Multi-scale tuberculosis morbidity prediction method and system

Abstract

The invention belongs to the technical field of medical information, and relates to a multi-scale tuberculosis morbidity prediction method and system.The system collects data through a data collecting and processing module to generate a spatio-temporal feature data set, and high-quality input is provided for subsequent model training; a space-time association graph structure between regions is constructed through a space-time association network construction module, and structured input is provided for a GCN-LSTM model; the GCN-LSTM model training module uses a graph convolutional neural network and a long short-term memory network to train multi-source spatio-temporal data, and is used for capturing spatial dependence between regions and a dynamic evolution trend of a time sequence; and the space-time prediction and result output module uses the trained GCN-LSTM model and the latest multi-source data to predict the tuberculosis morbidity of each region under different time scales in the future. According to the invention, the precision and real-time performance of tuberculosis morbidity prediction can be effectively improved, and technical support is provided for accurate prevention and control.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, specifically to a method and system for multi-scale prediction of tuberculosis incidence. Background Technology

[0002] In existing technologies, tuberculosis incidence prediction methods mostly rely on a single data source or model, failing to fully consider the dynamic evolution and cross-regional transmission characteristics in terms of space and time. Therefore, existing prediction models typically cannot accurately reflect the complexity of tuberculosis transmission, resulting in low accuracy of prediction results.

[0003] To this end, invention patent CN118737433A discloses a method for constructing a combined model for predicting tuberculosis incidence, including the following steps: S1, collecting monthly tuberculosis incidence data of the target area and r surrounding areas; S2, determining the weights of data from different areas based on spatial correlation analysis to form a merged data matrix; S3, constructing a SARIMA model, using the merged data matrix as input and the actual incidence rate of the target area or a certain surrounding area as output to train the SARIMA model, obtaining a trained SARIMA model; S4, constructing an SVR model, using the merged data matrix as input. The process involves several steps: S5, S6, S7, S8, S9, S10, S20, S30, S41, S52, S63, S74, S8, S9 ... This invention establishes a high-precision hybrid prediction model, in which SARIMA and SVR models are connected in parallel as the first stage, and their output is fed into the ELM model as the second stage. This hybrid structure makes the model flexible and adaptable. The hybrid model uses both statistical and machine learning models. The SARIMA model is suitable for handling linear trends, while the SVR model is suitable for handling nonlinear trends in tuberculosis incidence. The results of these two models are integrated into the ELM model, thereby significantly improving prediction accuracy. To address the problem of insufficient data, incidence data from the target area and surrounding areas are collected simultaneously, and the combined weights are determined through spatial correlation analysis to form a merged data matrix as the input to the model. The parameters of the ELM model are optimized using a sparrow search algorithm.

[0004] The above-mentioned technical solutions have made progress in combining statistical and machine learning models, but the following technical problems still exist: insufficient consideration of population mobility factors, most models treat population mobility as a static indicator or ignore its role, making it difficult to reflect the dynamic impact of changes in population mobility intensity over time on the disease transmission path and speed.

[0005] In view of this, it is very necessary for the present invention to provide a method and system for multi-scale prediction of tuberculosis incidence to overcome the above-mentioned defects in the prior art. Summary of the Invention

[0006] The purpose of this invention is to address the problem of insufficient consideration of dynamic population mobility characteristics, and to provide a method and system for designing a multi-scale tuberculosis incidence rate, in order to solve the aforementioned technical problems, in response to the technical deficiencies of the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for multi-scale prediction of tuberculosis incidence, comprising the following steps: Step S1: Multi-source dynamic data acquisition steps; in this step: Collect and integrate population flow, meteorological, socio-economic, and historical epidemic data of the target area and its surrounding areas to form a multi-source heterogeneous spatiotemporal dataset; Step S2: The step of constructing a spatiotemporal association network based on a regional association graph structure; in this step: Using regions as nodes and flow intensity as edges, a regional correlation graph is established, and spatial dependence and temporal evolution modeling are achieved through GCN and LSTM. Step S3: The step of training the model based on Bayesian optimization; in this step: The GCN-LSTM model is trained using historical feature data, and Bayesian optimization is used to automatically search for the optimal hyperparameter combination to improve prediction performance. Step S4: The step of outputting the multi-scale spatiotemporal prediction results; in this step: Input the latest multi-source data and output the predicted incidence rate of tuberculosis at different time scales in the future, so as to realize dynamic monitoring and trend early warning.

