node network based on these urban areas.
  • the server after obtaining the city map, obtains the construction instruction from the terminal, divides the city map according to the construction instruction, and selects the city area to construct the city node network.
  • Step S203 Obtain historical population information of each node in the city node network.
  • the historical population information may be the population information of the urban area represented by each node at a preset time point in the past.
  • the server may send the city area and task identifier corresponding to each node in the city node network to the information storage server to obtain historical population information of each node from the information storage server.
  • the task identifier is used to record the specific application of population flow prediction.
  • the population flow prediction can be recorded for transportation construction planning.
  • Transportation construction focuses on real-time changes. Therefore, the time slice for obtaining historical population information is shorter, and the time slice can be 30 minutes;
  • population flow prediction is used in urban construction planning, it is necessary to obtain the population flow law from a long period of time, so the time slice is longer, and the time slice can be six months.
  • the information storage server needs to obtain and store the historical population information of each city area in advance.
  • the information storage server can regularly count and store the population information of each city area through the Internet or the Global Navigation Satellite System (GNSS) to obtain historical population information.
  • GNSS Global Navigation Satellite System
  • the information storage server can obtain the user's geographic location information based on map applications in various mobile terminals, and then sort and count the geographic location information based on the city area to obtain historical population information.
  • the aforementioned historical population information can also be stored in a node of a blockchain.
  • the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • step S204 the historical population information of each node is calculated through the graph neural network to obtain the spatial characteristics and time series characteristics of each node.
  • the spatial feature can be feature data reflecting the spatial distribution of historical population information
  • the time series feature can be feature data reflecting the temporal distribution of historical population information.
  • the server inputs the historical population information of each node as the node feature into the trained graph neural network.
  • the graph neural network calculates and transmits information on the structure of the city node network, integrates the node features of each node, and generates each node The spatial characteristics and time series characteristics of.
  • Graph neural network is a neural network model that directly performs calculation and information transmission on the graph structure.
  • a graph G can be described by the set of vertices V (Vertices) and edges E (Edges) it contains, and the vertices can also be called nodes.
  • the nodes and edges in the graph structure can carry information.
  • each node and each edge can spread its information to the surrounding points and edges through the topological connection of the graph.
  • Each node and each edge will collect the information of the surrounding nodes and edges, and integrate it with its own information.
  • information will be diffused to a certain extent in the network, and each node and each edge will be affected to a certain extent by the surrounding information.
  • the dissemination of map structure information is as follows:
  • f is the way of information integration
  • neighbor(*) is the information of all nodes or edges adjacent to the node*.
  • Step S205 Generate a point embedding vector of each node according to the spatial feature and the time series feature.
  • the point embedding vector is similar to the word embedding vector in natural language processing, which is used to characterize the mapping of the node in the feature space, and integrate the spatio-temporal characteristics of the node.
  • the server After the server obtains the spatial features and time series features of each node, it re-inputs the spatial features and time series features into the graph neural network, and the graph neural network iterates the spatial features and time series features to obtain the point embedding vector of each node.
  • Step S206 generating population flow information based on the point embedding vector; wherein, the population flow information is used as a node connection line to connect each node.
  • the server first generates the edge embedding vector from the point embedding vector, and the method includes: point embedding vector point multiplication, point embedding vector splicing and then performing linear transformation, point edge transformation, and the like. Then activate the edge embedding vector to get the population flow information between each node.
  • the population flow information can be used as a node connecting line to connect each node, and the node connecting line is an abstract representation of physical roads between urban areas in practice.
  • the population flow information includes the direction, which represents the direction of the population flow; it also includes a quantitative factor, which represents the size of the flow of people.
  • the city map is divided by the city area as the node.
  • the city area can be flexibly selected, so as to flexibly generate the city node network according to the actual application;
  • the graph neural network can integrate the characteristics of the nodes and the relationship between the nodes.
  • the historical population information of each node is input into the graph neural network for calculation, and the spatial characteristics and time series characteristics of each node can be accurately obtained; the space characteristics and time series characteristics are used to generate the point embedding vector, and the point embedding vector is used to generate Population movement information, thereby improving the accuracy of the generated population movement predictions.
  • step S201 it may further include: obtaining a training data set; extracting the city node network in the training data set, historical population information corresponding to each node in the city node network, and population flow information corresponding to the city node network; According to the extracted city node network, historical population information and population flow information, the initial graph neural network is trained to obtain the graph neural network.
  • the training data set may be a data set for training the initial graph neural network;
  • the initial graph neural network may be a graph neural network that has not been trained yet.
  • the server needs to first obtain a graph neural network through training.
  • the server first obtains the training data set and extracts from it the city node network, historical population information of each node in the city node network, and population flow information of the city node network.
  • the population flow information is the real population flow data between the urban areas represented by each node.
  • the server takes the city node network and historical population information as the input of the initial graph neural network, and uses the population flow information as the expected output to train the initial graph neural network to obtain the graph neural network.
  • the city node network in the training data set and the historical population information of each node are used as input, and the real population flow information is used as the expected output to train the initial graph neural network to ensure that the trained graph neural network can be based on
  • the change of historical population information calculates population flow information.
  • step S204 may include:
  • step S2041 the historical population information of each node is calculated through the graph neural network to obtain the spatial characteristics of each node.
  • the server inputs the historical population information of each node as the node feature into the graph neural network, and processes it from the spatial dimension first. For each node, the server integrates the node characteristics of the node and other nodes in the city node network with the spatial attention mechanism, and performs graph convolution on the integrated node characteristics to obtain the spatial characteristics of each node.
  • Step S2042 Perform spatio-temporal transformation on the spatial features to obtain the time series features of each node.
  • the server transforms the spatial features in time and space, extracts the time series features of each node, and then adjusts the time series features of each node in conjunction with the time attention mechanism, and finally obtains it for generating The time series feature of the point embedding vector.
  • the spatial dimensionality of historical population information is calculated through the graph neural network to obtain spatial features, and then the temporal dimensionality is calculated to obtain time series features, which realizes orderly processing of historical population information.
  • step S2041 may include:
  • Step S20411 input the historical population information of each node as the node feature into the graph neural network to add spatial attention weight to each node based on the spatial attention mechanism.
  • the server uses the historical population information of each node as the node feature of each node and inputs it into the graph neural network.
  • the graph neural network uses the spatial attention mechanism to integrate the node characteristics and the node characteristics of other nodes in spatial association, and obtain the spatial attention weight of each node.
  • the spatial attention mechanism may adopt the attention mechanism in ASTGCN (Attention Based Spatial-Temporal Graph Convolutional Networks for Traffic Flow Forecasting, a spatiotemporal graph convolutional network model based on the attention mechanism).
  • the calculation method of spatial attention weight is as follows:
  • Is the input of the rth layer of the graph neural network C r-1 is the number of input channels of the rth layer of the graph neural network, and T r-1 is the time dimension of the rth layer of the graph neural network;
  • V s ,b s ⁇ R N ⁇ N as well as Is a learnable network parameter;
  • ⁇ can be a sigmoid function, N can be the number of nodes in a city node network;
  • S is a spatial attention weight matrix, and Si ,j are elements in the spatial attention weight matrix.
  • the node feature is the historical population information of each node; in the hidden layer of the graph neural network, the node feature also includes the effect of the node on the adjacent nodes and edges, that is, the population flow between each other.
  • Step S20412 Perform scaling processing on the node features according to the spatial attention weight.
  • the server scales the node feature according to the spatial attention weight, that is, enlarges or reduces the node feature.
  • Step S20413 Perform spatial graph convolution on the feature of the node after the scaling process to obtain the spatial feature of each node.
  • the server performs spatial graph convolution on the node feature after completing the scaling process on the node feature.
  • the graph neural network is used as a fully connected network, and the node characteristics of each node are integrated with the node characteristics of other nodes, and the distance between nodes is used as the convolution weight. In this way, the spatial characteristics obtained It combines the node characteristics of other nodes.
  • the nodes in the city node network are arranged in an orderly manner, and the location distribution of each node corresponds to the real city area corresponding to the node. Therefore, the distance between the nodes in the city node network is proportional to the real city area corresponding to the node. The distance between.
  • the server may perform Chebyshev convolution on the scaled node features.
  • the historical population information is input into the graph neural network as node features, and the node features are scaled through the spatial attention mechanism to strengthen the node features useful for population flow prediction, and then the spatial graph convolution is used to fuse each node The nodal characteristics of, ensure the accuracy of the spatial characteristics obtained.
  • step S2042 may include:
  • Step S20421 Perform spatio-temporal transformation on the spatial features of each node to obtain time series features.
  • the server After the server obtains the spatial feature, it performs a spatio-temporal transformation on the spatial feature to extract the time series feature. Specifically, it may first perform a ReLU nonlinear transformation on the spatial feature, and then perform a one-dimensional convolution operation in the time dimension, and then perform a ReLU non-linear transformation. Linear transformation, so as to obtain the time series characteristics of each node.
  • the operation of space-time transformation is as follows:
  • g ⁇ *G is the result of spatial graph convolution
  • ⁇ * is the one-dimensional graph convolution in the time dimension
  • ReLU() is the activation function
  • Step S20422 Add time attention weight to the time sequence feature based on the time attention mechanism.
  • the time series features include the time features of nodes at different time points, and the graph neural network adds time attention weights to different time points through the time attention mechanism. For the same node, the time attention weights at different time points can be different or the same.
  • the time attention mechanism is a kind of self-attention mechanism, and the addition of time attention weight can be learned through the training of graph neural network.
  • the temporal attention mechanism may also use the attention mechanism in ASTGCN (Attention Based Spatial-Temporal Graph Convolutional Networks for Traffic Flow Forecasting, a spatio-temporal graph convolutional network model based on the attention mechanism).
  • the time attention weight is as follows:
  • Tr -1 is the time dimension of the rth layer of the graph neural network
  • U 1 ⁇ R N as well as Is a learnable network parameter
  • ⁇ can be a sigmoid function
  • N can be the number of nodes in a city node network
  • E is a time attention weight matrix
  • E i,j are elements in a space attention weight matrix.
  • Step S20423 Iterate the time series features according to the added time attention weights to obtain the time series features after the iteration.
  • the obtained time attention weight will act on the time series feature of the corresponding node, so as to iteratively update the time series feature.
  • the graph neural network performs two iterations on the time series feature according to the time attention weight.
  • time attention weight is added to the time series feature, and the time series feature is updated iteratively, thereby strengthening the time points related to population flow prediction, and improving the extraction The accuracy of the time series features.
  • step S205 may include: inputting the spatial characteristics and time series characteristics of each node into the graph neural network; obtaining the output of the preset hidden layer in the graph neural network as the point embedding vector of each node.
  • the server inputs the convolved spatial features of the spatial graph and the iterated time series features into the graph neural network for iteration, and extracts the output of the layer from the preset hidden layer to obtain the point embedding vector of each node.
  • the point embedding vector combines the time series characteristics and spatial characteristics of the node.
  • the spatial feature and the time sequence feature are input to the graph neural network for iteration, so as to obtain the point embedding vector of each node from the preset hidden layer.
  • step S206 may include: respectively obtaining the starting point feature and the end point feature of each node based on the point embedding vector; taking the city node network as a fully connected network, and performing a point multiplication operation according to the starting point feature and end point feature of each node to generate inter-node
