WO2023123624A1 - Method and system for predicting influenza outbreak trend in city, and terminal and storage medium - Google Patents

Abstract

The present application relates to a method and system for predicting an influenza outbreak trend in a city, and a terminal and a storage medium. The method comprises: acquiring influenza case data within a city, and collecting individual movement trajectory data within the city; processing the individual movement trajectory data by means of a data driving method, so as to acquire a population movement relationship; on the basis of the population movement relationship, extracting spatial scale information of the influenza case data by using a graph neural network, and extracting a time sequence relationship of the influenza case data by using a long short-term memory network; and according to the spatial scale information and the time sequence relationship of the influenza case data, obtaining an influenza outbreak trend prediction result for the city, which has an outbreak of influenza. By means of the embodiments of the present application, a higher spatial resolution prediction for an influenza development situation within a city is realized, and a refined analysis on influenza is completed, such that a government and a public health department are assisted in gaining an insight into the influenza development situation within the city in a timely and accurate manner, and targeted intervention for epidemic prevention and control is performed, thereby guaranteeing the lives, health and safety of people to the greatest extent.

Description

Urban influenza incidence trend prediction method, system, terminal and storage medium technical field

The application belongs to the technical field of influenza computing, and in particular relates to a method, system, terminal and storage medium for predicting the incidence trend of influenza in an influenza city.

Background technique

From the end of 2019 to the beginning of 2021, COVID-19 has developed from a local outbreak to a global pandemic. More than 100 million people have been infected and more than 2.52 million people have died worldwide. The outbreak of this epidemic has reminded us that public health risks are still one of the major social risks facing mankind. Taking influenza (abbreviated as influenza) as an example, it is an acute respiratory flu caused by influenza virus. Influenza virus can break out in a short period of time, seriously threatening the health of the masses.

At present, influenza prevention and early warning methods in countries around the world mainly rely on traditional influenza surveillance. Influenza surveillance systems can provide a large amount of information and data useful for influenza prevention and control, including the number of infected cases, the number of deaths, clinical symptoms, and hospitalization information. . The government and public health departments assess recent influenza activity trends based on the information reported by the influenza surveillance system, and make corresponding public health decisions and emergency response plans. However, due to the fact that the verification and reporting of grassroots monitoring data often takes a lot of time (weeks or even months), the monitoring data will lag behind the actual situation, and most countries can only achieve provincial/state or Early warning at the city level. Therefore, it is urgent to make real-time, accurate and rapid predictions on the trend of influenza incidence in cities, so as to help the government and public health departments to timely and accurately understand the development trend of influenza in cities, and to maximize the protection of people's lives, health and safety.

The interaction and movement of people will lead to the spread and spread of influenza. In order to better realize the real-time, accurate and rapid prediction of the incidence trend of influenza in the city, it is necessary to study the spatial interaction in the process of population movement in the city, and construct a coupling spatial interaction model. Spatial-temporal forecasting model for urban influenza incidence trends. Existing influenza prediction modeling methods include Gaussian process-based models, LSTM-based neural network models, etc. However, most of the existing influenza incidence trend prediction methods often only rely on the time series characteristics of influenza-like illness statistics for prediction, without taking into account the spatial dependence between different regions and the spatial interaction between population movement and influenza transmission. Effect, such methods are difficult to accurately predict the incidence trend of influenza within the city due to the lack of information. Although other influenza prediction methods comprehensively utilize the spatiotemporal feature information of a specific location, the spatial feature extraction method is not perfect, and the accuracy of the extracted spatial feature information needs to be improved.

As the complexity of the human movement network within the city continues to deepen, traditional theoretical modeling methods cannot finely simulate the population movement and interaction within the city, and it has been difficult to meet the needs of the current spatial interaction modeling within the city. In addition, spatial interaction modeling methods based on OD (traffic trip volume) flows are often studied at lower spatial resolutions, such as national, provincial/state or city levels, and the graph structure constructed by this method at the urban scale The flow is too dispersed, leading to biased predictions of spatiotemporal patterns of influenza outbreaks.

