WO2023123625A1

WO2023123625A1 - Urban epidemic space-time prediction method and system, terminal and storage medium

Info

Publication number: WO2023123625A1
Application number: PCT/CN2022/076295
Authority: WO
Inventors: 李子垠; 尹凌
Original assignee: 中国科学院深圳先进技术研究院
Priority date: 2021-12-31
Filing date: 2022-02-15
Publication date: 2023-07-06
Also published as: CN114464329A

Abstract

The present application relates to an urban epidemic space-time prediction method and system, a terminal and a storage medium. The method comprises: collecting individual movement trajectory data and infectious disease case data in a city; processing the individual movement trajectory data, extracting population movement traffic between regions, dividing the population movement traffic according to time attributes, and obtaining a population movement relationship under time attributes on the basis of a proximity relationship of the regions; calculating the similarity of infectious disease case data between the regions by using a histogram-based similarity algorithm, and obtaining a position attention relationship of the regions according to the similarity of the infectious disease case data; and constructing a graph neural network and long short-term memory network-based infectious disease space-time prediction model according to the population movement relationship, the proximity relationship and the position attention relationship. According to the present application, time dependence and spatial dependence in infectious disease case data can be modeled, and the spatial perception capabilities and prediction performance of the space-time prediction model are improved.

Description

A spatio-temporal prediction method, system, terminal and storage medium of urban epidemic situation

technical field

The application belongs to the technical field of infectious disease computing, and in particular relates to a method, system, terminal and storage medium for temporal and spatial prediction of urban epidemics.

Background technique

Infectious disease forecasting has gained prominence in recent years due to the increased availability of public health surveillance data and the development of sophisticated methods. Real-time prediction of disease infections remains hampered by a lack of current estimates of mobility and interaction patterns. Mobility and interaction patterns are key drivers of disease transmission. Prediction of the incidence trend of infectious diseases is an important issue in the field of public health. Timely discovery, tracking and prediction of key information of infectious diseases, such as peak intensity and outbreak time, are crucial for effective formulation of prevention and control strategies and implementation of intervention measures. According to the principles and purposes of modeling, infectious disease models can be divided into two categories: mechanistic models and non-mechanistic models.

Taking seasonal infectious diseases as an example, monitoring and forecasting infectious disease activities, and making corresponding prevention and control preparations in time are crucial to the prevention and control of seasonal infectious diseases and infectious disease pandemics. Existing infectious disease prediction modeling methods include Gaussian process-based models, LSTM-based (Long Short-Term Memory, long-short-term memory network) neural network models, and the like. However, existing infectious disease prediction modeling methods include modeling based on traditional statistical models and modeling based on traditional machine learning. The infectious disease prediction methods based on traditional statistical models require data to meet some strict assumptions, such as The stationarity assumption of time series, etc. However, in the real world, these assumptions are not easy to satisfy. Most of the infectious disease prediction methods based on traditional machine learning still rely on feature engineering, and cannot achieve accurate predictions on difficult prediction problems. In addition, the above methods usually ignore other features from the spatial dimension, and thus cannot model the spatio-temporal dependencies contained in the spatio-temporal data, which limits the predictive performance of the model.

The main core of mechanistic models is the propagation dynamics model, the most commonly used is the compartment model. The compartment model divides the population or other hosts in the same unit into corresponding compartments according to the infection status of the infected, and then simulates the disease development dynamics in different compartments based on the characteristics of disease transmission. Its typical representatives are susceptible-infected-recovered (susceptible-infected-recovered, SIR) and susceptible-latent-infected-recovered (susceptible-exposed-infected-recovered, SEIR) models. Such models are often used to predict the long-term development trend of the disease and deduce the effect of different interventions. However, its modeling is complex, updates are slow, and the uncertainty of model parameters will greatly affect the accuracy of the model, making it difficult to achieve real-time and accurate prediction of short-term development trends of infectious diseases.