  • the edge embedding vector of; the edge embedding vector is activated by the activation function to generate population flow information.
  • the starting point feature can be the feature quantity of the reaction node as the starting point of population flow
  • the end feature can be the feature quantity of the reaction node being the end of population flow.
  • the server inputs the point embedding vector into the start point feature extraction network and the end point feature extraction network respectively to extract the feature value of the node as the starting point of the population flow and the feature value of the end point of the population flow to obtain the starting point and the end feature.
  • the start point feature extraction network and the end point feature extraction network may be a three-layer fully connected network.
  • the server When predicting the population flow information from A to B between two nodes A and B, the server extracts the starting point feature of node A, extracts the end point feature of node B, and performs dot multiplication on the starting point feature of node A and the end point feature of node B ,
  • the edge embedding vector from node A to node B is obtained, and the edge embedding vector is activated by the activation function to obtain the population flow information from node A to node B.
  • the activation function may be a tanh function.
  • the server regards the city node network as a fully connected network, performs pairwise calculations on the nodes, and obtains the population flow information between each two nodes.
  • the population flow information is used as a node connection line to connect each node.
  • population flow information includes both size and direction. The size represents the size of the population flow between nodes, and the direction represents the direction of population flow between nodes.
  • the edge embedding vector between nodes is generated on the basis of the point embedding vector, and the population flow information can be obtained by activating the edge embedding vector.
  • the city node network is calculated as a fully connected network to ensure The information on population movement is consistent with reality.
  • the population flow prediction method based on intelligent decision-making in this application involves neural networks, machine learning, and predictive analysis in the field of artificial intelligence.
  • This application can be applied to smart transportation in the field of smart cities, so as to promote the construction of smart cities.
  • population flow information can be used for traffic command, city planning, etc.
  • the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through computer-readable instructions, which can be stored in a computer-readable storage medium.
  • the computer-readable instructions When executed, they may include the processes of the above-mentioned method embodiments.
  • the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.
  • this application provides an embodiment of a population flow prediction device based on intelligent decision-making, and the device embodiment corresponds to the method embodiment shown in FIG. 2 ,
  • the device can be specifically applied to various electronic equipment.
  • the population flow prediction device 300 based on intelligent decision-making in this embodiment includes: a map acquisition module 301, a map division module 302, an information acquisition module 303, an information calculation module 304, a vector generation module 305, and information generation Module 306, where:
  • the map acquisition module 301 is used to acquire a city map.
  • the map division module 302 is used to divide the city map, and use the city areas in the city map as nodes to generate a city node network.
  • the information acquisition module 303 is used to acquire historical population information of each node in the city node network.
  • the information calculation module 304 is used to calculate the historical population information of each node through the graph neural network to obtain the spatial characteristics and time series characteristics of each node.
  • the vector generating module 305 is used to generate the point embedding vector of each node according to the spatial feature and the time series feature.
  • the information generating module 306 is used for generating population flow information based on the point embedding vector; wherein, the population flow information is used as a node connection line to connect each node.
  • the city map is divided by the city area as the node.
  • the city area can be flexibly selected, so as to flexibly generate the city node network according to the actual application;
  • the graph neural network can integrate the characteristics of the nodes and the relationship between the nodes.
  • the historical population information of each node is input into the graph neural network for calculation, and the spatial characteristics and time series characteristics of each node can be accurately obtained; the space characteristics and time series characteristics are used to generate the point embedding vector, and the point embedding vector is used to generate Population movement information, thereby improving the accuracy of the generated population movement predictions.
  • the device 300 for predicting population flow based on intelligent decision-making further includes: an acquisition module, an extraction module, and a training module, where:
  • the acquisition module is used to acquire the training data set.
  • the extraction module is used to extract the city node network in the training data set, the historical population information corresponding to each node in the city node network, and the population flow information corresponding to the city node network.
  • the training module is used to train the initial graph neural network according to the extracted city node network, historical population information and population flow information to obtain the graph neural network.
  • the city node network in the training data set and the historical population information of each node are used as input, and the real population flow information is used as the expected output to train the initial graph neural network to ensure that the trained graph neural network can be based on
  • the change of historical population information calculates population flow information.
  • the information calculation module 304 includes: an information calculation sub-module and a spatio-temporal transformation sub-module, where:
  • the information calculation sub-module is used to calculate the historical population information of each node through the graph neural network to obtain the spatial characteristics of each node.
  • the spatio-temporal transformation sub-module is used to perform spatio-temporal transformation on the spatial features to obtain the time series features of each node.
  • the spatial dimensionality of historical population information is calculated through the graph neural network to obtain spatial features, and then the temporal dimensionality is calculated to obtain time series features, which realizes orderly processing of historical population information.
  • the information calculation sub-module includes: a spatial weight adding unit, a feature processing unit, and a feature convolution unit, where:
  • the spatial weight adding unit is used to input the historical population information of each node as the node feature into the graph neural network to add spatial attention weight to each node based on the spatial attention mechanism.
  • the feature processing unit is used to perform scaling processing on node features according to the spatial attention weight.
  • the feature convolution unit is used to perform spatial graph convolution on the feature of the node after the scaling process to obtain the spatial feature of each node.
  • the historical population information is input into the graph neural network as node features, and the node features are scaled through the spatial attention mechanism to strengthen the node features useful for population flow prediction, and then the spatial graph convolution is used to fuse each node The nodal characteristics of, ensure the accuracy of the spatial characteristics obtained.
  • the spatio-temporal transformation sub-module includes: a feature transformation unit, a time weight adding unit, and a feature iteration unit, where:
  • the feature transformation unit is used to perform spatio-temporal transformation on the spatial features of each node to obtain time series features.
  • the time weight adding unit is used to add time attention weight to the time series feature based on the time attention mechanism.
  • the feature iteration unit is used to iterate the time series features according to the added time attention weights to obtain the time series features after the iteration.
  • time attention weight is added to the time series feature, and the time series feature is updated iteratively, thereby strengthening the time points related to population flow prediction, and improving the extraction The accuracy of the time series features.
  • the vector generation module 305 includes: a feature input sub-module and an output acquisition sub-module, where:
  • the feature input sub-module is used to input the spatial features and time series features of each node into the graph neural network.
  • the output obtaining sub-module is used to obtain the output of the preset hidden layer in the graph neural network as the point embedding vector of each node.
  • the spatial feature and the time sequence feature are input to the graph neural network for iteration, so as to obtain the point embedding vector of each node from the preset hidden layer.
  • the information generation module 306 includes: a feature acquisition sub-module, a feature operation sub-module, and a vector activation sub-module, where:
  • the feature acquisition sub-module is used to obtain the starting point feature and the end point feature of each node based on the point embedding vector.
  • the feature operation sub-module is used to take the city node network as a fully connected network, and perform point multiplication operations based on the start point and end point characteristics of each node to generate edge embedding vectors between nodes.
  • the vector activation sub-module is used to activate the edge embedding vector through the activation function to generate population flow information.
  • the edge embedding vector between nodes is generated on the basis of the point embedding vector, and the population flow information can be obtained by activating the edge embedding vector.
  • the city node network is calculated as a fully connected network to ensure The information on population movement is consistent with reality.
  • Fig. 7 is a block diagram of the basic structure of the computer device in this embodiment.
  • the computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are connected to each other in communication via a system bus. It should be pointed out that the figure only shows the computer device 4 with components 41-43, but it should be understood that it is not required to implement all the shown components, and more or fewer components may be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculation and/or information processing in accordance with pre-set or stored instructions.
  • Its hardware includes, but is not limited to, a microprocessor, a dedicated Integrated Circuit (Application Specific Integrated Circuit, ASIC), Programmable Gate Array (Field-Programmable Gate Array, FPGA), Digital Processor (Digital Signal Processor, DSP), embedded equipment, etc.
  • ASIC Application Specific Integrated Circuit
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • DSP Digital Processor
  • the computer device may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server.
  • the computer device can interact with the user through a keyboard, a mouse, a remote control, a touch panel, or a voice control device.
  • the memory 41 includes at least one type of computer-readable storage medium.
  • the computer-readable storage medium may be nonvolatile or volatile.
  • the computer-readable storage medium includes flash memory, hard disk, and multimedia card. , Card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), Programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc.
  • the memory 41 may be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4.
  • the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Media Card, SMC), and a secure digital (Secure Digital, SD) card, Flash Card, etc.
  • the memory 41 may also include both the internal storage unit of the computer device 4 and its external storage device.
  • the memory 41 is generally used to store an operating system and various application software installed in the computer device 4, such as computer-readable instructions of a population flow prediction method based on intelligent decision-making.
  • the memory 41 can also be used to temporarily store various types of data that have been output or will be output.
  • the processor 42 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments.
  • the processor 42 is generally used to control the overall operation of the computer device 4.
  • the processor 42 is configured to run computer-readable instructions or process data stored in the memory 41, for example, run the computer-readable instructions of the method for predicting population flow based on intelligent decision-making.
  • the network interface 43 may include a wireless network interface or a wired network interface, and the network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.
  • the computer device provided in this embodiment can execute the above-mentioned method for predicting population flow based on intelligent decision-making.
  • the population flow prediction method based on intelligent decision-making herein may be the population flow prediction method based on intelligent decision-making in each of the above embodiments.
  • the city map is divided by the city area as the node.
  • the city area can be flexibly selected, so as to flexibly generate the city node network according to the actual application;
  • the graph neural network can integrate the characteristics of the nodes and the relationship between the nodes.
  • the historical population information of each node is input into the graph neural network for calculation, and the spatial characteristics and time series characteristics of each node can be accurately obtained; the space characteristics and time series characteristics are used to generate the point embedding vector, and the point embedding vector is used to generate Population movement information, thereby improving the accuracy of the generated population movement predictions.
  • the present application also provides another implementation manner, that is, a computer-readable storage medium is provided with computer-readable instructions stored thereon, and the computer-readable instructions can be executed by at least one processor to The at least one processor is made to execute the steps of the population flow prediction method based on intelligent decision-making as described above.
  • the city map is divided by the city area as the node.
  • the city area can be flexibly selected, so as to flexibly generate the city node network according to the actual application;
  • the graph neural network can integrate the characteristics of the nodes and the relationship between the nodes.
  • the historical population information of each node is input into the graph neural network for calculation, and the spatial characteristics and time series characteristics of each node can be accurately obtained; the space characteristics and time series characteristics are used to generate the point embedding vector, and the point embedding vector is used to generate Population movement information, thereby improving the accuracy of the generated population movement predictions.
  • the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes several instructions to make a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods described in the various embodiments of the present application.
  • a terminal device which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.