Contents of the invention

The present application provides an influenza prediction method, system, terminal and storage medium for urban influenza incidence trends, 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 method for predicting the incidence trend of urban influenza, including:

Obtain the data of influenza cases in the city, and collect the data of individual movement trajectories in the city;

Process the individual movement trajectory data through a data-driven method, obtain the home address of each individual, set the home address as the starting point of each individual movement trajectory, and extract each individual from the starting point to Flow data to other regions to obtain the population movement relationship between regions;

Based on the population movement relationship, using a graph neural network to extract the spatial scale information of the influenza case data, and using a long short-term memory network to extract the time series relationship of the influenza case data;

According to the spatial scale information and time series relationship of the influenza case data, the urban influenza incidence trend prediction result is obtained.

The technical solution adopted in the embodiment of the present application also includes: the collection of individual movement trajectory data in the city includes:

Use mobile devices to collect individual movement trajectory data in the city; the mobile devices include mobile phones or smart watches; the individual movement trajectory data include each individual's mobile phone number, signaling time stamp, and base station latitude and longitude.

The technical solution adopted in the embodiment of the present application also includes: the processing of the individual movement trajectory data through the data-driven method is specifically:

Merge the original Thiessen polygons established by the city base stations in the grid area of the set time period and the set distance; divide the individual movement track data according to the stay time, and set the position with the longest night stay time of each individual Set as the individual's home address, and extract the OD flow and home center flow at a set interval in a grid area with a set distance.

The technical solution adopted in the embodiment of the present application also includes: the processing of the individual movement track data through the data-driven method further includes:

According to the population distribution information and geographical location information of each street in the city in the census data, the network flow extraction is performed on the individual movement track data through a theoretical model method; the theoretical model method includes a gravity model, a radiation model or a spatial proximity relationship model.

The technical solution adopted in the embodiment of the present application also includes: the extraction of the spatial scale information of the influenza case data by using the graph neural network is specifically:

The framework of the graph neural network is a message propagation neural network, and the message propagation neural network extracts spatial scale information based on spatial graph convolution; an undirected graph G is defined, the feature vector of a node v is x _v , and the feature of an edge is e _vw , connecting nodes v and w, N(v) represents the neighbor nodes of node v in graph G, t is the running time step, and the feature x _v of node v is used as the initial state of its hidden state

After that, the update of the hidden state by the spatial graph convolution is expressed as:

The message propagation neural network decomposes the spatial graph convolution into two parts of message delivery and state update operation, which are respectively completed by the message function _M1 and the node update function _U1 ; the message function _M1 is used to aggregate the characteristics of neighboring nodes , forming a message vector, which is ready to be delivered to the central node; the node update function U _l is used to update the node representation at the current moment, and combine the node representation at the current moment with the message obtained from the message function to obtain spatial scale information.

The technical solution adopted in the embodiment of the present application also includes: the extraction of the spatial scale information of the influenza case data by using the graph neural network is specifically:

Transform the individual movement trajectory data of a set time range into a weighted directed graph, the vertices represent street-level areas, and the edges are used to capture movement patterns; at time t, the flow between areas u and v forms an edge, which is multiplied by The number of cases in region u at time t

Indicates how many infected people may move from area u to area v; set

is a vector of node attributes containing the number of cases in each of the past w weeks in region u; the message passed through the message propagation neural network computes a feature vector for each region using the composite score from all regions:

where A represents the adjacency matrix of regional population movement flows, C _t is a matrix whose rows contain attributes of different regions; x _u ∈ R ^w is a vector combining the number of influenza cases moving within and towards region u, x The expression formula of _u is:

x _u ＝(x _u w _j,u +x _u w _i,u +….x _u w _v,u )+x _u w _u,u

where x _u ∈ R ^w denotes the estimate of the number of new potential cases in area u.

The technical solution adopted in the embodiment of the present application also includes: the time series relationship of the influenza case data extracted using the long-short-term memory network is specifically:

The long-short-term memory network calculates the output ht of the hidden layer at the current moment based on the input _xt at the current moment and the output ht _-1 of the hidden layer at the previous time period. The calculation formula of the long-short-term memory network is _:

_yi,t = LSTM(h _i,tn ,h _i,tn ,...,h _i,t-1 )

Among them, h _i,t-1 represents the influenza case data of the i-th region in the t-1 time period, and y _i,t represents the predicted influenza case data of the i-th region in the t-th time period.