[0008] Secondly, the present invention also provides a system for multi-scale prediction of tuberculosis incidence, comprising: The data acquisition and processing module collects population flow data, meteorological and environmental data, socio-economic indicators, and historical epidemic data from the target area and its surrounding areas. This data is used to generate a spatiotemporal feature dataset that has been cleaned, formatted, and standardized, providing high-quality input for subsequent model training.

[0009] The Spatiotemporal Relationship Network Construction Module uses multi-source data to construct a spatiotemporal relationship graph structure between regions. Each region serves as a node, and population flow intensity and other features serve as edge weights. This module is used to accurately model the spatial dependencies and dynamic evolution patterns between regions, providing structured input for the GCN-LSTM model.

[0010] The GCN-LSTM model training module uses Graph Convolutional Neural Network (GCN) and Long Short-Term Memory Network (LSTM) to train multi-source spatiotemporal data, capturing spatial dependencies between regions and dynamic evolution trends of time series. It also uses Bayesian optimization algorithm to adjust key hyperparameters to improve model prediction accuracy and generalization ability.

[0011] The spatiotemporal prediction output module uses a trained GCN-LSTM model and the latest multi-source data to predict the incidence of tuberculosis in various regions at different time scales in the future, and provides decision support, disease risk warning and resource optimization services.

[0012] The modules work together to achieve multi-scale dynamic prediction of tuberculosis incidence in the target area and surrounding areas, providing real-time monitoring, trend warning and precise decision support for public health management departments, and promoting the implementation of tuberculosis prevention and control.

[0013] The beneficial effects of this invention lie in its ability to effectively model complex inter-regional population interactions by integrating multi-source heterogeneous data, including population flow data, meteorological environmental parameters, socioeconomic indicators, and historical epidemic information. This overcomes the limitations of traditional single-data-source methods in capturing spatial transmission characteristics. Utilizing the powerful spatial feature aggregation capabilities of GCN, combined with the precise temporal dynamic modeling of LSTM, the accuracy and stability of tuberculosis incidence prediction can be significantly improved. It achieves in-depth mining and dynamic capture of the spatiotemporal patterns of tuberculosis transmission, providing efficient and intelligent technical support for infectious disease monitoring, risk assessment, and prevention and control strategy formulation, and possesses significant value for widespread application. Its application not only helps public health departments achieve early warning and precise intervention but also optimizes the allocation of medical resources and improves the level of public health security, demonstrating significant socioeconomic benefits and public health value.

[0014] Therefore, it is evident that the present invention has outstanding substantive features and significant progress compared with the prior art, and the beneficial effects of its implementation are also obvious. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This is a flowchart of a multi-scale method for predicting tuberculosis incidence. Figure 2 This is a block diagram illustrating the principle of a multi-scale tuberculosis incidence prediction system. Figure 3 This is a flowchart of a multi-scale system for predicting tuberculosis incidence. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The following embodiments are explanations of the present invention, but the present invention is not limited to the following implementation methods.

[0018] Example 1: like Figure 1 As shown in the figure, this embodiment provides a method for predicting tuberculosis incidence at multiple scales, which includes the following steps: Step S1: Multi-source dynamic data acquisition steps; in this step: Collect and integrate population flow, meteorological, socio-economic, and historical epidemic data of the target area and its surrounding areas to form a multi-source heterogeneous spatiotemporal dataset; Step S2: The step of constructing a spatiotemporal association network based on a regional association graph structure; in this step: Using regions as nodes and flow intensity as edges, a regional correlation graph is established, and spatial dependence and temporal evolution modeling are achieved through GCN and LSTM. Step S3: The step of training the model based on Bayesian optimization; in this step: The GCN-LSTM model is trained using historical feature data, and Bayesian optimization is used to automatically search for the optimal hyperparameter combination to improve prediction performance. Step S4: The step of outputting the multi-scale spatiotemporal prediction results; in this step: Input the latest multi-source data and output the predicted incidence rate of tuberculosis at different time scales in the future, so as to realize dynamic monitoring and trend early warning.