Non-mechanism models include statistical models and machine learning models. Traditional statistical models require data to meet some strict assumptions (eg, stationarity assumptions of time series). However, in the real world, these assumptions are not easy to satisfy. Although traditional machine learning-based infectious disease prediction methods are data-driven and do not need to meet strict data assumptions, most traditional machine learning methods still rely on feature engineering and cannot achieve accurate predictions on difficult prediction problems. predict. In addition, in the process of modeling, regardless of the infectious disease prediction method based on the traditional statistical model or the traditional machine learning model based infectious disease prediction method, these methods usually ignore other features from the spatial dimension, and thus cannot analyze the spatial and temporal data. The temporal and spatial dependencies of the implications are modeled. This way of modeling limits the predictive performance of these methods. Infectious disease prediction methods based on deep learning can automatically learn nonlinear features from spatio-temporal data, which has greater performance advantages. Some studies have designed a prediction framework based on deep learning from multiple perspectives to better model the spatiotemporal dependencies contained in spatiotemporal data, so as to improve the predictive performance of the model. However, most of these studies did not take into account the spatial interaction caused by population movement and the spatio-temporal diffusion pattern of infectious diseases within cities when modeling the spatial dependencies between different regions. Accurately predict the epidemic trend of infectious diseases.

Contents of the invention

The present application provides a spatio-temporal prediction method, system, terminal and storage medium of an urban epidemic situation, aiming to solve one of the above-mentioned technical problems in the prior art at least to a certain extent.

In order to solve the above problems, the application provides the following technical solutions:

A spatial-temporal prediction method for urban epidemics, comprising:

Collect individual movement trajectory data and infectious disease case data in the city;

Process the individual movement trajectory data through a data-driven method, extract the population movement flow between various regions, and divide the population movement flow according to the time attribute, and obtain the population under each time attribute based on the proximity of the region mobile relationship;

Use the histogram-based similarity algorithm to calculate the similarity of infectious disease case data between various regions, and obtain the position attention relationship of each region according to the similarity of the infectious disease case data; according to the population movement relationship, proximity relationship and Location-attention relationship constructs a spatial-temporal prediction model of infectious diseases based on graph neural network and long-term short-term memory network;

The intra-city infectious disease case data are input into the infectious disease spatio-temporal prediction model, and the intra-urban infectious disease prediction results are obtained through the infectious disease spatio-temporal prediction model.

The technical solution adopted in the embodiment of the present application also includes: dividing the population movement flow according to the time attribute, and obtaining the population movement relationship under each time attribute based on the proximity relationship of the area, specifically:

The time attribute includes working days, weekends or holidays; the population movement flow of each time attribute is converted into a weighted directed graph G=(V, E), wherein V represents a node set, E represents an edge set, and the vertex Represents each area within the city, and the edge is used to capture the movement pattern; and based on the proximity of the area, it is determined whether there is an edge between the two nodes in the directed graph according to whether the two areas are in contact with each other, and the corresponding The adjacency matrix of the population movement relation of the proximity relation.

The technical solution adopted in the embodiment of the present application also includes: calculating the similarity of infectious disease case data between various regions using a histogram-based similarity algorithm, and obtaining the location attention of each region according to the similarity of infectious disease case data Force relationships include:

Construct corresponding histograms for the infectious disease case data in each region;

Calculate the similarity of infectious disease case data between two regions, if the similarity of infectious disease case data between two regions is higher than the set threshold, the infectious disease outbreak trends in the two regions are considered to be similar and correlated , constructing a connection edge on the graph between the two regions in the directed graph, and generating an adjacency matrix representing positional attention relationships between all regions.

The technical solution adopted in the embodiment of the present application also includes: the calculation formula of the position-attention relationship is:

Among them, θ is the threshold for establishing a connection based on the histogram-based similarity algorithm; when the similarity w _i,j between region i and region j is higher than the threshold θ, a connection is created between region i and region j. Edges, and finally get the adjacency matrix of the corresponding location attention relationship of each area in the city.