Abstract

一种基于智能决策的人口流动预测方法、装置、计算机设备及存储介质，该方法包括获取城市地图（S201）；对城市地图进行划分，以城市地图中的城市区域为节点，生成城市节点网络（S202）；获取城市节点网络中各节点的历史人口信息（S203）；通过图神经网络对各节点的历史人口信息进行计算，得到各节点的空间特征以及时间序列特征（S204）；根据空间特征以及时间序列特征生成各节点的点嵌入向量（S205）；基于点嵌入向量生成人口流动信息；其中，人口流动信息作为节点连接线连接各节点（S206）。该方法提高了人口流动预测的准确性。

Description

基于智能决策的人口流动预测方法、装置及计算机设备
本申请要求于2020年09月18日提交中国专利局、申请号为202010988205.0，发明名称为“基于智能决策的人口流动预测方法、装置及计算机设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域，尤其涉及一种基于智能决策的人口流动预测方法、装置、计算机设备及存储介质。
背景技术
随着计算机技术的发展，计算机在城市人口流动研究中的应用也越来越广泛。城市人口流动研究具有很高的实用价值，城市人口流动反应城市人口的群体动态性特征，在城市规划、交通建设等领域具有重要的指导作用。
发明人意识到，传统的人口流动预测技术，通常是对某个区域进行流入流出的人口流动预测，不包含方向性；或者直接依据重力模型或辐射模型对某个区域的人口流动信息进行预测，因为模型较为简单，使得人口流动预测的准确性较低。
发明内容
本申请实施例的目的在于提出一种基于智能决策的人口流动预测方法、装置、计算机设备及存储介质，以解决人口流动预测准确性较低的问题。
为了解决上述技术问题，本申请实施例提供一种基于智能决策的人口流动预测方法，采用了如下所述的技术方案：
获取城市地图；
对所述城市地图进行划分，以所述城市地图中的城市区域为节点，生成城市节点网络；
获取所述城市节点网络中各节点的历史人口信息；
通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征；
根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量；
基于所述点嵌入向量生成人口流动信息；其中，所述人口流动信息作为节点连接线连接所述各节点。
为了解决上述技术问题，本申请实施例还提供一种基于智能决策的人口流动预测装置，采用了如下所述的技术方案：
地图获取模块，用于获取城市地图；
地图划分模块，用于对所述城市地图进行划分，以所述城市地图中的城市区域为节点，生成城市节点网络；
信息获取模块，用于获取所述城市节点网络中各节点的历史人口信息；
信息计算模块，用于通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征；
向量生成模块，用于根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量；
信息生成模块，用于基于所述点嵌入向量生成人口流动信息；其中，所述人口流动信息作为节点连接线连接所述各节点。
为了解决上述技术问题，本申请实施例还提供一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：
获取城市地图；
对所述城市地图进行划分，以所述城市地图中的城市区域为节点，生成城市节点网络；
获取所述城市节点网络中各节点的历史人口信息；
通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征；
根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量；
基于所述点嵌入向量生成人口流动信息；其中，所述人口流动信息作为节点连接线连接所述各节点。
为了解决上述技术问题，本申请实施例还提供一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如下步骤：
获取城市地图；
对所述城市地图进行划分，以所述城市地图中的城市区域为节点，生成城市节点网络；
获取所述城市节点网络中各节点的历史人口信息；
通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征；
根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量；
基于所述点嵌入向量生成人口流动信息；其中，所述人口流动信息作为节点连接线连接所述各节点。
与现有技术相比，本申请实施例主要有以下有益效果：获取城市地图后，以城市区域为节点对城市地图进行划分，城市区域可以灵活选取，从而根据实际应用灵活地生成城市节点网络；图神经网络可以整合节点特征以及节点之间的相互作用，将各节点的历史人口信息输入图神经网络进行计算，可以准确地得到各节点的空间特征以及时间序列特征；空间特征和时间序列特征用以生成点嵌入向量，点嵌入向量用于生成人口流动信息，从而提高了生成的人口流动预测的准确性。
附图说明
为了更清楚地说明本申请中的方案，下面将对本申请实施例描述中所需要使用的附图作一个简单介绍，显而易见地，下面描述中的附图是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本申请可以应用于其中的示例性系统架构图；
图2是根据本申请的基于智能决策的人口流动预测方法的一个实施例的流程图；
图3是图2中步骤S204的一种具体实施方式的流程图；
图4是图2中步骤S2041的一种具体实施方式的流程图；
图5是图2中步骤S2042的一种具体实施方式的流程图；
图6是根据本申请的基于智能决策的人口流动预测装置的一个实施例的结构示意图；
图7是根据本申请的计算机设备的一个实施例的结构示意图。
具体实施方式
除非另有定义，本文所使用的所有的技术和科学术语与属于本申请的技术领域的技术人员通常理解的含义相同；本文中在申请的说明书中所使用的术语只是为了描述具体的实施例的目的，不是旨在于限制本申请；本申请的说明书和权利要求书及上述附图说明中的术语“包括”和“具有”以及它们的任何变形，意图在于覆盖不排他的包含。本申请的说明书和权利要求书或上述附图中的术语“第一”、“第二”等是用于区别不同对象，而不是用于描述特定顺序。
在本文中提及“实施例”意味着，结合实施例描述的特定特征、结构或特性可以包含 在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例，也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是，本文所描述的实施例可以与其它实施例相结合。
为了使本技术领域的人员更好地理解本申请方案，下面将结合附图，对本申请实施例中的技术方案进行清楚、完整地描述。
如图1所示，系统架构100可以包括终端设备101、102、103，网络104和服务器105。网络104用以在终端设备101、102、103和服务器105之间提供通信链路的介质。网络104可以包括各种连接类型，例如有线、无线通信链路或者光纤电缆等等。
用户可以使用终端设备101、102、103通过网络104与服务器105交互，以接收或发送消息等。终端设备101、102、103上可以安装有各种通讯客户端应用，例如网页浏览器应用、购物类应用、搜索类应用、即时通信工具、邮箱客户端、社交平台软件等。
终端设备101、102、103可以是具有显示屏并且支持网页浏览的各种电子设备，包括但不限于智能手机、平板电脑、电子书阅读器、MP3播放器(Moving Picture Experts Group Audio Layer III，动态影像专家压缩标准音频层面3)、MP4(Moving Picture Experts Group Audio Layer IV，动态影像专家压缩标准音频层面4)播放器、膝上型便携计算机和台式计算机等等。
服务器105可以是提供各种服务的服务器，例如对终端设备101、102、103上显示的页面提供支持的后台服务器。
需要说明的是，本申请实施例所提供的基于智能决策的人口流动预测方法一般由服务器执行，相应地，基于智能决策的人口流动预测装置一般设置于服务器中。
应该理解，图1中的终端设备、网络和服务器的数目仅仅是示意性的。根据实现需要，可以具有任意数目的终端设备、网络和服务器。
继续参考图2，示出了根据本申请的基于智能决策的人口流动预测方法的一个实施例的流程图。所述的基于智能决策的人口流动预测方法，包括以下步骤：
步骤S201，获取城市地图。
在本实施例中，基于智能决策的人口流动预测方法运行于其上的电子设备(例如图1所示的服务器)可以通过有线连接方式或者无线连接方式与终端或信息存储服务器进行通信。需要指出的是，上述无线连接方式可以包括但不限于3G/4G连接、WiFi连接、蓝牙连接、WiMAX连接、Zigbee连接、UWB(ultra wideband)连接、以及其他现在已知或将来开发的无线连接方式。