Another technical solution adopted in the embodiment of the present application is: a system for predicting the incidence trend of urban influenza, including:

Data collection module: used to obtain the data of influenza cases in the city and collect the data of individual movement trajectories in the city;

Data processing module: used to process the individual movement trajectory data through a data-driven method, obtain the home address of each individual, set the home address as the starting point of each individual movement trajectory, and extract each individual From the flow data of the starting point to other regions, the population movement relationship among the regions is obtained;

Spatio-temporal feature extraction module: used to extract the spatial scale information of the influenza case data by using the graph neural network based on the population movement relationship, and extract the time series relationship of the influenza case data by using the long short-term memory network;

Influenza prediction module: used to obtain urban influenza incidence trend prediction results based on the spatial scale information and time series relationship of the influenza case data.

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 is stored with program instructions for realizing the urban influenza incidence trend prediction method;

The processor is configured to execute the program instructions stored in the memory to control urban influenza incidence trend prediction.

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 predicting the incidence trend of urban influenza.

Compared with the prior art, the beneficial effects produced by the embodiments of the present application are: the method, system, terminal and storage medium for predicting the incidence trend of urban influenza in the embodiments of the present application extract the population movement relationship through the individual movement trajectory data in the city, and based on the population movement relationship , using the graph convolutional neural network to obtain the spatial scale information of the influenza case data, and using the long short-term memory network to extract the time series relationship of the influenza case data, and making a refined prediction of the influenza trend according to the spatial scale information and the time series relationship. The embodiment of this application realizes the prediction of the development trend of influenza in the city with higher spatial resolution, completes the refined analysis of influenza, helps the government and public health departments to timely and accurately understand the development trend of influenza in the city, and conduct targeted Epidemic prevention and control intervention can maximize the protection of people's lives, health and safety. Compared with the prior art, the present application has at least the following beneficial effects:

(1) After integrating large-scale mobile phone location data, the defect that the traditional theoretical model method cannot be refined to simulate the population movement and interaction within the city has been improved, and the mobility of individuals has been more realistically restored.

(2) Compared with other data-driven modeling methods such as the spatial interaction modeling method based on OD flow, the spatial interaction modeling method of the family center flow can be constructed at a higher spatial resolution and at the urban scale. The graph structure traffic is more concentrated, which can obviously capture multiple strong commuting traffic, and then effectively improve the prediction accuracy of the deep learning model of influenza within the city.

(3) The present invention only needs to obtain the location information of the mobile phone, and does not need to obtain multi-dimensional information such as weekly average temperature, air pressure, rainfall, relative humidity, maximum temperature difference, and sunshine time, which reduces the difficulty of data acquisition and processing.

Description of drawings

Fig. 1 is the flowchart of the method for predicting the incidence of urban influenza in the embodiment of the present application;

Fig. 2 is the schematic structural diagram of the urban influenza incidence trend 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.

Please refer to FIG. 1 , which is a flow chart of the method for predicting the incidence trend of urban influenza according to the embodiment of the present application. The urban influenza incidence trend prediction method of the embodiment of the present application comprises the following steps:

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

In this step, mobile devices include but are not limited to devices such as mobile phones or smart watches. In the embodiment of this application, the individual movement trajectory data includes but not limited to each individual’s mobile phone number, signaling timestamp, identified base station latitude and longitude and other information. The specific way to obtain the individual movement trajectory data set is: based on China Unicom HDFS+Hive +Spark big data platform obtains the location information carried by the mobile phone signaling of 5.6 million users, deduplicates the location information, and removes the records missing key information in the location information to obtain individual movement trajectory data.

S2: Process individual movement trajectory data through data-driven methods and theoretical model methods, obtain the home address of each individual, set the home address as the starting point of all movement trajectories of each individual, and extract each individual from the starting point Flow data to other regions to obtain the population movement relationship between regions;

In this step, the individual’s movement trajectory data is divided according to the stay time by the data-driven method, the location with the longest night stay time is set as the individual’s home address, and the home address is set as the starting point of all the individual’s movement trajectories. The starting point; and then extract the flow data of each individual from the starting point to other areas (such as work and entertainment places) to obtain the population movement relationship between different areas.