[0019] Example 2: like Figure 1 As shown in the figure, this embodiment provides a method for predicting tuberculosis incidence at multiple scales, which includes the following steps: Step S1: Multi-source dynamic data acquisition steps; in this step: Collect and integrate population flow, meteorological, socio-economic, and historical epidemic data of the target area and its surrounding areas to form a multi-source heterogeneous spatiotemporal dataset; Step S2: The step of constructing a spatiotemporal association network based on a regional association graph structure; in this step: Using regions as nodes and flow intensity as edges, a regional correlation graph is established, and spatial dependence and temporal evolution modeling are achieved through GCN and LSTM. Step S3: The step of training the model based on Bayesian optimization; in this step: The GCN-LSTM model is trained using historical feature data, and Bayesian optimization is used to automatically search for the optimal hyperparameter combination to improve prediction performance. Step S4: The step of outputting the multi-scale spatiotemporal prediction results; in this step: Input the latest multi-source data and output the predicted incidence rate of tuberculosis at different time scales in the future, so as to realize dynamic monitoring and trend early warning.

[0020] In step S1: Multi-source heterogeneous data are collected at the target research scale and integrated to generate a multi-source heterogeneous spatiotemporal dataset, thereby comprehensively reflecting the spatiotemporal driving factors of tuberculosis transmission. Further, the feature set of the multi-source spatiotemporal data includes, but is not limited to: Inter-regional population mobility intensity data: By collecting inter-regional population migration data in real time, we can depict the scale and frequency of population migration between different regions and assess the potential transmission channels and intensity of cross-regional infection risks; Meteorological environmental parameter data: including time-series meteorological data such as daily precipitation, average temperature, and humidity, used to analyze the impact of natural environmental changes on tuberculosis transmission and to combine climate change analysis on the impact on tuberculosis transmission pathways; Socioeconomic indicators include resident population size, population density, regional GDP, and allocation of medical resources, to assess the impact of socioeconomic development levels on disease prevalence characteristics; Historical epidemic data: This includes the trend of tuberculosis incidence in the study area over a previous period, used as a benchmark for model training and prediction, thereby ensuring that the model can make effective predictions based on historical data.

[0021] After collecting the above data, it was organized into a multi-source spatiotemporal data feature set, which served as input for subsequent modeling and analysis. The collection and integration of multi-source heterogeneous data comprehensively reflects the spatiotemporal driving factors of tuberculosis transmission, providing complete basic data support for subsequent modeling. The collected data can reveal the potential pathways and influencing factors of tuberculosis transmission from different dimensions, ensuring the accuracy of subsequent models across multiple spatial and temporal scales. Furthermore, it can capture the combined effects of different factors on tuberculosis transmission, ensuring the comprehensiveness and accuracy of tuberculosis incidence prediction.

[0022] Example 3: like Figure 1 As shown in the figure, this embodiment provides a method for predicting tuberculosis incidence at multiple scales, which includes the following steps: Step S1: Multi-source dynamic data acquisition steps; in this step: Collect and integrate population flow, meteorological, socio-economic, and historical epidemic data of the target area and its surrounding areas to form a multi-source heterogeneous spatiotemporal dataset; Step S2: The step of constructing a spatiotemporal association network based on a regional association graph structure; in this step: Using regions as nodes and flow intensity as edges, a regional correlation graph is established, and spatial dependence and temporal evolution modeling are achieved through GCN and LSTM. Step S3: The step of training the model based on Bayesian optimization; in this step: The GCN-LSTM model is trained using historical feature data, and Bayesian optimization is used to automatically search for the optimal hyperparameter combination to improve prediction performance. Step S4: The step of outputting the multi-scale spatiotemporal prediction results; in this step: Input the latest multi-source data and output the predicted incidence rate of tuberculosis at different time scales in the future, so as to realize dynamic monitoring and trend early warning.

[0023] In step S1: Multi-source heterogeneous data are collected at the target research scale and integrated to generate a multi-source heterogeneous spatiotemporal dataset, thereby comprehensively reflecting the spatiotemporal driving factors of tuberculosis transmission. Further, the feature set of the multi-source spatiotemporal data includes, but is not limited to: Inter-regional population mobility intensity data: By collecting inter-regional population migration data in real time, we can depict the scale and frequency of population migration between different regions and assess the potential transmission channels and intensity of cross-regional infection risks; Meteorological environmental parameter data: including time-series meteorological data such as daily precipitation, average temperature, and humidity, used to analyze the impact of natural environmental changes on tuberculosis transmission and to combine climate change analysis on the impact on tuberculosis transmission pathways; Socioeconomic indicators include resident population size, population density, regional GDP, and allocation of medical resources, to assess the impact of socioeconomic development levels on disease prevalence characteristics; Historical epidemic data: This includes the trend of tuberculosis incidence in the study area over a previous period, used as a benchmark for model training and prediction, thereby ensuring that the model can make effective predictions based on historical data.