具体地，服务器获取城市地图，对城市地图所对应的城市进行人口流动预测。服务器可以从终端接收地图选取指令，根据地图选取指令中的地图标识从信息存储服务器中查询城市地图。信息存储服务器与本申请中提到的服务器可以是同一个服务器，也可以是不同的服务器。
步骤S202，对城市地图进行划分，以城市地图中的城市区域为节点，生成城市节点网络。
具体地，城市地图可以附带地图说明，地图说明记录了城市地图中每一部分所对应的实体。例如，地图说明记录了城市地图中某一部分为某市的一个行政区，或城市地图中的某栋建筑为高铁站。地图说明中记录的每一部分可视作一个城市区域。
服务器依据地图说明对城市地图进行划分，以城市地图中的城市区域作为节点，构建出城市节点网络。可以依据地理实际对城市地图进行任意的划分，城市区域可以具有任意形状，并非一定为规则的几何形状。
当前构建出的城市节点网络的节点间并无节点连接线，服务器将预测人口流动信息，人口流动信息作为有权的节点连接线连接各节点。
在一个实施例中，服务器根据地图说明对城市区域进行筛选，去除与人口流动预测相关性不大的城市区域。例如，一个城市区域对应城市中的几座山，则该城市区域人流量较 少，对人口流动预测作用不大，因此可以将该城市区域予以删除，不参与城市节点网络的构建。
在一个实施例中，地图说明已经记录了进行人口流动预测时需要使用哪些城市区域，服务器可直接根据这些城市区域构建城市节点网络。
在一个实施例中，服务器获取到城市地图后，从终端获取构建指令，依据构建指令对城市地图进行划分，并选取城市区域构建城市节点网络。
步骤S203，获取城市节点网络中各节点的历史人口信息。
其中，历史人口信息可以是各节点所代表的城市区域，在过去的预设时间点所具有的人口信息。
具体地，服务器可以将城市节点网络中的各节点所对应的城市区域以及任务标识发送至信息存储服务器，以从信息存储服务器获取各节点的历史人口信息。
任务标识用以记录人口流动预测的具体应用，例如可以记录该次人口流动预测用于交通建设规划，交通建设侧重实时变化，因此获取历史人口信息时时间切片较短，时间切片可以是30分钟；当人口流动预测用于城市建设规划时，需要从一段较长的时间段内获取人口流动规律，因此时间切片较长，时间切片可以是六个月。
信息存储服务器需要预先获取并存储各城市区域的历史人口信息。信息存储服务器可以通过互联网或全球导航卫星系统(Global Navigation Satellite System，GNSS)定时统计各城市区域的人口信息并存储，得到历史人口信息。例如，信息存储服务器可以基于各种移动终端中的地图应用获取用户的地理位置信息，再根据城市区域对统计到的地理位置信息进行整理统计，从而得到历史人口信息。
需要强调的是，为进一步保证上述历史人口信息的私密和安全性，上述历史人口信息还可以存储于一区块链的节点中。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
步骤S204，通过图神经网络对各节点的历史人口信息进行计算，得到各节点的空间特征以及时间序列特征。
其中，空间特征可以是反应历史人口信息空间分布的特征数据；时间序列特征可以是反应历史人口信息时间分布的特征数据。
具体地，服务器将各节点的历史人口信息作为节点特征输入训练完毕的图神经网络，图神经网络在城市节点网络的结构上进行计算和信息传递，对各节点的节点特征进行整合，生成各节点的空间特征以及时间序列特征。
图神经网络是直接在图结构上进行计算和信息传递的神经网络模型。一个图G(Graph)可以用它包含的顶点V(Vertices)和边E(Edges)的集合来描述，其中顶点又可称为节点。图结构中的节点和边都可以携带信息，在每一层神经网络的计算中，每个节点和每条边可以通过图的拓扑连接把自己的信息扩散到周围的点和边上。每个节点和每条边会收集到周围的节点和边的信息，并与自身的信息进行整合。经过多层的神经网络计算，信息就会在网络中进行一定程度的扩散，每个节点和每条边就会被周围的信息在一定程度上影响。图结构信息传播的如下：
Figure PCTCN2020124428-appb-000001
其中，f为信息整合的方式，
Figure PCTCN2020124428-appb-000002
为图神经网络中第l层第i个点的特征向量，neighbor(*)为所有与节点*相邻的节点或边的信息。
步骤S205，根据空间特征以及时间序列特征生成各节点的点嵌入向量。
其中，点嵌入向量与自然语言处理中的词嵌入向量相似，用于表征节点在特征空间的 映射，整合了节点的时空特征。
服务器得到各节点的空间特征以及时间序列特征后，将空间特征和时间序列特征重新输入图神经网络，由图神经网络对空间特征和时间序列特征进行迭代，从而得到各节点的点嵌入向量。
步骤S206，基于点嵌入向量生成人口流动信息；其中，人口流动信息作为节点连接线连接各节点。
具体地，服务器先由点嵌入向量生成边嵌入向量，方法包括：点嵌入向量点乘、点嵌入向量拼接后再进行线性变换、点边变换等。再对边嵌入向量进行激活，得到各节点之间的人口流动信息。
人口流动信息可以作为节点连接线连接各节点，节点连接线是实际中各城市区域之间物理道路的抽象表示。人口流动信息包括方向，代表人口流动的方向；还包括数量因子，数量因子的大小表征了人流量的大小。
本实施例中，获取城市地图后，以城市区域为节点对城市地图进行划分，城市区域可以灵活选取，从而根据实际应用灵活地生成城市节点网络；图神经网络可以整合节点特征以及节点之间的相互作用，将各节点的历史人口信息输入图神经网络进行计算，可以准确地得到各节点的空间特征以及时间序列特征；空间特征和时间序列特征用以生成点嵌入向量，点嵌入向量用于生成人口流动信息，从而提高了生成的人口流动预测的准确性。
进一步的，在上述步骤S201之前还可以包括：获取训练数据集；提取训练数据集中的城市节点网络、城市节点网络中各节点所对应的历史人口信息以及与城市节点网络所对应的人口流动信息；根据提取到的城市节点网络、历史人口信息以及人口流动信息训练初始图神经网络，得到图神经网络。
其中，训练数据集可以是对初始图神经网络进行训练的数据集；初始图神经网络可以是尚未完成训练的图神经网络。
具体地，在进行人口流动预测之前，服务器需要先通过训练得到图神经网络。服务器先获取训练数据集并从中提取城市节点网络、城市节点网络中各节点的历史人口信息以及城市节点网络的人口流动信息。其中，人口流动信息是各节点所代表的城市区域间真实的人口流动数据。
服务器将城市节点网络和历史人口信息作为初始图神经网络的输入，将人口流动信息作为期望输出训练初始图神经网络，从而得到图神经网络。
本实施例中，将训练数据集中的城市节点网络和各节点的历史人口信息作为输入，将真实的人口流动信息作为期望输出对初始图神经网络进行训练，保证了训练完毕的图神经网络可以依据历史人口信息的变化计算出人口流动信息。
进一步的，如图3所示，上述步骤S204可以包括：
步骤S2041，通过图神经网络对各节点的历史人口信息进行计算，得到各节点的空间特征。
具体地，服务器将各节点的历史人口信息作为节点特征输入图神经网络，先从空间维度进行处理。对于每一个节点，服务器结合空间注意力机制对该节点及城市节点网络中其他各节点的节点特征进行整合，并对整合后的节点特征进行图卷积，得到各节点的空间特征。
步骤S2042，对空间特征进行时空变换，得到各节点的时间序列特征。
具体地，得到空间特征后再从时间维度进行处理，服务器对空间特征进行时空转换，提取到各节点的时间序列特征，然后结合时间注意力机制调整各节点的时间序列特征，最后得到用于生成点嵌入向量的时间序列特征。
本实施例中，先通过图神经网络对历史人口信息进行空间维度的计算得到空间特征，再进行时间维度的计算得到时间序列特征，实现了对历史人口信息的有序处理。
进一步的，如图4所示，上述步骤S2041可以包括：
步骤S20411，将各节点的历史人口信息作为节点特征输入图神经网络，以基于空间注意力机制给各节点添加空间注意力权重。
具体地，服务器将各节点的历史人口信息作为各节点的节点特征，输入图神经网络。对于每一个节点，图神经网络使用空间注意力机制对节点特征以及其他节点的节点特征进行空间关联上的整合，得到各节点的空间注意力权重。在一个实施例中，空间注意力机制可以采用ASTGCN(Attention Based Spatial-Temporal Graph Convolutional Networks for Traffc Flow Forecasting，基于注意力机制的时空图卷积网络模型)中的注意力机制。
其中，空间注意力权重的计算方式如下：
Figure PCTCN2020124428-appb-000003
Figure PCTCN2020124428-appb-000004
其中，