The technical solution adopted in the embodiment of the present application also includes: the construction of the spatial-temporal prediction model of infectious diseases based on the graph neural network and long-term short-term memory network according to the population movement relationship, proximity relationship and location attention relationship includes:

The graph structure is input into the graph neural network, and the graph neural network uses the neighborhood aggregation method to normalize the directed graph, so that the incoming edge weight of each node is equal to 1:

Among them, H ⁱ is a matrix, which contains the node representation of the previous layer, the initial H ⁰ is set to represent the historical data of the change of infectious disease cases in each region, W ⁱ represents the trainable parameter matrix of the i-th layer, and f is the nonlinear activation function.

The technical solution adopted in the embodiment of the present application also includes: the construction of the spatial-temporal prediction model of infectious diseases based on the graph neural network and the long-term and short-term memory network according to the population movement relationship, proximity relationship and location attention relationship further includes:

Use a message propagation neural network at each time step to obtain a representation sequence h _i,tn , _hi,t-n+1 ,...,hi _,t-1 , which will represent the sequence h _i,tn , h _{i, t-n+1} ,..., h _{i, t-1} are input into the long-term short-term memory network, and the time series relationship therein is extracted; the calculation formula of the long-term short-term memory network is:

X _i,t ＝LSTM(h _i,tn ,h _i,t-n+1 ,...,h _i,t-1 ) where, X _i,t represents the prediction of the i-th region in the t-th time period Infectious disease case data, h _i,t-1 represents the infectious disease case data of the i-th region in the t-1 time period.

The technical solution adopted in the embodiment of the present application also includes: inputting the infectious disease case data in the city into the infectious disease spatiotemporal prediction model, and obtaining the urban infectious disease prediction result through the infectious disease spatiotemporal prediction model is specifically:

Through the spatial-temporal prediction model of infectious diseases, the neighborhood relationship, population movement relationship and location attention relationship are fused to obtain the prediction results of urban infectious diseases; the fusion method is specifically:

Among them, W _adj , W _od and W _at are the parameter matrices that need to be trained,

and

are the infectious disease prediction results at time t based on the adjacency matrix, population movement flow matrix and location attention matrix, tanh is the activation function,

is the final prediction result of the entire spatio-temporal prediction model at time t.

Another technical solution adopted in the embodiment of the present application is: a spatio-temporal prediction system for urban epidemic situation, including:

Data collection module: used to collect individual movement trajectory data and infectious disease case data in the city;

Flow calculation module: used to process the individual movement trajectory data through a data-driven method, extract the population movement flow between various regions, and divide the population movement flow according to the time attribute, and obtain it based on the proximity relationship of the region The population movement relationship under each time attribute;

Similarity calculation module: used to calculate the similarity of infectious disease case data between various regions using a histogram-based similarity algorithm, and obtain the position attention relationship of each region according to the similarity of the infectious disease case data;

Model building module: used to construct a spatial-temporal prediction model of infectious diseases based on graph neural network and long-term short-term memory network according to the population movement relationship, proximity relationship and location attention relationship;

Infectious disease prediction module: used to input the urban infectious disease case data into the infectious disease spatiotemporal prediction model, and obtain the urban infectious disease prediction results through the infectious disease spatiotemporal prediction model.

Another technical solution adopted by the embodiment of the present application is: a terminal, the terminal includes a processor and a memory coupled to the processor, wherein,

The memory stores program instructions for realizing the spatio-temporal prediction method of the urban epidemic situation;

The processor is used to execute the program instructions stored in the memory to control the spatio-temporal prediction of urban epidemics.

Another technical solution adopted in the embodiment of the present application is: a storage medium storing program instructions executable by a processor, and the program instructions are used to execute the method for spatio-temporal prediction of urban epidemic situation.

