Infectious disease transmission simulation-prediction integrated modeling method based on ensemble population
By combining infectious disease dynamics equations and neural networks, a multi-regional interaction network is constructed, which solves the problems of lack of interpretability and overfitting in traditional models, and achieves high-precision simulation and prediction of infectious disease transmission, supporting public health decision-making.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SICHUAN UNIV
- Filing Date
- 2026-01-28
- Publication Date
- 2026-04-17
AI Technical Summary
Traditional infectious disease transmission models lack interpretability, are prone to overfitting, and often produce predictions that defy common sense in small sample scenarios, making it difficult to accurately simulate and predict the risk of cross-regional transmission of infectious diseases.
An integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations is adopted. By embedding the infectious disease dynamics equation into a neural network and combining real data from multiple regions and population flow data, an explicit and implicit interaction network is constructed for feature extraction and fusion. The disease transmission dynamics differential equation is then used for reconstruction and prediction.
It enables reliable prediction of infectious disease transmission, possesses parametric interpretability, can accurately simulate the transmission process in unknown areas, and can be promoted and applied in different regions to provide scientific public health decision support.
Smart Images
Figure CN121885207A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of infectious disease surveillance and prediction technology, and in particular to an integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations. Background Technology
[0002] In the context of globalization and high urbanization, infectious diseases have evolved into an extremely complex spatial interaction system. Modeling methods based on aggregate populations, which treat different geographical regions as network nodes and couple them using population flow data, have become a strategic cornerstone for capturing spatial heterogeneity and assessing the risk of cross-regional transmission.
[0003] Traditional research has long relied on data-driven models, which are essentially "black boxes" lacking interpretability and often produce predictions that defy common sense due to the lack of physical constraints (such as negative infection numbers or population non-conservation). In early, small-sample scenarios, they are prone to overfitting. Summary of the Invention
[0004] The purpose of this application is to provide an integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations. By embedding the infectious disease dynamics equation as prior knowledge into the loss function or architecture of a neural network, it provides reliable predictions for infectious disease transmission and endows the model with parameter-level interpretability. This allows the method to clearly quantify the risk of infectious disease transmission and provides support for public health decision-making that combines high accuracy and scientific logic.
[0005] To achieve the above objectives, this application provides the following solution: A method for integrated simulation and prediction of infectious disease transmission based on ensemble populations, the method comprising: S1: Preprocess the acquired historical data on the spread of infectious diseases to obtain real data from multiple regions; S2: Construct a known region node representation based on the real data of the multi-region, and fill in the infectious disease data of the adjacent unknown regions with the known region node representation to obtain the node representation of the adjacent unknown regions; S3: Perform feature extraction and fusion on the known region node representations, unknown region node representations, and acquired environmental data to obtain temporal fusion features. ; S4: Based on the obtained multi-regional real data, acquire regional adjacency data and regional population flow data, and combine the regional adjacency data and regional population flow data to construct an explicit interaction relationship network between regional nodes. The explicit interaction relationship network Used to obtain spatial features of multiple regions; Based on time fusion features Constructing an implicit interaction network between regional nodes The implicit interaction network Used to obtain temporal characteristics of multiple regions; The spatial and temporal features of the multiple regions are fused to obtain multi-layer fused spatiotemporal features; S5: Reconstruct and decode the multi-layer fused spatiotemporal features and perform predictive decoding to obtain infectious disease data for all regions at time t after reconstruction. and infectious disease transmission parameters ( , ); S6: Obtain the reconstructed infectious disease data for all regions at time t. and infectious disease transmission parameters ( , Input the differential equation of disease transmission dynamics Solve the problem and output the simulation results of infectious disease data in unknown areas and the prediction results of future spread in all areas.
[0006] Furthermore, the step of constructing known region node representations based on the real data from the multiple regions, and then filling in the infectious disease data of adjacent unknown regions with these known region node representations to obtain node representations of adjacent unknown regions includes: The real data of each region is used as the corresponding node data, and a propagation dynamics model based on the population SIR of the current region is established based on the node data to obtain the node representation of the known region; By using the node representations of known regions, the infectious disease data of adjacent unknown regions are filled in using a filling model to obtain the node representations of adjacent unknown regions.