Figure PCTCN2020124428-appb-000005
是图神经网络第r层的输入，C r-1是图神经网络第r层的输入通道数，T r-1是图神经网络第r层的时间维度；V s,b s∈R N×N、
Figure PCTCN2020124428-appb-000006
Figure PCTCN2020124428-appb-000007
以及
Figure PCTCN2020124428-appb-000008
是可学习的网络参数；σ可以是sigmoid函数，N可以是城市节点网络中的节点数；S是空间注意力权重矩阵，S i,j是空间注意力权重矩阵中的元素。
在图神经网络的输入层，节点特征即为各节点的历史人口信息；在图神经网络的隐藏层中，节点特征还包含了节点对相邻节点和边的作用，即相互间的人口流动。
步骤S20412，根据空间注意力权重对节点特征进行放缩处理。
具体地，服务器得到各节点的空间注意力权重后，根据空间注意力权重对节点特征进行放缩，即放大或缩小节点特征。
步骤S20413，对放缩处理后的节点特征进行空间图卷积，得到各节点的空间特征。
具体地，服务器对节点特征完成放缩处理后，再对节点特征进行空间图卷积。在进行空间图卷积时，将图神经网络作为全连通网络，将各节点的节点特征与其他节点的节点特征进行整合，并将节点之间的距离作为卷积权重，如此，得到的空间特征便结合了其他各节点的节点特征。
城市节点网络中各节点是有序排布的，各节点的位置分布与节点所对应的真实城市区域相对应，因此城市节点网络中各节点之间的距离正比于节点所对应的真实城市区域之间的距离。
在一个实施例中，服务器可以对放缩后的节点特征进行切比雪夫卷积。
本实施例中，将历史人口信息作为节点特征输入图神经网络，并通过空间注意力机制对节点特征进行放缩，以强化对人口流动预测有用的节点特征，再通过空间图卷积融合各节点的节点特征，保证了得到的空间特征的准确性。
进一步的，如图5所示，上述步骤S2042可以包括：
步骤S20421，对各节点的空间特征进行时空变换，得到时间序列特征。
具体地，服务器得到空间特征后，对空间特征进行时空变换以提取时间序列特征，具体可以是先对空间特征进行ReLU非线性变换，然后在时间维度上进行一维卷积操作，再进行ReLU非线性变换，从而得到各节点的时间序列特征。时空变换的运算如下：
Figure PCTCN2020124428-appb-000009
其中，g θ*G为空间图卷积的结果，Φ*为时间维度上的一维图卷积，ReLU()为激活函数。
步骤S20422，基于时间注意力机制给时间序列特征添加时间注意力权重。
具体地，时间序列特征包含了节点在不同时间点的时间特征，图神经网络通过时间注意力机制给不同的时间点添加时间注意力权重。对于同一个节点，不同时间点的时间注意力权重可以不同，也可以相同。
时间注意力机制属于自注意力机制的一种，时间注意力权重的添加可以通过图神经网络的训练学习到。在一个实施例中，时间注意力机制也可以采用ASTGCN(Attention Based Spatial-Temporal Graph Convolutional Networks for Traffc Flow Forecasting，基于注意力机制的时空图卷积网络模型)中的注意力机制。
在一个实施例中，时间注意力权重如下：
Figure PCTCN2020124428-appb-000010
Figure PCTCN2020124428-appb-000011
其中，
Figure PCTCN2020124428-appb-000012
是图神经网络第r层的输入，T r-1是图神经网络第r层的时间维度；
Figure PCTCN2020124428-appb-000013
U 1∈R N、
Figure PCTCN2020124428-appb-000014
以及
Figure PCTCN2020124428-appb-000015
是可学习的网络参数；σ可以是sigmoid函数，N可以是城市节点网络中的节点数；E是时间注意力权重矩阵，E i,j是空间注意力权重矩阵中的元素。
步骤S20423，根据添加的时间注意力权重对时间序列特征进行迭代，得到迭代完毕的时间序列特征。
具体的，得到的时间注意力权重将作用于对应节点的时间序列特征，从而对时间序列特征进行迭代更新。在一个实施例中，图神经网络依据时间注意力权重对时间序列特征进行两次迭代。
本实施例中，通过时空变换得到时间序列特征后，给时间序列特征添加时间注意力权重，并对时间序列特征进行迭代更新，从而对与人口流动预测相关的时间点进行强化，提高了提取到的时间序列特征的准确性。
进一步的，上述步骤S205可以包括：将各节点的空间特征以及时间序列特征输入图神经网络；获取图神经网络中预设隐藏层的输出，作为各节点的点嵌入向量。
具体地，服务器将空间图卷积后的空间特征以及迭代后的时间序列特征，输入图神经网络再进行迭代，并从预设的隐藏层提取该层的输出，得到各节点的点嵌入向量。点嵌入向量融合了该节点的时间序列特征以及空间特征。
本实施例中，将空间特征和时间序列特征输入图神经网络进行迭代，从而实现从预设的隐藏层获取各节点的点嵌入向量。
进一步的，上述步骤S206可以包括：基于点嵌入向量分别获取各节点的起点特征和终点特征；将城市节点网络作为全连通网络，根据各节点的起点特征和终点特征进行点乘运算，生成节点间的边嵌入向量；通过激活函数激活边嵌入向量，生成人口流动信息。
其中，起点特征可以是反应节点作为人口流动起点的特征量；终点特征可以是反应节点作为人口流动终点的特征量。
具体地，服务器将点嵌入向量分别输入起点特征提取网络以及终点特征提取网络，以提取节点作为人口流动起点的特征量以及作为人口流动终点的特征量，得到起点特征以及终点特征。
在一个实施例中，起点特征提取网络以及终点特征提取网络可以是三层全连接网络。
预测两个节点A、B之间从A到B的人口流动信息时，服务器提取节点A的起点特征，提取节点B的终点特征，对节点A的起点特征和节点B的终点特征进行点乘运算，得到从节点A到节点B的边嵌入向量，边嵌入向量再经过激活函数激活，即可得到从节点A到节点B的人口流动信息。激活函数可以是tanh函数。
在实际中，各节点所代表的城市区域之间均可能存在人口流动，因此服务器将城市节点网络作为全连通网络，对节点进行两两计算，分别得到每两个节点之间的人口流动信息。人口流动信息作为节点连接线连接各节点，城市节点网络中，每两个节点之间均存在人口流动信息。人口流动信息既包含大小也包含方向，大小表征了节点间人口流量的大小，方向表征了节点间人口流动的方向。
本实施例中，在点嵌入向量的基础上生成节点之间的边嵌入向量，边嵌入向量激活即可得到人口流动信息，在生成边嵌入向量时将城市节点网络作为全连通网络进行计算，保证了人口流动信息与实际的相符性。
本申请中基于智能决策的人口流动预测方法涉及人工智能领域中的神经网络、机器学习和预测分析。
本申请可应用于智慧城市领域中的智慧交通，从而推动智慧城市建设。例如，人口流动信息可以用于交通指挥，城市规划等。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，该计算机可读指令可存储于一计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等非易失性存储介质，或随机存储记忆体(Random Access Memory，RAM)等。
应该理解的是，虽然附图的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，其可以以其他的顺序执行。而且，附图的流程图中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，其执行顺序也不必然是依次进行，而是可以与其他步骤或者其他步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
进一步参考图6，作为对上述图2所示方法的实现，本申请提供了一种基于智能决策的人口流动预测装置的一个实施例，该装置实施例与图2所示的方法实施例相对应，该装置具体可以应用于各种电子设备中。
如图6所示，本实施例所述的基于智能决策的人口流动预测装置300包括：地图获取模块301、地图划分模块302、信息获取模块303、信息计算模块304、向量生成模块305以及信息生成模块306，其中：
地图获取模块301，用于获取城市地图。
地图划分模块302，用于对城市地图进行划分，以城市地图中的城市区域为节点，生成城市节点网络。
信息获取模块303，用于获取城市节点网络中各节点的历史人口信息。
信息计算模块304，用于通过图神经网络对各节点的历史人口信息进行计算，得到各节点的空间特征以及时间序列特征。
向量生成模块305，用于根据空间特征以及时间序列特征生成各节点的点嵌入向量。
信息生成模块306，用于基于点嵌入向量生成人口流动信息；其中，人口流动信息作为节点连接线连接各节点。