Compared with the prior art, the beneficial effects produced by the embodiment of the present application lie in that the urban epidemic spatiotemporal prediction method, system, terminal and storage medium of the embodiment of the present application extract the urban population movement flow through the individual movement trajectory data in the city, and analyze the urban population flow according to the time attribute. The population movement flow within a certain period of time is divided, and the proximity relationship, population movement relationship, and location attention relationship between regions are obtained according to the population movement flow of different time attributes, and the neighborhood relationship, population movement relationship, and location attention relationship are constructed. The spatio-temporal prediction model of infectious diseases based on graph neural network and long-term short-term memory network can make refined predictions on the trend of infectious diseases through the spatio-temporal prediction model of infectious diseases. In the embodiment of the present application, by dividing the time attribute of the population movement flow, it is possible to accurately capture the impact of population interaction and movement under different time attributes on the spread and spread of infectious diseases within the city, and make full use of the local spatial context information of each region, and Multiple spatial relationships between regions are considered to better model the temporal and spatial dependencies in infectious disease case data, which greatly improves the spatial awareness and model prediction performance of the infectious disease spatiotemporal prediction model. Realized the prediction of the development trend of infectious diseases in the city with a higher spatial resolution, completed the refined analysis of the epidemic situation of infectious diseases, and helped the government and public health departments to gain timely and accurate insight into the development trend of infectious diseases in the city, and carry out targeted Epidemic prevention and control intervention can maximize the protection of people's lives, health and safety.

Description of drawings

Fig. 1 is the flow chart of the urban epidemic situation spatio-temporal prediction method of the embodiment of the present application;

Fig. 2 is the schematic structural diagram of the urban epidemic situation spatio-temporal prediction system of the embodiment of the present application;

FIG. 3 is a schematic diagram of a terminal structure in an embodiment of the present application;

FIG. 4 is a schematic structural diagram of a storage medium according to an embodiment of the present application.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, not to limit the present application.

Aiming at the deficiencies of existing technologies, the urban epidemic spatio-temporal prediction method of the embodiment of this application is based on the proximity relationship, population movement relationship and location attention relationship between various regions in the city to construct a spatio-temporal prediction model based on graph neural network and long-term short-term memory network , to better model the temporal and spatial dependencies in spatio-temporal data, when modeling spatial dependencies, consider multiple spatial relationships between regions and construct a corresponding relationship graph structure; When modeling time dependence, make full use of the local spatial context information of each region to improve the prediction performance of the model; and divide the population movement flow into different time attributes according to the population movement pattern of infectious disease-susceptible groups, and construct A directed graph structure with different temporal attributes to improve the spatio-temporal awareness of the spatio-temporal prediction model and further improve the prediction performance of the model.

Specifically, please refer to FIG. 1 , which is a flow chart of the spatio-temporal prediction method for urban epidemic situation in the embodiment of the present application. The urban epidemic spatio-temporal prediction method of the embodiment of the present application includes the following steps:

S1: Obtain the data of infectious disease cases in the city, and use mobile devices to collect individual movement trajectory data in the city;

In this step, the infectious disease case data include, but are not limited to, the number of infected cases, the number of deaths, clinical symptoms, and hospitalized personnel information. Individual movement track data includes, but is not limited to, each individual's mobile phone number, signaling timestamp, and mobile phone location data. In order to better predict the incidence trend of infectious diseases within the city for susceptible groups such as minors, large-scale individual movement trajectory data of infectious disease cases is obtained in this embodiment of the application. Mobile devices include, but are not limited to, devices such as cell phones or smart watches.

S2: Process the individual movement trajectory data through data-driven methods, extract the population movement flow between different regions, and obtain the urban population movement relationship;

In this step, the data-driven method is used to process the individual movement trajectory data as follows: extract the population movement flow between different regions through large-scale mobile phone location data, and capture the urban population movement relationship.