[0007] Furthermore, the filling model includes: ; in, —Nodes under channel D At any moment The Dimensional filling features; Channel D = {S, I}, where S represents the number of susceptible individuals and I represents the number of infected individuals; —Calculate the fill options; like If =1, then the fill value is calculated based on the geographic spatial adjacency relationship; Then the fill value is calculated based on the inflow of people; Then, the fill value is calculated based on the dynamic time-normalized similarity of the environmental data; Then, the fill value is calculated based on the similarity of humanities data.
[0008] Furthermore, the known region node representations, unknown region node representations, and acquired environmental data are subjected to feature extraction and fusion to obtain temporal fusion features. The steps include: By combining known region nodes, unknown region nodes, and corresponding environmental data, feature extraction is performed separately to obtain corresponding data features. Then, a gating fusion mechanism is used to fuse the obtained data features to obtain the temporal fusion feature. ; ; ; in, —Infectious disease data features obtained based on the known region node representation and the unknown region node representation; —Humanities-related data features obtained based on the known region node representations and unknown region node representations; —Characteristics of environmental data; FC—Fully connected layer in a neural network; —The pass rate of information for the feature; — An embedded representation that integrates features from multiple data sources.
[0009] Furthermore, the step of fusing the spatial and temporal features of the multiple regions to obtain multi-layered fused spatiotemporal features includes: And combined with the explicit interaction relationship network between the regional nodes and implicit interaction networks Constructing a network of interactions among populations, including: ; ; ; ; in, —Population flow relationships between regions This indicates that p people from region i moved to region j; —The geographical adjacency between regions. This indicates that region i and region j are adjacent on the map and share the same regional boundary line; —The weight matrix corresponding to the explicit interaction relationship network, and the values of its elements. The degree of explicit interaction between region i and region j; The similarity matrix corresponding to the implicit interaction relationship network at time t; —Hadamard product, which represents the multiplication of corresponding elements in a matrix; —A matrix consisting entirely of 1s; FC—Fully connected layer in a neural network; —Weight of explicit population associations; —A matrix of interaction relationships among aggregate populations that integrates spatial and temporal characteristics; —SoftMax activation function —Sigmoid activation function; Based on the interaction matrix among ensemble populations, a graph neural network method is used to extract multi-layer spatiotemporal features of infectious diseases. ; Among them, GCN—Graph Neural Network; — Output of the l-th layer GCN .
[0010] Furthermore, the reconstruction decoder in S5 includes: Obtain the intermediate output of each GCN module layer. and the final output of the last layer Feature compression and mapping are performed through skip connections and 1×1 convolutional layers to output the reconstructed infectious disease data of all regions at time t. ; Among them, CNN stands for Convolutional Neural Network; —Reconstructed infectious disease data for all regions at time t.
[0011] Furthermore, the prediction decoder in S5 includes: Output from the last layer of the GCN module Generate hidden representations by combining time-aware modules. ; The time-aware module perceives the periodic time structure by performing a Fourier transform on time t. ; ; By hiding the representation Predictive decoding is performed to obtain infectious disease transmission parameters ( , ): ; In the formula, The SoftPlus activation function is used to constrain the non-negativity of infectious disease dynamics parameters. It is a fully connected layer in a multi-layer neural network.
[0012] Furthermore, the differential equation of disease transmission dynamics include: ; ; ; ; In the formula —The total number of people in region i, i∈[1,n]; —The number of susceptible individuals in the region; —Number of infections; —Number of people recovered; — The parameter value corresponding to region i in the middle; — The parameter value corresponding to region i in the middle; —The number of people flowing from region i to region j, where j∈[1,n]; n—Total number of regions.
[0013] Furthermore, the steps of simulating the output of infectious disease data in unknown areas and predicting the future spread of the disease in all areas include: The reconstructed infectious disease data for all regions at time t. and infectious disease transmission parameters ( , Substitute them together into the differential equation The future propagation state of region i is obtained by solving the problem. : ; In the formula, —Differential equation solver; —Infectious disease transmission rates in various regions; —Infectious disease removal rate in each region.
[0014] According to the specific embodiments provided in this application, this application has the following technical effects: By reconstructing branches, the algorithm no longer relies solely on the historical data of a single node, but instead uses the aggregated information of neighboring nodes in the graph structure and its own static environmental characteristics to make comprehensive inferences.