本实施例中，获取城市地图后，以城市区域为节点对城市地图进行划分，城市区域可以灵活选取，从而根据实际应用灵活地生成城市节点网络；图神经网络可以整合节点特征以及节点之间的相互作用，将各节点的历史人口信息输入图神经网络进行计算，可以准确地得到各节点的空间特征以及时间序列特征；空间特征和时间序列特征用以生成点嵌入向量，点嵌入向量用于生成人口流动信息，从而提高了生成的人口流动预测的准确性。
在本实施例的一些可选的实现方式中，基于智能决策的人口流动预测装置300还包括：获取模块、提取模块以及训练模块，其中：
获取模块，用于获取训练数据集。
提取模块，用于提取训练数据集中的城市节点网络、城市节点网络中各节点所对应的历史人口信息以及与城市节点网络所对应的人口流动信息。
训练模块，用于根据提取到的城市节点网络、历史人口信息以及人口流动信息训练初始图神经网络，得到图神经网络。
本实施例中，将训练数据集中的城市节点网络和各节点的历史人口信息作为输入，将 真实的人口流动信息作为期望输出对初始图神经网络进行训练，保证了训练完毕的图神经网络可以依据历史人口信息的变化计算出人口流动信息。
在本申请的一些可选的实现方式中，信息计算模块304包括：信息计算子模块以及时空变换子模块，其中：
信息计算子模块，用于通过图神经网络对各节点的历史人口信息进行计算，得到各节点的空间特征。
时空变换子模块，用于对空间特征进行时空变换，得到各节点的时间序列特征。
本实施例中，先通过图神经网络对历史人口信息进行空间维度的计算得到空间特征，再进行时间维度的计算得到时间序列特征，实现了对历史人口信息的有序处理。
在本申请的一些可选的实现方式中，信息计算子模块包括：空间权重添加单元、特征处理单元以及特征卷积单元，其中：
空间权重添加单元，用于将各节点的历史人口信息作为节点特征输入图神经网络，以基于空间注意力机制给各节点添加空间注意力权重。
特征处理单元，用于根据空间注意力权重对节点特征进行放缩处理。
特征卷积单元，用于对放缩处理后的节点特征进行空间图卷积，得到各节点的空间特征。
本实施例中，将历史人口信息作为节点特征输入图神经网络，并通过空间注意力机制对节点特征进行放缩，以强化对人口流动预测有用的节点特征，再通过空间图卷积融合各节点的节点特征，保证了得到的空间特征的准确性。
在本申请的一些可选的实现方式中，时空变换子模块包括：特征变换单元、时间权重添加单元以及特征迭代单元，其中：
特征变换单元，用于对各节点的空间特征进行时空变换，得到时间序列特征。
时间权重添加单元，用于基于时间注意力机制给时间序列特征添加时间注意力权重。
特征迭代单元，用于根据添加的时间注意力权重对时间序列特征进行迭代，得到迭代完毕的时间序列特征。
本实施例中，通过时空变换得到时间序列特征后，给时间序列特征添加时间注意力权重，并对时间序列特征进行迭代更新，从而对与人口流动预测相关的时间点进行强化，提高了提取到的时间序列特征的准确性。
在本申请的一些可选的实现方式中，向量生成模块305包括：特征输入子模块以及输出获取子模块，其中：
特征输入子模块，用于将各节点的空间特征以及时间序列特征输入图神经网络。
输出获取子模块，用于获取图神经网络中预设隐藏层的输出，作为各节点的点嵌入向量。
本实施例中，将空间特征和时间序列特征输入图神经网络进行迭代，从而实现从预设的隐藏层获取各节点的点嵌入向量。
在本申请的一些可选的实现方式中，信息生成模块306包括：特征获取子模块、特征运算子模块以及向量激活子模块，其中：
特征获取子模块，用于基于点嵌入向量分别获取各节点的起点特征和终点特征。
特征运算子模块，用于将城市节点网络作为全连接网络，根据各节点的起点特征和终点特征进行点乘运算，生成节点间的边嵌入向量。
向量激活子模块，用于通过激活函数激活边嵌入向量，生成人口流动信息。
本实施例中，在点嵌入向量的基础上生成节点之间的边嵌入向量，边嵌入向量激活即可得到人口流动信息，在生成边嵌入向量时将城市节点网络作为全连通网络进行计算，保证了人口流动信息与实际的相符性。
为解决上述技术问题，本申请实施例还提供计算机设备。具体请参阅图7，图7为本 实施例计算机设备基本结构框图。
所述计算机设备4包括通过系统总线相互通信连接存储器41、处理器42、网络接口43。需要指出的是，图中仅示出了具有组件41-43的计算机设备4，但是应理解的是，并不要求实施所有示出的组件，可以替代的实施更多或者更少的组件。其中，本技术领域技术人员可以理解，这里的计算机设备是一种能够按照事先设定或存储的指令，自动进行数值计算和/或信息处理的设备，其硬件包括但不限于微处理器、专用集成电路(Application Specific Integrated Circuit，ASIC)、可编程门阵列(Field－Programmable Gate Array，FPGA)、数字处理器(Digital Signal Processor，DSP)、嵌入式设备等。
所述计算机设备可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。所述计算机设备可以与用户通过键盘、鼠标、遥控器、触摸板或声控设备等方式进行人机交互。
所述存储器41至少包括一种类型的计算机可读存储介质，所述计算机可读存储介质可以是非易失性，也可以是易失性，所述计算机可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。在一些实施例中，所述存储器41可以是所述计算机设备4的内部存储单元，例如该计算机设备4的硬盘或内存。在另一些实施例中，所述存储器41也可以是所述计算机设备4的外部存储设备，例如该计算机设备4上配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。当然，所述存储器41还可以既包括所述计算机设备4的内部存储单元也包括其外部存储设备。本实施例中，所述存储器41通常用于存储安装于所述计算机设备4的操作系统和各类应用软件，例如基于智能决策的人口流动预测方法的计算机可读指令等。此外，所述存储器41还可以用于暂时地存储已经输出或者将要输出的各类数据。
所述处理器42在一些实施例中可以是中央处理器(Central Processing Unit，CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器42通常用于控制所述计算机设备4的总体操作。本实施例中，所述处理器42用于运行所述存储器41中存储的计算机可读指令或者处理数据，例如运行所述基于智能决策的人口流动预测方法的计算机可读指令。
所述网络接口43可包括无线网络接口或有线网络接口，该网络接口43通常用于在所述计算机设备4与其他电子设备之间建立通信连接。
本实施例中提供的计算机设备可以执行上述基于智能决策的人口流动预测方法。此处基于智能决策的人口流动预测方法可以是上述各个实施例的基于智能决策的人口流动预测方法。
本实施例中，获取城市地图后，以城市区域为节点对城市地图进行划分，城市区域可以灵活选取，从而根据实际应用灵活地生成城市节点网络；图神经网络可以整合节点特征以及节点之间的相互作用，将各节点的历史人口信息输入图神经网络进行计算，可以准确地得到各节点的空间特征以及时间序列特征；空间特征和时间序列特征用以生成点嵌入向量，点嵌入向量用于生成人口流动信息，从而提高了生成的人口流动预测的准确性。
本申请还提供了另一种实施方式，即提供一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令可被至少一个处理器执行，以使所述至少一个处理器执行如上述的基于智能决策的人口流动预测方法的步骤。
本实施例中，获取城市地图后，以城市区域为节点对城市地图进行划分，城市区域可以灵活选取，从而根据实际应用灵活地生成城市节点网络；图神经网络可以整合节点特征以及节点之间的相互作用，将各节点的历史人口信息输入图神经网络进行计算，可以准确地得到各节点的空间特征以及时间序列特征；空间特征和时间序列特征用以生成点嵌入向量，点嵌入向量用于生成人口流动信息，从而提高了生成的人口流动预测的准确性。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本申请各个实施例所述的方法。
显然，以上所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例，附图中给出了本申请的较佳实施例，但并不限制本申请的专利范围。本申请可以以许多不同的形式来实现，相反地，提供这些实施例的目的是使对本申请的公开内容的理解更加透彻全面。尽管参照前述实施例对本申请进行了详细的说明，对于本领域的技术人员来而言，其依然可以对前述各具体实施方式所记载的技术方案进行修改，或者对其中部分技术特征进行等效替换。凡是利用本申请说明书及附图内容所做的等效结构，直接或间接运用在其他相关的技术领域，均同理在本申请专利保护范围之内。