S3: Divide the population movement flow according to the time attribute, construct a directed graph of the population movement flow with different time attributes, and determine whether there is an edge between two nodes in the directed graph based on the proximity of the region, and generate different time attributes The adjacency matrix of the population movement relationship;

In this step, the time attribute includes but not limited to weekdays, weekends or holidays. Suppose a city is given, create a graph structure with day, week or month as the time interval unit, and divide the population movement flow into two time attribute population movement flows according to weekdays and weekends, and divide the population movement flow of each time attribute The flows are respectively transformed into a weighted directed graph G = (V, E), where V represents a set of nodes, E represents a set of edges, vertices represent various regions within the city, and edges are used to capture mobility patterns. For example, the weight from vertex v to vertex u

Represents the total number of people who moved from area v to area u on weekdays of week t, so as to obtain the adjacency matrix of the population movement relationship under the two time attributes of weekdays and weekends in week t

and

Further, according to the first law of geography, all things are related to other things, but things with close distances are more correlated than things with far distances, therefore, the embodiment of the present application considers the space between things with close distances relationship, that is, proximity relationship. Based on the adjacency relationship of the regions, it is determined whether there is an edge between the two regions (nodes) according to whether the two regions are in contact with each other, and the adjacency matrix A ^adj of the population movement relationship corresponding to the adjacency relationship is obtained. Specifically, assuming that regions v and u are adjacent regions, the weight from vertex v to vertex u

Indicates the total number of individuals who move from area v to area u on weekdays, and the directed graph G can contain the population flow behavior of each adjacent area in the city. The mobility of areas v and u on weekdays forms an edge, which is multiplied by the number of cases in area v during that time period to obtain a relative score, which indicates how many infected people are likely to flow from area v to area u. It can be understood that the case change pattern between two adjacent areas may not be fluid, for example, areas i and j are adjacent areas, the flow rate of area i is fixed and always zero, and the flow rate of area j is dynamic and non-zero, there is no correlation between the change patterns of cases in these two adjacent regions.

S4: Construct the histogram of the infectious disease case data, use the histogram-based similarity algorithm to calculate the similarity of the infectious disease case data between regions, and obtain the location attention relationship representing all regions according to the similarity of the infectious disease case data adjacency matrix;

In this step, the correlation between two regions may be affected by geographical distance, that is, adjacent regions may have similar terrain or climate characteristics, making them have similar infectious disease outbreak trends. However, non-contiguous areas may also have potential dependencies due to population movement and similar geographic features, but it is difficult to model all relevant factors for infectious disease outbreaks. Therefore, the present invention utilizes a histogram-based similarity algorithm to calculate the correlation of infectious disease case data between regions. If the similarity of infectious disease case data in two regions is relatively high, correspondingly, their case change patterns will be relatively similar, and the embodiment of the present application calculates the infectious disease cases in these non-adjacent but highly correlated regions The similarity of the data, so as to capture the trend of case changes from a more global spatial perspective.

The similarity algorithm based on the histogram is as follows: firstly, construct corresponding histograms for the time series of infectious disease case data samples in different regions; secondly, calculate the similarity of the infectious disease case data in two regions, if the infection If the similarity of disease case data is higher than the set threshold, it is considered that the infectious disease outbreak trends in these two areas are relatively similar and there is a strong correlation between them, and a graph is constructed between these two areas in the directed graph , so as to generate an adjacency matrix representing the location-attention relationship between all regions. Calculated as follows:

Among them, θ is the threshold for establishing a connection edge based on the histogram-based similarity algorithm. When the similarity w _i,j between area i and area j is higher than the threshold θ, a connection edge is created between area i and area j, and finally the adjacency matrix of the attention relationship of each area in the city is obtained

Specifically, the histogram construction algorithm is as follows:

Histogram construction algorithm

The similarity algorithm based on histogram is as follows:

similarity algorithm

S5: Use the proximity relationship, population movement relationship and location attention relationship to train the spatiotemporal prediction model of infectious diseases based on Graph Neural Networks (GNN) and Long short-term memory (LSTM);

In this step, the graph neural network is a neighbor aggregation strategy, and the representation vector of a node is calculated by its neighbor nodes through cyclic aggregation and transfer representation vectors. The framework of graph neural network is Message Passing Neural Network (MPNN). Message Propagation Neural Network is a formal framework for spatial graph convolution. In the embodiment of the present application, the graph neural network uses the following neighborhood aggregation method to normalize the input directed graph structure matrix A:

Among them, H ⁱ is a matrix, which contains the node representation of the previous layer, the initial H ⁰ is set to represent the historical data of the change of infectious disease cases in each region, W ⁱ represents the trainable parameter matrix of the i-th layer, and f is the nonlinear activation function , such as the ReLU function. Normalize the input directed graph structure matrix A so that the incoming edge weight of each node is equal to 1, and the normalized directed graph structure matrix is obtained

Long short-term memory network is a special kind of recurrent neural network (Recurrent Neural Network, RNN), which can be used to process sequence data. The long short-term memory network is mainly to solve the problem of gradient disappearance and gradient explosion in the long sequence training process. Use one MPNN at each time step to obtain a representation sequence h _i,tn , _hi,t-n+1 ,...,hi _,t-1 . Input these representation sequences into the long short-term memory network to extract the temporal features. The LSTM calculation formula is expressed as:

X _i,t ＝LSTM(h _i,tn ,h _i,t-n+1 ,...,h _i,t-1 ) (3)

Among them, X _i,t represents the infectious disease case data predicted in the i-th region in the t-th time period, h _i,t-1 represents the infectious disease case data in the i-th region in the t-1 time period, etc. input representation of features.

S6: Through the spatial-temporal prediction model of infectious diseases, the neighborhood relationship, population movement relationship and location attention relationship are fused to obtain the prediction results of urban infectious diseases;

Among them, in order to be able to simultaneously consider the influence of the proximity relationship, population movement relationship and location attention relationship on the prediction of infectious diseases within the city, the present invention adopts a fusion method based on a parameter matrix, specifically:

and

is the final prediction result of the entire infectious disease spatiotemporal prediction model at time t.

Since the embodiment of the present application uses artificial intelligence deep learning to predict the trend of infectious diseases, the model parameters can be updated by learning new data after multiple predictions, making the model more intelligent and efficient.

Based on the above, the urban epidemic spatio-temporal prediction method of the embodiment of the present application extracts the urban population movement flow through the individual movement trajectory data in the city, divides the population movement flow within a certain period of time according to the time attribute, and obtains the population movement flow according to different time attributes The proximity relationship, population movement relationship and location attention relationship between various regions, and according to the proximity relationship, population movement relationship and location attention relationship, a spatiotemporal prediction model of infectious diseases based on graph neural network and long short-term memory network is constructed. Predictive models make refined forecasts on infectious disease trends. In the embodiment of the present application, by dividing the time attribute of the population movement flow, it is possible to accurately capture the impact of population interaction and movement under different time attributes on the spread and spread of infectious diseases within the city, and make full use of the local spatial context information of each region, and Multiple spatial relationships between regions are considered to better model the temporal and spatial dependencies in infectious disease case data, which greatly improves the spatial awareness and model prediction performance of the infectious disease spatiotemporal prediction model. Realized the prediction of the development trend of infectious diseases in the city with a higher spatial resolution, completed the refined analysis of the epidemic situation of infectious diseases, and helped the government and public health departments to gain timely and accurate insight into the development trend of infectious diseases in the city, and carry out targeted Epidemic prevention and control intervention can maximize the protection of people's lives, health and safety.

Please refer to FIG. 2 , which is a schematic structural diagram of the spatio-temporal prediction system of the urban epidemic situation in the embodiment of the present application. The spatial-temporal prediction system 40 of the urban epidemic situation in the embodiment of the present application includes:

Data collection module 41: used to collect individual movement trajectory data and infectious disease case data in the city;

Flow calculation module 42: used to process individual movement trajectory data through a data-driven method, extract population movement flow between various regions, divide population movement flow according to time attributes, and obtain various time attributes based on the proximity of regions The population movement relationship under ;

Similarity calculation module 43: used to calculate the similarity of infectious disease case data between various regions using a histogram-based similarity algorithm, and obtain the position attention relationship of each region according to the similarity of infectious disease case data;

Model building module 44: used to construct a spatiotemporal prediction model of infectious diseases based on graph neural network and long short-term memory network according to population movement relationship, proximity relationship and location attention relationship;

Infectious disease prediction module 45: used to input the infectious disease case data in the city into the infectious disease spatiotemporal prediction model, and obtain the urban infectious disease prediction result through the infectious disease spatiotemporal prediction model.