[0015] It can generate reconstruction results based on the inherent "propagation-driven mechanism", thus having the ability to handle unknown nodes (zero samples); by introducing the physical propagation mechanism specification, the model output is forced to meet the physical propagation law, thereby enhancing the model's generalization ability; the combination of the two enables the present invention to more accurately predict the spread of diseases while simulating the spread in unknown areas, and combined with regional environmental characteristics, it is easier to generalize to different regions and different diseases, thus providing technical support for disease prevention. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a flowchart of an integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations, as described in one embodiment of this application. Figure 2 This is a structural diagram of an integrated infectious disease transmission simulation-prediction model in one embodiment of this application. Detailed Implementation
[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0019] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0020] like Figure 1 As shown, this embodiment provides an integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations, including: S1: Preprocess the acquired historical data on the spread of infectious diseases to obtain real data from multiple regions; In one specific embodiment, environmental data, regional adjacency and regional population flow data within the corresponding time and area are collected through the official meteorological data platform and the government information disclosure data platform, and the data is preprocessed. The official meteorological data platform provides environmental data for n regions, including climate data such as temperature, relative humidity, rainfall, and wind speed at two meters above the ground. Human data includes land area, population density, residents' income level, and socioeconomic level. This data, after preprocessing operations such as cleaning and normalization, is used as input features for the model. The data normalization formula is as follows: ; in, This represents the normalized environmental data. Represents environmental data. This represents the mean of environmental data. This represents the variance of environmental data; Regional cultural data The processing method is as follows: ; in, This represents the normalized regional human data. Represents regional human data, This represents the mean of regional human data. This represents the variance of regional human data; Simultaneously, based on data collected from official government platforms regarding the borders between regions on the map, as well as the total population within the region from the census and population movement data for the past year, borders are established using regions as nodes and existing border relationships on the map as edges (i...). ), construct a regional adjacency network (matrix) Using regions as nodes and directed edges representing the direction of population flow (i... The population flow network (matrix) is constructed by using the number of migrants, p, as the edge weights. .
[0021] S2: Construct a known region node representation based on the real data of the multi-region, and fill in the infectious disease data of the adjacent unknown regions with the known region node representation to obtain the node representation of the adjacent unknown regions; S3: Perform feature extraction and fusion on the known region node representations, unknown region node representations, and acquired environmental data to obtain temporal fusion features. ; S4: Based on the obtained multi-regional real data, acquire regional adjacency data and regional population flow data, and combine the regional adjacency data and regional population flow data to construct an explicit interaction relationship network between regional nodes. The explicit interaction relationship network Used to obtain spatial features of multiple regions; Based on time fusion features Constructing an implicit interaction network between regional nodes The implicit interaction network Used to obtain temporal characteristics of multiple regions; The spatial and temporal features of the multiple regions are fused to obtain multi-layer fused spatiotemporal features; S5: Reconstruct and decode the multi-layer fused spatiotemporal features and perform predictive decoding to obtain infectious disease data for all regions at time t after reconstruction. and infectious disease transmission parameters ( , ); S6: Obtain the reconstructed infectious disease data for all regions at time t. and infectious disease transmission parameters ( , Input the differential equation of disease transmission dynamics Solve the problem and output the simulation results of infectious disease data in unknown areas and the prediction results of future spread in all areas.
[0022] S6: Combining bi-branch reconstruction decoding and prediction decoding yields reconstruction and prediction results. Based on these results, the model parameters are jointly adjusted to obtain an optimized infectious disease transmission simulation-prediction model (e.g., ...). Figure 2 As shown in the figure, the output of the simulation of infectious disease data in unknown areas and the prediction of future spread in all areas are based on the optimized infectious disease transmission simulation-prediction model.
[0023] In this embodiment, by combining the "propagation-driven mechanism" to generate reconstruction results and by introducing the physical propagation mechanism specification, the model output is forced to meet the physical propagation law, thereby enhancing the model's generalization ability. This allows the application to more accurately predict the disease propagation process while simulating the propagation in unknown areas, and is more easily generalized to different regions and different diseases by combining regional environmental characteristics, thus providing technical support for disease prevention.