Claims (20)

  1. 一种基于智能决策的人口流动预测方法，其中，包括下述步骤：
    获取城市地图；
    对所述城市地图进行划分，以所述城市地图中的城市区域为节点，生成城市节点网络；
    获取所述城市节点网络中各节点的历史人口信息；
    通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征；
    根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量；
    基于所述点嵌入向量生成人口流动信息；其中，所述人口流动信息作为节点连接线连接所述各节点。
  2. 根据权利要求1所述的基于智能决策的人口流动预测方法，其中，在所述获取城市地图的步骤之前还包括：
    获取训练数据集；
    提取所述训练数据集中的城市节点网络、所述城市节点网络中各节点所对应的历史人口信息以及与所述城市节点网络所对应的人口流动信息；
    根据提取到的城市节点网络、历史人口信息以及人口流动信息训练初始图神经网络，得到图神经网络。
  3. 根据权利要求1所述的基于智能决策的人口流动预测方法，其中，所述通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征的步骤包括：
    通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征；
    对所述空间特征进行时空变换，得到所述各节点的时间序列特征。
  4. 根据权利要求3所述的基于智能决策的人口流动预测方法，其中，所述通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征的步骤包括：
    将所述各节点的历史人口信息作为节点特征输入图神经网络，以基于空间注意力机制给所述各节点添加空间注意力权重；
    根据所述空间注意力权重对所述节点特征进行放缩处理；
    对放缩处理后的节点特征进行空间图卷积，得到所述各节点的空间特征。
  5. 根据权利要求3所述的基于智能决策的人口流动预测方法，其中，所述对所述空间特征进行时空变换，得到所述各节点的时间序列特征的步骤包括：
    对所述各节点的空间特征进行时空变换，得到时间序列特征；
    基于时间注意力机制给所述时间序列特征添加时间注意力权重；
    根据添加的时间注意力权重对所述时间序列特征进行迭代，得到迭代完毕的时间序列特征。
  6. 根据权利要求1所述的基于智能决策的人口流动预测方法，其中，所述根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量的步骤包括：
    将所述各节点的空间特征以及时间序列特征输入所述图神经网络；
    获取所述图神经网络中预设隐藏层的输出，作为所述各节点的点嵌入向量。
  7. 根据权利要求1所述的基于智能决策的人口流动预测方法，其中，所述基于所述点嵌入向量生成人口流动信息的步骤包括：
    基于所述点嵌入向量分别获取所述各节点的起点特征和终点特征；
    将所述城市节点网络作为全连通网络，根据所述各节点的起点特征和终点特征进行点乘运算，生成节点间的边嵌入向量；
    通过激活函数激活所述边嵌入向量，生成人口流动信息。
  8. 一种基于智能决策的人口流动预测装置，其中，包括：
    地图获取模块，用于获取城市地图；
    地图划分模块，用于对所述城市地图进行划分，以所述城市地图中的城市区域为节点，生成城市节点网络；
    信息获取模块，用于获取所述城市节点网络中各节点的历史人口信息；
    信息计算模块，用于通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征；
    向量生成模块，用于根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量；
    信息生成模块，用于基于所述点嵌入向量生成人口流动信息；其中，所述人口流动信息作为节点连接线连接所述各节点。
  9. 一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：
    获取城市地图；
    对所述城市地图进行划分，以所述城市地图中的城市区域为节点，生成城市节点网络；
    获取所述城市节点网络中各节点的历史人口信息；
    通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征；
    根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量；
    基于所述点嵌入向量生成人口流动信息；其中，所述人口流动信息作为节点连接线连接所述各节点。
  10. 根据权利要求9所述的计算机设备，其中，所述通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征的步骤包括：
    通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征；
    对所述空间特征进行时空变换，得到所述各节点的时间序列特征。
  11. 根据权利要求10所述的计算机设备，其中，所述通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征的步骤包括：
    将所述各节点的历史人口信息作为节点特征输入图神经网络，以基于空间注意力机制给所述各节点添加空间注意力权重；
    根据所述空间注意力权重对所述节点特征进行放缩处理；
    对放缩处理后的节点特征进行空间图卷积，得到所述各节点的空间特征。
  12. 根据权利要求10所述的计算机设备，其中，所述对所述空间特征进行时空变换，得到所述各节点的时间序列特征的步骤包括：
    对所述各节点的空间特征进行时空变换，得到时间序列特征；
    基于时间注意力机制给所述时间序列特征添加时间注意力权重；
    根据添加的时间注意力权重对所述时间序列特征进行迭代，得到迭代完毕的时间序列特征。
  13. 根据权利要求9所述的计算机设备，其中，所述根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量的步骤包括：
    将所述各节点的空间特征以及时间序列特征输入所述图神经网络；
    获取所述图神经网络中预设隐藏层的输出，作为所述各节点的点嵌入向量。
  14. 根据权利要求9所述的计算机设备，其中，所述基于所述点嵌入向量生成人口流动信息的步骤包括：
    基于所述点嵌入向量分别获取所述各节点的起点特征和终点特征；
    将所述城市节点网络作为全连通网络，根据所述各节点的起点特征和终点特征进行点乘运算，生成节点间的边嵌入向量；
    通过激活函数激活所述边嵌入向量，生成人口流动信息。
  15. 一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机可读指令；其中，所述计算机可读指令被处理器执行时实现如下步骤：
    获取城市地图；
    对所述城市地图进行划分，以所述城市地图中的城市区域为节点，生成城市节点网络；
    获取所述城市节点网络中各节点的历史人口信息；
    通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征；
    根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量；
    基于所述点嵌入向量生成人口流动信息；其中，所述人口流动信息作为节点连接线连接所述各节点。
  16. 根据权利要求15所述的一种计算机可读存储介质，其中，所述通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征以及时间序列特征的步骤包括：
    通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征；
    对所述空间特征进行时空变换，得到所述各节点的时间序列特征。
  17. 根据权利要求16所述的一种计算机可读存储介质，其中，所述通过图神经网络对所述各节点的历史人口信息进行计算，得到所述各节点的空间特征的步骤包括：
    将所述各节点的历史人口信息作为节点特征输入图神经网络，以基于空间注意力机制给所述各节点添加空间注意力权重；
    根据所述空间注意力权重对所述节点特征进行放缩处理；
    对放缩处理后的节点特征进行空间图卷积，得到所述各节点的空间特征。
  18. 根据权利要求16所述的一种计算机可读存储介质，其中，所述对所述空间特征进行时空变换，得到所述各节点的时间序列特征的步骤包括：
    对所述各节点的空间特征进行时空变换，得到时间序列特征；
    基于时间注意力机制给所述时间序列特征添加时间注意力权重；
    根据添加的时间注意力权重对所述时间序列特征进行迭代，得到迭代完毕的时间序列特征。
  19. 根据权利要求15所述的一种计算机可读存储介质，其中，所述根据所述空间特征以及时间序列特征生成所述各节点的点嵌入向量的步骤包括：
    将所述各节点的空间特征以及时间序列特征输入所述图神经网络；
    获取所述图神经网络中预设隐藏层的输出，作为所述各节点的点嵌入向量。
  20. 根据权利要求15所述的一种计算机可读存储介质，其中，所述基于所述点嵌入向量生成人口流动信息的步骤包括：
    基于所述点嵌入向量分别获取所述各节点的起点特征和终点特征；
    将所述城市节点网络作为全连通网络，根据所述各节点的起点特征和终点特征进行点乘运算，生成节点间的边嵌入向量；
    通过激活函数激活所述边嵌入向量，生成人口流动信息。
PCT/CN2020/124428 2020-09-18 2020-10-28 基于智能决策的人口流动预测方法、装置及计算机设备 WO2021174876A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010988205.0A CN112116155B (zh) 2020-09-18 2020-09-18 基于智能决策的人口流动预测方法、装置及计算机设备
CN202010988205.0 2020-09-18