Further, the processing of individual movement trajectory data through the data-driven method is as follows: merging the original Thiessen polygons established by the city base stations in the 500-meter grid area within a certain period of time, and the location where each user spends the longest time at night Set as the user's home address, extract the OD traffic and home center traffic every half hour in the 500-meter grid area of the city; then map to each street unit to get every half-hour in the 500-meter grid area within a certain period of time OD traffic and home center traffic.

Secondly, the processing of individual movement trajectory data through theoretical model methods is as follows: according to the population distribution information and geographical location information of each street in the city in the census data, through theoretical model methods such as gravity model, radiation model and spatial proximity model, the individual The mobile trajectory data set is used for network flow extraction, where:

The gravity model assumes that the intensity of population movement between two regions is directly proportional to their respective population distributions and inversely proportional to the distance between them. The gravity model formula between two urban areas can usually be expressed as:

In the above formula, T _ij is the population flow intensity between area i and area j in the city, m _i and n _j are the population distribution of area i and area j respectively, r _ij is the distance between area i and area j, a is a constant, which can take a value of 1.

Radiation models view population movement as a stochastic process governed by joint probabilities, depending on the population distribution of origin, destination, and sphere of influence. The radiation model formula is:

T _i =m _i (N _c /N) (3)

In the above formula, s _ij represents the total population within the area with area i as the center and the distance between area i and area j as the radius. N is the total population of area i, and N _c is the commuter population of area i.

The spatial proximity relationship model is based on the first law of geography, all things are related to other things, but things that are close in distance are more related than things that are far away. Therefore, the present invention considers the spatial relationship between things with close distances, that is, the proximity relationship. Based on the adjacent relationship of the regions, the present invention judges whether there is an edge between the two regions (nodes) according to whether the two regions are in contact with each other, and obtains an adjacency matrix corresponding to the adjacent relationship.

S3: Input the influenza case data and the population movement relationship into the Graph Neural Networks (GNN), and the graph neural network extracts the spatial features of the influenza case data according to the population movement relationship, and obtains the spatial scale information of the influenza case data;

In this step, the graph neural network is used for spatial dependence modeling, which has a wider application range and better generalization ability. The core idea of graph neural network is to learn a function that enables each node to aggregate its own features and its neighbors' features to generate a new feature representation of the node. Construct a neural network model containing multiple graph convolutional layers. Through continuous iteration and learning, the topological relationship on the graph can finally be used to learn the input features of each node and make predictions. 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. First define the undirected graph G, the feature vector of the node v is x _v , the feature of the edge is e _vw , connecting the nodes v and w, N(v) represents the neighbor nodes of the node v in the graph G, t is the running time step, Use the feature x _v of node v as the initial state of its hidden state

After that, the update of the hidden state by spatial graph convolution is expressed by the following formula:

The message propagation neural network decomposes the spatial domain graph convolution into two parts: message passing and state update operation, which are completed by message function M _l and node update function U _l respectively. The function of the message function M _l is to aggregate the characteristics of the neighbor nodes to form a message vector, which is ready to be transmitted to the central node. The role of the node update function U _l is to update the node representation at the current moment, and combine the node representation at the current moment with the message obtained from the message function to obtain the spatial scale information of the influenza case data.

Specifically, the graph neural network extracts the spatial features of the influenza case data according to the population movement relationship, and obtains the spatial scale information of the influenza case data as follows: convert the individual movement trajectory data within a set time range into a weighted directed graph, and its vertices represent Street-level regions, edges are used to capture movement patterns. For example, the weight w _{v,w of an edge (v,u) from vertex v to vertex u} represents the total number of individuals whose home address is in area v who move to area u at time t. At time t, the flow between regions u and v forms an edge that is multiplied by the number of cases in region u at time t

A relative score is provided indicating how many infected people are likely to move from area u to area v. set up

is a vector of node attributes, which contains the number of cases in each week in the past w weeks in region u. Messages passing through this network use the combined scores from all regions to compute a feature vector for each region as follows:

where A represents the adjacency matrix of regional population movement flows, and _Ct is a matrix whose rows contain attributes of different regions. x _u ∈ R ^w is a vector combining the number of influenza cases within and moving towards area u. The expression formula of x _u is:

x _u ＝(x _u w _j,u +x _u w _i,u +….x _u w _v,u )+x _u w _u,u (7)

where region u receives people from different regions, and _Xt contains vectors of past cases in that region. x _u ∈ R ^w represents an estimate of the number of new potential cases in area u, broken down into cases received from other areas and new cases due to mobility within area u. In order to update the vertex representation of each input graph, the embodiment of this application uses the following neighborhood aggregation scheme:

In the above formula, H ⁱ is a matrix containing the node representation of the previous layer, H ⁰ =X, W ⁰ is the matrix of trainable parameters of the first layer, and f is a nonlinear activation function, such as ReLU. By normalizing the adjacency matrix A, the sum of the weights of the incoming edges of each node is equal to 1. Based on a model with a K-order neighborhood aggregation layer, as the order of the neighborhood aggregation layer increases, the final nodes will capture more and more global information.

S4: Input the influenza case data and spatial scale information into the long short-term memory network (Long short-term memory, LSTM), and extract the time series relationship of the influenza case data through the long short-term memory network;

In this step, the long-short-term memory network is a recurrent neural network (Recurrent Neural Network, RNN) that can be used to process sequence data, which can solve the problem of gradient disappearance and gradient explosion during long sequence training. Compared with ordinary RNN, LSTM can achieve better performance in longer sequences. LSTM includes three stages of forgetting, selective memory, and output. The forgetting stage is used to selectively forget the input transmitted by the previous node. The memory selection stage is used to selectively "memorize" the input of the current stage. The output stage is used to determine which will be regarded as the output of the current state, and scale the output obtained in the previous stage through a tanh activation function. The LSTM calculation formula can be expressed as:

y _i,t ＝LSTM(h _i,tn ,h _i,tn ,...,h _i,t-1 ) (9)

Among them, h _i,t-1 represents the influenza case data of the i-th region in the t-1 time period, and y _i,t represents the predicted influenza case data of the i-th region in the t-th time period.

In the embodiment of the present application, the time series relationship of influenza case _data extracted through the long-term short-term memory network is specifically: the long-term short-term memory network calculates _the current The output h _t of the hidden layer at all times, and newly added the input gate _it , the forgetting gate f _t , the output gate o _t and the memory unit c _t . Among them, the input gate is obtained by linearly transforming the input x _t and the output h _t-1 of the hidden layer of the previous step, and then calculated by the activation function, which is used to control the extent to which the new state of the current calculation is updated to the memory unit. The forget gate and output gate are calculated in a similar way to the input gate, which are used to control how much the information in the memory unit of the previous step is forgotten and how much the current output depends on the current memory unit. The input gate, the forget gate, and the output gate all have their own parameters W and b, and the memory unit is mainly controlled by the input gate and the forget gate: the input gate controls the information that needs to be memorized in the current input sequence, and the forget gate controls the information in the previous historical memory. Information that needs to be forgotten. The output ht of the hidden layer of the long-term short-term memory network at time _t is finally determined by the output gate and the memory unit. The update formula of each calculation unit of the long-term short-term memory network is as follows:

i _t =σ(W _i [h _t-1 ,x _t ]+b _i ) (10)

f _t =σ(W _f [h _t-1 ,x _t ]+b _f ) (11)

o _t =σ(W _o [h _t-1 ,x _t ]+b _o ) (12)

h _t ＝o _t *tanh(C _t ) (15)

Based on the above, for the information X _t at each input moment, first use the graph neural network to extract the spatial scale information of each region, and then use the newly obtained input sequence of each region as the input of the long short-term memory network, through training and learning to obtain Hidden features corresponding to each region. At this time, each region shares the same long-term short-term memory network model, which helps to improve the generalization ability of the model, while reducing model parameters and reducing the complexity of the model. While learning the original features in the time series relationship, it is also making full use of the introduced spatial context information to make the prediction model better. As the number of network layers deepens, the final node function will capture more and more global information. However, in order to preserve local intermediate information, the present invention splices the hidden states H ₁ and H ₂ of the last time step of the two LSTM layers with the input historical information, and the formula is as follows:

H＝CONCAT(X _tn ，C _t-n+1 ，…,C _t-1 ,H ₁ ,H ₂ ) (16)

The rows of matrix H can be viewed as vertex representations that encode multi-scale structural information (including initial features of nodes), and then pass this vertex representation to the output layer consisting of a two-layer fully-connected network.