[0024] In one embodiment provided in this application, the step of constructing a known region node representation based on the real data of the multiple regions, and then filling the infectious disease data of adjacent unknown regions with the known region node representation to obtain the node representation of the adjacent unknown regions includes: The real data of each region is used as the corresponding node data, and a propagation dynamics model based on the population SIR of the current region is established based on the node data to obtain the node representation of the known region; By using the node representations of known regions, the infectious disease data of adjacent unknown regions are filled in using a filling model to obtain the node representations of adjacent unknown regions.
[0025] In one embodiment provided in this application, the filling model includes: ; in, —Nodes under channel D At any moment The Dimensional filling features; Channel D = {S, I}, where S represents the number of susceptible individuals and I represents the number of infected individuals; —Calculate the fill options; like If =1, then the fill value is calculated based on the geographic spatial adjacency relationship; Then the fill value is calculated based on the inflow of people; Then, the fill value is calculated based on the dynamic time-normalized similarity of the environmental data; Then, the fill value is calculated based on the similarity of humanities data.
[0026] Specifically, a propagation dynamics model based on population SIR was used to process the collected real data from multiple regions. The population was divided into three categories according to its state during the propagation process, and differential equations were used to describe the changes in the number of these three categories of population over time. S (Susceptible) - Susceptible individuals refer to those who have not yet been infected (or have not received information) but lack immunity and are likely to be infected upon contact with an infected person.
[0027] I (Infective) - An infected person refers to a group of people who have been infected (or have received information) and have the ability to spread the disease (or information) to susceptible individuals.
[0028] R (Recovered / Removed) - This refers to individuals who have been removed from the infected population. This typically includes those who have gained permanent immunity through recovery, as well as those who have died or been effectively isolated and no longer participate in transmission. In information dissemination models, this usually represents those who no longer spread information.
[0029] Based on the possibility of infection data in some areas of the real world, and considering the situation of missing infectious disease data in areas (nodes), in the specific implementation of step S2, 10% of the areas (rounded) are set as missing areas; under different channels D={S, I}, the geographically adjacent ( ), Crowd inflow ( Similarity calculation relationship () Selecting nodes Related Each node, through averaged observation data For nodes Fill in, For channel D={S, I}, the first Each node at the corresponding time Observed infectious disease data (Channel S: number of susceptible individuals, Channel I: number of infected individuals).
[0030] The infectious disease transmission data characteristics of the missing regions are filled in by combining the number of infected individuals (I) in each region (node), the number of infected individuals removed (R), and the number of remaining susceptible individuals (S) in the population. This includes: This embodiment constructs filling features based on four dimensions (j=1 to j=4): geographic spatial adjacency, population inflow, dynamic temporal regularization similarity of environmental data, and similarity of human data. This comprehensively considers both explicit and implicit similarities between regions, avoiding the limitations of single-dimensional data and making the filling of infectious disease data (susceptible population S, infected population I) in unknown areas more closely reflect actual transmission scenarios. Based on observational data from known areas, the unknown areas are filled using averaging calculations. This utilizes the interaction and correlation information between aggregate populations and ensures the independence and accuracy of data from different transmission states through explicit channel division (D={S,I}), providing high-quality input data for subsequent feature fusion and model training, and reducing the impact of missing data on model performance. The filling process comprehensively considers multiple core factors influencing the spread of infectious diseases, including geography, population, environment, and human factors, enabling the model to adapt to the transmission differences in different regions (such as urban and rural areas, economically developed and underdeveloped areas). This lays a solid foundation for subsequent simulation of the spread in unknown areas and improves the model's applicability in diverse scenarios.
[0031] In one embodiment provided in this application, temporal fusion features are obtained in step S3. It also includes: The processed infectious disease data and environmental data are then used to extract features from time series models of different branches of a neural network model. ; ; ; In the formula, and The data consisted of normalized human and environmental data, and time-series features were extracted using a gated recurrent unit (GRU) module. For multi-channel infectious disease data, multi-channel temporal features are extracted using a multi-channel friendly temporal convolutional network (TCN).
[0032] Subsequently, a gating fusion mechanism is used to aggregate and fuse features from multiple data sources: By combining known region nodes, unknown region nodes, and corresponding environmental data, feature extraction is performed separately to obtain corresponding data features. Then, a gating fusion mechanism is used to fuse the obtained data features to obtain the temporal fusion feature. ; ; ; in, —Infectious disease data features obtained based on the known region node representation and the unknown region node representation; —Humanities-related data features obtained based on the known region node representations and unknown region node representations; —Characteristics of environmental data; FC—Fully connected layer in a neural network; —The pass rate of information for the feature; — An embedded representation that integrates features from multiple data sources.