Publications (1)

Publication Number Publication Date
WO2021174876A1 true WO2021174876A1 (zh) 2021-09-10

Family

ID=73801369

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/124428 WO2021174876A1 (zh) 2020-09-18 2020-10-28 基于智能决策的人口流动预测方法、装置及计算机设备

Country Status (2)

Country Link
CN (1) CN112116155B (zh)
WO (1) WO2021174876A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113780662A (zh) * 2021-09-14 2021-12-10 北京航空航天大学 一种流量预测方法、装置、设备和介质
CN114529043A (zh) * 2022-01-05 2022-05-24 深圳大学 一种城市空间组团划分方法
CN115221766A (zh) * 2022-06-15 2022-10-21 南京大学 一种改进辐射模型的跨境人口流动模拟方法
CN115759350A (zh) * 2022-10-25 2023-03-07 清华大学 一种数据稀疏区域的人口流动预测方法及装置
CN116110588A (zh) * 2022-05-10 2023-05-12 北京理工大学 基于动态邻接矩阵和时空注意力的医学时间序列预测方法

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112990530B (zh) * 2020-12-23 2023-12-26 北京软通智慧科技有限公司 区域人口数量预测方法、装置、电子设备和存储介质
CN112541852B (zh) * 2020-12-24 2024-04-12 南方科技大学 城市人流监控方法、装置、电子设备及存储介质
CN112633608B (zh) * 2021-01-06 2022-04-12 南方科技大学 人流转移预测方法、装置、设备及存储介质
CN113342873B (zh) * 2021-05-27 2022-07-05 武汉大学 一种基于城市形态与汇聚模式的人口分析单元划分方法
CN113517076A (zh) * 2021-07-30 2021-10-19 齐鲁工业大学 基于图神经网络与迁移学习的疾病病例数预测方法及系统
CN113744882B (zh) * 2021-09-17 2023-09-19 腾讯科技（深圳）有限公司 目标区域的确定方法、装置、设备以及存储介质
CN117151344B (zh) * 2023-10-26 2024-02-02 乘木科技(珠海)有限公司 一种数字孪生城市人口管理方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107529651A (zh) * 2017-08-18 2018-01-02 北京航空航天大学 一种基于深度学习的城市交通客流预测方法和设备
CN110322064A (zh) * 2019-06-27 2019-10-11 清华大学 城市出行需求预测方法
CN111612206A (zh) * 2020-03-30 2020-09-01 清华大学 一种基于时空图卷积神经网络的街区人流预测方法及系统
CN111612249A (zh) * 2020-05-20 2020-09-01 北京百度网讯科技有限公司 用于预测人流量的方法、装置、设备以及存储介质

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3547226A1 (en) * 2018-03-28 2019-10-02 Koninklijke Philips N.V. Cross-modal neural networks for prediction
CN109754605B (zh) * 2019-02-27 2021-12-07 中南大学 一种基于注意力时态图卷积网络的交通预测方法
CN110675623B (zh) * 2019-09-06 2020-12-01 中国科学院自动化研究所 基于混合深度学习的短时交通流量预测方法、系统、装置
CN110969854A (zh) * 2019-12-13 2020-04-07 深圳先进技术研究院 一种交通流量的预测方法、系统及终端设备
CN111161535B (zh) * 2019-12-23 2021-03-02 山东大学 基于注意力机制的图神经网络交通流量预测方法及系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107529651A (zh) * 2017-08-18 2018-01-02 北京航空航天大学 一种基于深度学习的城市交通客流预测方法和设备
CN110322064A (zh) * 2019-06-27 2019-10-11 清华大学 城市出行需求预测方法
CN111612206A (zh) * 2020-03-30 2020-09-01 清华大学 一种基于时空图卷积神经网络的街区人流预测方法及系统
CN111612249A (zh) * 2020-05-20 2020-09-01 北京百度网讯科技有限公司 用于预测人流量的方法、装置、设备以及存储介质

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
CAI LE, GUO HAOTIAN: "Prediction of Population Flow Based on Convolutional Neural Network", COMPUTER AND INFORMATION TECHNOLOGY, vol. 27, no. 6, 31 December 2019 (2019-12-31), pages 1 - 3, XP055842243, ISSN: 1005-1228, DOI: 10.19414/j.cnki.1005-1228.2019.06.001 *
LI ZHISHUAI; XIONG GANG; CHEN YUANYUAN; LV YISHENG; HU BIN; ZHU FENGHUA; WANG FEI-YUE: "A Hybrid Deep Learning Approach with GCN and LSTM for Traffic Flow Prediction*", 2019 IEEE INTELLIGENT TRANSPORTATION SYSTEMS CONFERENCE (ITSC), IEEE, 27 October 2019 (2019-10-27), pages 1929 - 1933, XP033668160, DOI: 10.1109/ITSC.2019.8916778 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113780662A (zh) * 2021-09-14 2021-12-10 北京航空航天大学 一种流量预测方法、装置、设备和介质
CN113780662B (zh) * 2021-09-14 2024-04-23 北京航空航天大学 一种流量预测方法、装置、设备和介质
CN114529043A (zh) * 2022-01-05 2022-05-24 深圳大学 一种城市空间组团划分方法
CN116110588A (zh) * 2022-05-10 2023-05-12 北京理工大学 基于动态邻接矩阵和时空注意力的医学时间序列预测方法
CN116110588B (zh) * 2022-05-10 2024-04-26 北京理工大学 基于动态邻接矩阵和时空注意力的医学时间序列预测方法
CN115221766A (zh) * 2022-06-15 2022-10-21 南京大学 一种改进辐射模型的跨境人口流动模拟方法
CN115221766B (zh) * 2022-06-15 2024-05-07 南京大学 一种改进辐射模型的跨境人口流动模拟方法
CN115759350A (zh) * 2022-10-25 2023-03-07 清华大学 一种数据稀疏区域的人口流动预测方法及装置

Also Published As

Publication number Publication date
CN112116155B (zh) 2023-07-25
CN112116155A (zh) 2020-12-22

Similar Documents

Publication Publication Date Title
WO2021174876A1 (zh) 基于智能决策的人口流动预测方法、装置及计算机设备
US10559097B2 (en) Methods and systems for providing mapping, data management, and analysis
US20190034816A1 (en) Methods and system for associating locations with annotations
Li et al. A regionalization method for clustering and partitioning based on trajectories from NLP perspective
CN113886721B (zh) 个性化兴趣点推荐方法、装置、计算机设备及存储介质
Huang et al. A time-dependent attention convolutional LSTM method for traffic flow prediction
CN110737820B (zh) 用于生成事件信息的方法和装置
Yuan et al. A multi-period emergency medical service location problem based on Wasserstein-metric approach using generalised benders decomposition method
CN112016503B (zh) 人行道检测方法、装置、计算机设备及存储介质
CN116150429A (zh) 异常对象识别方法、装置、计算设备以及存储介质
Zhong et al. ReCovNet: Reinforcement learning with covering information for solving maximal coverage billboards location problem
KR102502515B1 (ko) 유저 단말기의 스캐닝 이미지를 처리하여 생활편의 서비스를 증강현실 기반으로 제공하는 플랫폼의 운영방법
KR102610965B1 (ko) 빅데이터 기반 ３ｄ 인테리어 시뮬레이션 서비스 제공 시스템
EP4318375A1 (en) Graph data processing method and apparatus, computer device, storage medium and computer program product
Church et al. Grand Challenges
Zhao et al. Unveiling the evolution of urban rail transit network: considering ridership attributes
CN116777644A (zh) 基于蚁群算法的车险道路救援规划方法及其相关设备
Cheng et al. Point-of-interest recommendation based on bidirectional self-attention mechanism by fusing spatio-temporal preference
Engelgardt BIG DATA RANKING SYSTEM AS AN EFFECTIVE METHOD OF VISUALIZING THE QUALITY OF URBAN STRUCTURAL UNITS
CN116955850A (zh) 事件信息处理方法、设备、存储介质及程序产品
CN117235257A (zh) 基于人工智能的情感预测方法、装置、设备及存储介质
CN117349479A (zh) 图结构的更新方法、装置、设备、存储介质及程序产品
CN115563317A (zh) 多媒体内容的推送方法、装置、设备及存储介质
CN116910095A (zh) 埋点处理方法、装置、计算机设备及存储介质
CN113469752A (zh) 内容推荐方法、装置、存储介质及电子设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20923589

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205N DATED 12/05/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20923589

Country of ref document: EP

Kind code of ref document: A1