S5: According to the spatial scale information and time series relationship of influenza case data, the urban influenza incidence trend prediction results are obtained;

Among them, since the embodiment of the present application uses artificial intelligence deep learning to predict the trend of influenza incidence, 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 influenza incidence trend prediction method of the embodiment of the present application extracts the population movement relationship through the individual movement trajectory data in the city. The memory network extracts the time series relationship of influenza case data, and makes refined predictions of influenza trends based on spatial scale information and time series relationship. The embodiment of this application realizes the prediction of the development trend of influenza in the city with higher spatial resolution, completes the refined analysis of influenza, helps the government and public health departments to timely and accurately understand the development trend of influenza in the city, and conduct targeted Epidemic prevention and control intervention can maximize the protection of people's lives, health and safety. Compared with the prior art, the present application has at least the following beneficial effects:

(1) After integrating large-scale mobile phone location data, the defect that the traditional theoretical model method cannot be refined to simulate the population movement and interaction within the city has been improved, and the mobility of individuals has been more realistically restored.

(2) Compared with other data-driven modeling methods such as the spatial interaction modeling method based on OD flow, the spatial interaction modeling method of the family center flow can be constructed at a higher spatial resolution and at the urban scale. The graph structure traffic is more concentrated, which can obviously capture multiple strong commuting traffic, and then effectively improve the prediction accuracy of the deep learning model of influenza within the city.

(3) The present invention only needs to obtain the location information of the mobile phone, and does not need to obtain multi-dimensional information such as weekly average temperature, air pressure, rainfall, relative humidity, maximum temperature difference, and sunshine time, which reduces the difficulty of data acquisition and processing.

Please refer to FIG. 2 , which is a schematic structural diagram of an urban influenza incidence trend prediction system according to an embodiment of the present application. The urban influenza incidence trend prediction system 40 of the embodiment of the present application includes:

Data collection module 41: used to obtain data on influenza cases in the city, and collect data on individual movement trajectories in the city;

Data processing module 42: used to process the individual movement trajectory data set through a data-driven method, obtain the home address of each individual, and set the home address as the starting point of all movement trajectories of each individual, and extract each individual from Flow data from the starting point to other regions to obtain the population movement relationship between regions;

Spatio-temporal feature extraction module 43: used to extract the spatial scale information of influenza case data by using the graph neural network based on the population movement relationship, and extract the time series relationship of the influenza case data by using the long-term short-term memory network;

Influenza prediction module 44: used to obtain urban influenza incidence trend prediction results based on the spatial scale information and time series relationship of influenza case data.

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 method for predicting the incidence trend of urban influenza.

The processor 51 is used to execute the program instructions stored in the memory 52 to control the prediction of urban flu incidence trends.

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 (10)

1. A method for predicting the incidence trend of urban influenza, characterized in that it includes:

   Obtain the data of influenza cases in the city, and collect the data of individual movement trajectories in the city;

   Process the individual movement trajectory data through a data-driven method, obtain the home address of each individual, set the home address as the starting point of each individual movement trajectory, and extract each individual from the starting point to Flow data to other regions to obtain the population movement relationship between regions;

   Based on the population movement relationship, using a graph neural network to extract the spatial scale information of the influenza case data, and using a long short-term memory network to extract the time series relationship of the influenza case data;

   According to the spatial scale information and time series relationship of the influenza case data, the urban influenza incidence trend prediction result is obtained.
2. The urban influenza incidence trend prediction method according to claim 1, wherein the collection of individual movement track data in the city comprises:

   Use mobile devices to collect individual movement trajectory data in the city; the mobile devices include mobile phones or smart watches; the individual movement trajectory data include each individual's mobile phone number, signaling time stamp, and base station latitude and longitude.
3. The method for predicting the incidence trend of urban influenza according to claim 2, wherein the processing of the individual movement trajectory data through a data-driven method is specifically:

   Merge the original Thiessen polygons established by the city base stations in the grid area of the set time period and the set distance; divide the individual movement track data according to the stay time, and set the position with the longest night stay time of each individual Set as the individual's home address, and extract the OD flow and home center flow at a set interval in a grid area with a set distance.
4. The method for predicting the incidence trend of urban influenza according to claim 1, wherein the processing of the individual movement track data by a data-driven method also includes:

   According to the population distribution information and geographical location information of each street in the city in the census data, the network flow extraction is performed on the individual movement track data through a theoretical model method; the theoretical model method includes a gravity model, a radiation model or a spatial proximity relationship model.
5. The urban influenza incidence trend prediction method according to any one of claims 1 to 4, wherein the extraction of the spatial scale information of the influenza case data using a graph neural network is specifically:

   The framework of the graph neural network is a message propagation neural network, and the message propagation neural network extracts spatial scale information based on spatial graph convolution; an undirected graph G is defined, the feature vector of a node v is x _v , and the feature of an edge is e _vw , connecting nodes v and w, N(v) represents the neighbor nodes of node v in graph G, t is the running time step, and the feature x _v of node v is used as the initial state of its hidden state

   

   After that, the update of the hidden state by the spatial graph convolution is expressed as:

   

   

   The message propagation neural network decomposes the spatial graph convolution into two parts of message delivery and state update operation, which are respectively completed by the message function _M1 and the node update function _U1 ; the message function _M1 is used to aggregate the characteristics of neighboring nodes , forming a message vector, which is ready to be delivered to the central node; the node update function U _l is used to update the node representation at the current moment, and combine the node representation at the current moment with the message obtained from the message function to obtain spatial scale information.
6. The method for predicting the incidence of influenza in influenza cities according to claim 5, wherein the spatial scale information of the influenza case data extracted using a graph neural network is specifically:

   Transform the individual movement trajectory data of a set time range into a weighted directed graph, the vertices represent street-level areas, and the edges are used to capture movement patterns; at time t, the flow between areas u and v forms an edge, which is multiplied by The number of cases in region u at time t

   

   Indicates how many infected people may move from area u to area v; set

   

   is a vector of node attributes containing the number of cases in each of the past w weeks in region u; the message passed through the message propagation neural network computes a feature vector for each region using the composite score from all regions:

   

   where A represents the adjacency matrix of regional population movement flows, X _t is a matrix whose rows contain the attributes of different regions; x _u ∈ R ^w is a vector combining the number of influenza cases moving within and towards region u, x The expression formula of _u is:

   x _u ＝(x _u w _j,u +x _u w _i,u +….x _u w _v,u )+x _u w _u,u

   where x _u ∈ R ^w denotes the estimate of the number of new potential cases in area u.
7. According to claim 6, the method for predicting the incidence trend of influenza in urban influenza, is characterized in that, the time series relationship of the influenza case data extracted using the long-short-term memory network is specifically:

   The long-short-term memory network calculates the output ht of the hidden layer at the current moment based on the input _xt at the current moment and the output ht _-1 of the hidden layer at the previous time period. The calculation formula of the long-short-term memory network is _:

   y _i,t ＝LSTM(h _i,tn ,h _i,tn ,...,h _i,t-1 )

   Among them, h _i,t-1 represents the influenza case data of the i-th region in the t-1 time period, and y _i,t represents the predicted influenza case data of the i-th region in the t-th time period.
8. An urban influenza incidence trend forecasting system is characterized in that it includes:

   Data collection module: used to obtain the data of influenza cases in the city and collect the data of individual movement trajectories in the city;

   Data processing module: used to process the individual movement trajectory data through a data-driven method, obtain the home address of each individual, set the home address as the starting point of each individual movement trajectory, and extract each individual From the flow data of the starting point to other regions, the population movement relationship among the regions is obtained;

   Spatio-temporal feature extraction module: used to extract the spatial scale information of the influenza case data by using the graph neural network based on the population movement relationship, and extract the time series relationship of the influenza case data by using the long short-term memory network;

   Influenza prediction module: used to obtain urban influenza incidence trend prediction results based on the spatial scale information and time series relationship of the influenza case data.
9. 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 influenza incidence trend 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 prediction of flu incidence trends in flu cities.
10. A storage medium, which is characterized by storing program instructions executable by a processor, and the program instructions are used to execute the urban influenza incidence trend prediction method according to any one of claims 1 to 7.

Images