[0033] In this embodiment, features of infectious disease data, humanistic data, and environmental data are extracted separately through different branches of the neural network. This targeted capture of core information for each type of data (such as the temporal evolution patterns of infectious disease data and the climatic impact characteristics of environmental data) avoids information redundancy or interference caused by directly splicing multi-source data. The information throughput z, calculated using the sigmoid function, dynamically adjusts the contribution of each feature, enabling the model to automatically focus on key influencing factors at different stages of transmission (such as the initial outbreak, stable period, and decline period). (For example, during the outbreak, the focus is on infectious disease data features; during the stable period, the focus is on environmental and humanistic data features), thus improving the temporal fusion features. The relevance and effectiveness of this approach. It fuses multi-source data into a unified embedded representation h. t This enables data of different types and dimensions to form standardized features, which facilitates the construction of subsequent interaction relationship networks and the extraction of spatiotemporal features, reduces the impact of data format differences on model calculation, and improves the overall running efficiency of the model.
[0034] In one embodiment provided in this application, the step of fusing the spatial and temporal features of the multiple regions to obtain multi-layer fused spatiotemporal features includes: And combined with the explicit interaction relationship network between the regional nodes and implicit interaction networks Constructing a network of interactions among populations, including: ; ; ; ; in, —Population flow relationships between regions This indicates that p people from region i moved to region j; —The geographical adjacency between regions. This indicates that region i and region j are adjacent on the map and share the same regional boundary line; —The weight matrix corresponding to the explicit interaction relationship network, and the values of its elements. The degree of explicit interaction between region i and region j; The similarity matrix corresponding to the implicit interaction relationship network at time t; —Hadamard product, which represents the multiplication of corresponding elements in a matrix; —A matrix consisting entirely of 1s; FC—Fully connected layer in a neural network; —Weight of explicit population associations; —A matrix of interaction relationships among aggregate populations that integrates spatial and temporal characteristics; —SoftMax activation function —Sigmoid activation function; Based on the interaction matrix among ensemble populations, a graph neural network method is used to extract multi-layer spatiotemporal features of infectious diseases. ; Among them, GCN—Graph Neural Network; — Output of the l-th layer GCN ; These are trainable parameters.
[0035] In this embodiment, an explicit interaction network is constructed using regional adjacency and population flow data to capture objectively existing regional relationships; and temporal fusion features are utilized. Constructing implicit interaction networks and uncovering potential propagational connections (such as implicit propagation patterns in areas of similar environment) allows for a more comprehensive and realistic depiction of the interaction relationships among ensemble populations. This is achieved through M... t(Explicit association weights) dynamically balance the contributions of explicit and implicit interaction networks, adaptively adjusting according to the propagation characteristics at different time points (e.g., emphasizing explicit networks during peak population flows and implicit networks during periods of slower propagation), thus enhancing the adaptability of the interaction network to spatiotemporal dynamic changes. Multi-layered feature extraction of network nodes based on graph neural networks (GCN) can gradually uncover deep spatiotemporal associations among aggregate populations (e.g., short-term propagation trends and long-term diffusion patterns), combining local regional propagation characteristics with global interaction associations. This provides more representative feature support for subsequent reconstruction decoding and prediction decoding, improving the model's accuracy in characterizing the dynamics of infectious disease transmission.
[0036] In one embodiment provided in this application, the reconstruction decoder in S5 includes: Obtain the intermediate output of each GCN module layer. and the final output of the last layer Feature compression and mapping are performed through skip connections and 1×1 convolutional layers to output the reconstructed infectious disease data of all regions at time t. ; ; Among them, CNN stands for Convolutional Neural Network; —Reconstructed infectious disease data for all regions at time t.