Please refer to FIG. 3 , which is a schematic diagram of a terminal structure in an embodiment of the present application. The terminal 50 includes a processor 51 and a memory 52 coupled to the processor 51 .

The memory 52 stores program instructions for realizing the above-mentioned spatio-temporal prediction method of urban epidemic situation.

The processor 51 is used to execute the program instructions stored in the memory 52 to control the spatio-temporal prediction of the urban epidemic situation.

Wherein, the processor 51 may also be referred to as a CPU (Central Processing Unit, central processing unit). The processor 51 may be an integrated circuit chip with signal processing capabilities. The processor 51 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components . A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.

Please refer to FIG. 4 , which is a schematic structural diagram of a storage medium according to an embodiment of the present application. The storage medium of the embodiment of the present application stores a program file 61 capable of realizing all the above-mentioned methods, wherein the program file 61 can be stored in the above-mentioned storage medium in the form of a software product, and includes several instructions to make a computer device (which can It is a personal computer, a server, or a network device, etc.) or a processor (processor) that executes all or part of the steps of the methods in various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. , or terminal devices such as computers, servers, mobile phones, and tablets.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in this invention may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention will not be limited to these embodiments shown in the present invention, but will conform to the widest scope consistent with the principles and novel features disclosed in the present invention.

Claims

A spatio-temporal forecasting method for an urban epidemic, characterized by comprising:

Collect individual movement trajectory data and infectious disease case data in the city;

Process the individual movement trajectory data through a data-driven method, extract the population movement flow between various regions, and divide the population movement flow according to the time attribute, and obtain the population under each time attribute based on the proximity of the region mobile relationship;

Using a histogram-based similarity algorithm to calculate the similarity of the infectious disease case data between the regions, and obtain the position attention relationship of each region according to the similarity of the infectious disease case data;

Constructing a spatiotemporal prediction model of infectious diseases based on graph neural network and long short-term memory network according to the population movement relationship, proximity relationship and location attention relationship;

The intra-city infectious disease case data are input into the infectious disease spatio-temporal prediction model, and the intra-urban infectious disease prediction results are obtained through the infectious disease spatio-temporal prediction model.
The spatio-temporal prediction method of urban epidemic situation according to claim 1, characterized in that, the population movement flow is divided according to the time attribute, and the population movement relationship under each time attribute is obtained based on the proximity relationship of the region as follows:

The time attribute includes working days, weekends or holidays; the population movement flow of each time attribute is converted into a weighted directed graph G=(V, E), wherein V represents a node set, E represents an edge set, and the vertex Represents each area within the city, and the edge is used to capture the movement pattern; and based on the proximity of the area, it is determined whether there is an edge between the two nodes in the directed graph according to whether the two areas are in contact with each other, and the corresponding The adjacency matrix of the population movement relation of the proximity relation.
The spatio-temporal prediction method of urban epidemic situation according to claim 2, characterized in that, the similarity of the infectious disease case data between the regions is calculated using a histogram-based similarity algorithm, and according to the similarity of the infectious disease case data Obtaining the location-attention relationship of each region includes:

Construct corresponding histograms for the infectious disease case data in each region;

Calculate the similarity of infectious disease case data between two regions, if the similarity of infectious disease case data between two regions is higher than the set threshold, the infectious disease outbreak trends in the two regions are considered to be similar and correlated , constructing a connection edge on the graph between the two regions in the directed graph, and generating an adjacency matrix representing positional attention relationships between all regions.
The spatio-temporal prediction method of urban epidemic situation according to claim 3, is characterized in that, the calculation formula of described location attention relation is:

Among them, θ is the threshold for establishing a connection based on the histogram-based similarity algorithm; when the similarity w i,j between region i and region j is higher than the threshold θ, a connection is created between region i and region j. Edges, and finally get the adjacency matrix of the corresponding location attention relationship of each area in the city.
The spatio-temporal prediction method of urban epidemic according to any one of claims 1 to 4, characterized in that, according to the population movement relationship, proximity relationship and location attention relationship, constructing a contagion algorithm based on a graph neural network and a long-term short-term memory network Disease spatiotemporal prediction models include:

The graph structure is input into the graph neural network, and the graph neural network uses the neighborhood aggregation method to normalize the directed graph, so that the incoming edge weight of each node is equal to 1:

Among them, H i is a matrix, which contains the node representation of the previous layer, the initial H 0 is set to represent the historical data of the change of infectious disease cases in each region, W i represents the trainable parameter matrix of the i-th layer, and f is the nonlinear activation function.
The spatiotemporal prediction method of urban epidemic situation according to claim 5, characterized in that, constructing the spatiotemporal prediction model of infectious diseases based on graph neural network and long short-term memory network according to the population movement relationship, proximity relationship and position attention relationship also include:

Use a message propagation neural network at each time step to obtain a representation sequence h i,tn , hi,t-n+1 ,...,hi ,t-1 , which will represent the sequence h i,tn , h i, t-n+1 ,..., h i, t-1 are input into the long-term short-term memory network, and the time series relationship therein is extracted; the calculation formula of the long-term short-term memory network is:

X i,t ＝LSTM(h i,tn ,h i,t-n+1 ,...,h i,t-1 )

Among them, X i,t represents the predicted infectious disease case data of the i-th region in the t-th time period, and h i,t-1 represents the infectious disease case data of the i-th region in the t-1 time period.
The spatio-temporal prediction method of urban epidemic situation according to claim 6, characterized in that the data of infectious disease cases in the city are input into the spatio-temporal prediction model of infectious diseases, and the prediction results of infectious diseases in the city are obtained through the spatio-temporal prediction model of infectious diseases Specifically:

Through the spatial-temporal prediction model of infectious diseases, the neighborhood relationship, population movement relationship and location attention relationship are fused to obtain the prediction results of urban infectious diseases; the fusion method is specifically:

Among them, W adj , W od and W at are the parameter matrices that need to be trained,
and
are the infectious disease prediction results at time t based on the adjacency matrix, population movement flow matrix and location attention matrix, tanh is the activation function,
is the final prediction result of the entire infectious disease spatiotemporal prediction model at time t.
A spatial-temporal prediction system for urban epidemics, characterized in that it includes:

Data collection module: used to collect individual movement trajectory data and infectious disease case data in the city;

Flow calculation module: used to process the individual movement trajectory data through a data-driven method, extract the population movement flow between various regions, and divide the population movement flow according to the time attribute, and obtain it based on the proximity relationship of the region The population movement relationship under each time attribute;

Similarity calculation module: used to calculate the similarity of infectious disease case data between various regions using a histogram-based similarity algorithm, and obtain the position attention relationship of each region according to the similarity of the infectious disease case data;

Model building module: used to construct a spatial-temporal prediction model of infectious diseases based on graph neural network and long-term short-term memory network according to the population movement relationship, proximity relationship and location attention relationship;

Infectious disease prediction module: used to input the urban infectious disease case data into the infectious disease spatiotemporal prediction model, and obtain the urban infectious disease prediction results through the infectious disease spatiotemporal prediction model.
A terminal, characterized in that the terminal includes a processor and a memory coupled to the processor, wherein,

The memory is stored with program instructions for realizing the urban epidemic spatiotemporal prediction method described in any one of claims 1-7;

The processor is used to execute the program instructions stored in the memory to control the spatio-temporal prediction of urban epidemics.
A storage medium, characterized in that it stores program instructions executable by a processor, and the program instructions are used to execute the urban epidemic spatio-temporal prediction method according to any one of claims 1 to 7.