[0037] In this embodiment, skip connections are used to integrate the intermediate and final outputs of each layer of the GCN, making full use of the spatiotemporal features of different layers (shallow features capture local propagation details, while deep features characterize global propagation patterns). This avoids the information limitations of single-layer features, enabling the reconstructed infectious disease data (S, I) at time t to more accurately reflect the actual propagation state. A 1×1 convolutional layer compresses and maps multi-layer features, reducing feature dimensional redundancy while strengthening key propagation information. This makes the reconstructed output more focused on core propagation features, reduces noise interference, and improves the reliability of the reconstruction results, providing high-quality initial state data for subsequent prediction and decoding. The system can reconstruct infectious disease data for all regions (including known regions and filled unknown regions), not only verifying the rationality of known region data but also providing more accurate propagation state simulations for unknown regions. This solves the model training bias problem caused by missing data in some regions and enhances the model's adaptability to scenarios with incomplete data.
[0038] In one embodiment provided in this application, the prediction decoder in S5 includes: Output through the last layer of the GCN module Generate hidden representations by combining time-aware modules. ; The time-aware module perceives the periodic time structure by performing a Fourier transform on time t. ; ; By hiding the representation Predictive decoding is performed to obtain infectious disease transmission parameters ( , ): ; In the formula, The SoftPlus activation function is used to constrain the non-negativity of infectious disease dynamics parameters. It is a fully connected layer in a multi-layer neural network.
[0039] By processing time t using Fourier transform, a time feature Φ containing sine and cosine components is generated. t It can accurately capture the periodic patterns of infectious disease transmission (such as the seasonal cycle of seasonal influenza and the weekly fluctuations in the spread of respiratory diseases), providing key time dimension support for prediction and decoding.
[0040] Output the infectious disease transmission rate β using a multilayer perceptron (MLP). t and removal rate δ t Furthermore, the SoftPlus activation function is used to constrain the non-negativity of the parameters, ensuring that the parameters conform to the laws of physical propagation. At the same time, the explicit propagation parameters make the model no longer a "black box," enabling it to quantify the propagation intensity and recovery efficiency in different regions and at different times, providing interpretable key indicators for public health decision-making.
[0041] Hidden representation Integrates time fusion features The final output and temporal features of GCN comprehensively cover core information such as the historical state of infectious disease transmission, population interaction, and temporal patterns, providing sufficient feature support for transmission parameter prediction, making the predicted parameters more consistent with the actual transmission scenario, and improving the accuracy of subsequent differential equation solutions.
[0042] In one embodiment provided in this application, the differential equation of disease transmission dynamics include: ; ; ; ; In the formula —The total number of people in region i, i∈[1,n]; —The number of susceptible individuals in the region; —Number of infections; —Number of people recovered; — The parameter value corresponding to region i in the middle; — The parameter value corresponding to region i in the middle; —The number of people flowing from region i to region j, where j∈[1,n]; n—Total number of regions.
[0043] Furthermore, the steps of simulating the output of infectious disease data in unknown areas and predicting the future spread of the disease in all areas include: The reconstructed infectious disease data for all regions at time t. and infectious disease transmission parameters ( , Substitute them together into the differential equation The future propagation state of region i is obtained by solving the problem. : ; In the formula, —Differential equation solver; —Infectious disease transmission rates in various regions; —Infectious disease removal rate in each region.
[0044] In this embodiment, differential equation constraints are used to ensure the physical rationality of the prediction. Continuous prediction is performed based on the SIR propagation dynamics differential equation of the ensemble population, forcing the prediction results to follow the basic physical laws of infectious disease transmission (such as population conservation, non-negative infection numbers, and dynamic equilibrium between transmission and recovery). This avoids illogical prediction results that might occur in purely data-driven models (such as negative infection numbers and population non-conservation), thus improving the scientific rigor and reliability of the prediction results. The initial values of the joint reconstruction decoder (fitting the actual propagation state) and the propagation parameters of the prediction decoder (precise β) are used. t δ t Solving the differential equations enables a dual-precise "state-parameter" driving force, ensuring that future propagation predictions conform to both historical propagation trends and dynamic laws, significantly improving prediction accuracy, especially in early, small-sample scenarios where overfitting is effectively reduced. The differential equations are modeled separately for each region i, combined with the total population N in that region. iPopulation mobility M ij Personalized information can output the dynamic trends of the number of susceptible people, the number of infected people, and the number of recovered people in each region, providing precise quantitative basis for different regions to formulate differentiated prevention and control strategies (such as strengthening control in areas with high transmission rates and increasing medical resource investment in areas with high infection numbers), thereby improving the pertinence and effectiveness of prevention and control decisions.
[0045] In one embodiment provided in this application, the model utilizes the reconstruction results ( ), Prediction results ( ) and the results of reconstruction and decoding The result of the inverse derivative ( These factors, together, form the loss function that constrains model training. Ultimately, the optimized model provides a simulation of infectious disease data in unknown areas and predictions of future spread across all areas. The loss function for model training is designed as follows:
[0046] in, Predicting loss for the data:
[0047] Reconstructing losses from the spread of infectious diseases:
[0048] In the formula, The set of regions with known historical propagation patterns. This represents the number of corresponding regions.
[0049] Loss due to physical propagation mechanism:
[0050] in, Controlling the strength of physical constraints, This represents the result obtained from the reconstruction decoder. The back-derivative of the neural network for the input variable time t, Let be the function on the right-hand side of the differential equation of the SIR model in the figure.
[0051] In one embodiment provided in this application, the model uses a loss function to measure the magnitude of the error between the model and actual observations in both simulation and prediction. An optimizer with good optimization performance, such as AdamW, is used to minimize the loss function value, thereby simultaneously optimizing the data errors in simulation and prediction, as well as the PINN physical residual loss. To achieve synergistic optimization, the errors of these three factors are balanced through hyperparameters, enabling the model to accurately fit the observed case data while adhering to the laws of epidemiological dynamics, thus maintaining robustness and interpretability in noisy environments and under sparse data conditions.
[0052] In summary, this application provides an integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations. On one hand, it learns the characteristics of different data types through a three-branch structure, capturing the impact of environmental factors, human factors, and historical transmission patterns on infectious disease transmission. A gating fusion mechanism couples different factors to uncover the correlation between infectious diseases and the environment and human factors. Simultaneously, it utilizes inter-regional geographical correlations, population flow correlations, and similarity correlations of environmental and human factors to jointly characterize the interaction relationships between regional populations, providing a guarantee for further simulation of infectious disease transmission in missing areas. Through the differential equation of infectious disease dynamics, this invention can more accurately and scientifically predict the disease transmission process. On the other hand, based on the easily extensible architecture of this invention, it can be effectively transferred to different diseases in different regions for prediction at a finer-grained unit.
[0053] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0054] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (Read-Only Memory). Memory includes ROM, magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0055] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0056] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0057] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for integrated modeling and prediction of infectious disease transmission based on ensemble populations, characterized in that, The method includes: S1: Preprocess the acquired historical data on the spread of infectious diseases to obtain real data from multiple regions; S2: Construct a known region node representation based on the real data of the multi-region, and fill in the infectious disease data of the adjacent unknown regions with the known region node representation to obtain the node representation of the adjacent unknown regions; S3: Perform feature extraction and fusion on the known region node representations, unknown region node representations, and acquired environmental data to obtain temporal fusion features. ; S4: Based on the obtained multi-regional real data, acquire regional adjacency data and regional population flow data, and combine the regional adjacency data and regional population flow data to construct an explicit interaction relationship network between regional nodes. The explicit interaction relationship network Used to obtain spatial features of multiple regions; Based on time fusion features Constructing an implicit interaction network between regional nodes The implicit interaction network Used to obtain temporal characteristics of multiple regions; The spatial and temporal features of the multiple regions are fused to obtain multi-layer fused spatiotemporal features; S5: Reconstruct and decode the multi-layer fused spatiotemporal features and perform predictive decoding to obtain infectious disease data for all regions at time t after reconstruction. and infectious disease transmission parameters ( , ); S6: Obtain the reconstructed infectious disease data for all regions at time t. and infectious disease transmission parameters ( , Input the differential equation of disease transmission dynamics Solve the problem and output the simulation results of infectious disease data in unknown areas and the prediction results of future spread in all areas.
2. The integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations as described in claim 1, characterized in that, The steps of constructing known region node representations based on the multi-region real data, and then filling in the infectious disease data of adjacent unknown regions with these known region node representations to obtain node representations of adjacent unknown regions include: The real data of each region is used as the corresponding node data, and a propagation dynamics model based on the population SIR of the current region is established based on the node data to obtain the node representation of the known region; By using the node representations of known regions, the infectious disease data of adjacent unknown regions are filled in using a filling model to obtain the node representations of adjacent unknown regions.
3. The integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations as described in claim 2, characterized in that, The filling model includes: ； in, —Nodes under channel D At any moment The Dimensional filling features; Channel D = {S, I}, where S represents the number of susceptible individuals and I represents the number of infected individuals; —Calculate the fill options; like If =1, then the fill value is calculated based on the geographic spatial adjacency relationship; Then the fill value is calculated based on the inflow of people; Then, the fill value is calculated based on the dynamic time-normalized similarity of the environmental data; Then, the fill value is calculated based on the similarity of humanities data.
4. The integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations as described in claim 1, characterized in that, The process involves extracting and fusing features from the known region node representations, the unknown region node representations, and the acquired environmental data to obtain temporal fusion features. The steps include: By combining known region nodes, unknown region nodes, and corresponding environmental data, feature extraction is performed separately to obtain corresponding data features. Then, a gating fusion mechanism is used to fuse the obtained data features to obtain the temporal fusion feature. ; ； ； in, —Infectious disease data features obtained based on the known region node representation and the unknown region node representation; —Humanities-related data features obtained based on the known region node representations and unknown region node representations; —Characteristics of environmental data; FC—Fully connected layer in a neural network; —The pass rate of information for the feature; — An embedded representation that integrates features from multiple data sources.
5. The integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations as described in claim 1, characterized in that, The step of fusing the spatial and temporal features of the multiple regions to obtain multi-layer fused spatiotemporal features includes: And combined with the explicit interaction relationship network between the regional nodes and implicit interaction networks Constructing a network of interactions among populations, including: ； ； ； ； in, —Population flow relationships between regions This indicates that p people from region i moved to region j; —The geographical adjacency between regions. This indicates that region i and region j are adjacent on the map and share the same regional boundary line; —The weight matrix corresponding to the explicit interaction relationship network, and the values of its elements. The degree of explicit interaction between region i and region j; The similarity matrix corresponding to the implicit interaction relationship network at time t; —Hadamard product, which represents the multiplication of corresponding elements in a matrix; —A matrix consisting entirely of 1s; FC—Fully connected layer in a neural network; —Weight of explicit population associations; —A matrix of interaction relationships among aggregate populations that integrates spatial and temporal characteristics; —SoftMax activation function; —Sigmoid activation function; Based on the interaction matrix among ensemble populations, a graph neural network method is used to extract multi-layer spatiotemporal features of infectious diseases. ； Among them, GCN—Graph Neural Network; — Output of the l-th layer GCN .
6. The integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations as described in claim 1, characterized in that, The reconstruction decoder in S5 includes: Obtain the intermediate output of each GCN module layer. and the final output of the last layer Feature compression and mapping are performed through skip connections and 1×1 convolutional layers to output the reconstructed infectious disease data of all regions at time t. ； Among them, CNN stands for Convolutional Neural Network; —Reconstructed infectious disease data for all regions at time t.
7. The integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations as described in claim 1, characterized in that, The prediction decoder in S5 includes: Output through the last layer of the GCN module Generate hidden representations by combining time-aware modules. ; The time-aware module perceives the periodic time structure by performing a Fourier transform on time t. ； ； By hiding the representation Predictive decoding is performed to obtain infectious disease transmission parameters ( , ): ； In the formula, The SoftPlus activation function is used to constrain the non-negativity of infectious disease dynamics parameters. It is a fully connected layer in a multi-layer neural network.
8. The integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations as described in claim 1, characterized in that, The differential equation of disease transmission dynamics include: ； ； ； ； In the formula —The total number of people in region i, i∈[1,n]; —The number of susceptible individuals in the region; —Number of infections; —Number of people recovered; — The parameter value corresponding to region i in the middle; — The parameter value corresponding to region i in the middle; —The number of people flowing from region i to region j, where j∈[1,n]; n—Total number of regions.
9. The integrated modeling method for infectious disease transmission simulation and prediction based on ensemble populations as described in claim 1, characterized in that, The steps for simulating the output of infectious disease data in unknown areas and predicting the future spread of the disease in all areas include: The reconstructed infectious disease data for all regions at time t. and infectious disease transmission parameters ( , Substitute them together into the differential equation The future propagation state of region i is obtained by solving the problem. : ； In the formula, —Differential equation solver; —Infectious disease transmission rates in various regions; —Infectious disease removal rate in each region.