[0024] After collecting the above data, it was organized into a multi-source spatiotemporal data feature set, which served as input for subsequent modeling and analysis. The collection and integration of multi-source heterogeneous data comprehensively reflects the spatiotemporal driving factors of tuberculosis transmission, providing complete basic data support for subsequent modeling. The collected data can reveal the potential pathways and influencing factors of tuberculosis transmission from different dimensions, ensuring the accuracy of subsequent models across multiple spatial and temporal scales. Furthermore, it can capture the combined effects of different factors on tuberculosis transmission, ensuring the comprehensiveness and accuracy of tuberculosis incidence prediction.

[0025] Step S2 specifically includes the following steps: S21, Construct a region association graph: Based on the geographical location and population flow data of the target study area, a spatiotemporal relationship map between regions is constructed: Each study area is treated as a node in the graph. The node attributes include multi-source feature data of the area, such as population density, meteorological and environmental parameters, socioeconomic indicators and historical epidemic data, which are used to characterize the features of the area.

[0026] Using the intensity of population flow and geographical proximity between regions as the basis for connecting nodes, a regional association structure with spatiotemporal semantics is formed. For example, if two regions have a high frequency of population flow or are spatially adjacent, connecting edges are established in the graph to reflect potential propagation paths.

[0027] Edges between regions are defined based on factors such as population flow intensity, geographical proximity, and transportation networks. The weight of an edge can be determined by various factors, such as the frequency of population movement, physical distance, and transportation accessibility, to quantify the transmission potential and connection strength between regions. For example, if the population flow intensity between two regions is high, the edge weight between these two regions will also be large, indicating a strong transmission potential.

[0028] The constructed regional association graph can not only retain the multidimensional feature information of each region, but also explicitly model cross-regional propagation relationships, providing high-quality structured input data for subsequent graph convolutional neural network training.

[0029] S22, Introducing the graph convolutional neural network structure: Using the constructed regional association graph as input, the message passing mechanism of a graph convolutional neural network (GCN) is employed to spatially aggregate node features. This allows each regional node to simultaneously absorb its own information and that of its neighboring nodes during feature updates, capturing spatial dependencies and potential propagation paths between regions. Multi-layered GCNs can progressively fuse spatial information over a wider range, ensuring that node features encompass both internal and neighboring regional characteristics, thereby characterizing the potential propagation paths and spatial dependencies arising from cross-regional population flows.

[0030] The message passing mechanism of the graph convolutional neural network aggregates the features of neighboring nodes, allowing nodes to propagate local information throughout the graph. It learns and updates node features through multiple layers of convolutional operations. This not only captures local dependencies between nodes but also captures global structural information through multi-level aggregation, providing effective spatial modeling capabilities for spatiotemporal prediction tasks.

[0031] S23, Spatiotemporal Feature Comprehensive Modeling: The node feature sequences, aggregated in the spatial domain, are input into an LSTM for time series modeling. The LSTM, through a gating mechanism, captures the long-term dependence and short-term fluctuations of tuberculosis incidence over time, further revealing the dynamic evolution process over time, achieving spatiotemporal joint modeling of disease transmission, and forming the GCN-LSTM spatiotemporal prediction model. Simultaneously, a multilayer perceptron (MLP) is employed to enhance the nonlinear fitting capability of the spatiotemporal joint modeling, providing high-precision and reliable input features for predicting tuberculosis incidence at future time steps, thus improving prediction performance.

[0032] This step, by constructing a graph-based spatiotemporal correlation network, effectively integrates and represents the spatiotemporal features of multiple regions, enabling accurate modeling of transmission paths and inter-regional interactions. The introduction of a graph convolutional neural network ensures that the spatial relationships between nodes are fully expressed, capturing the transmission potential and dependencies of different regions. Simultaneously, the node feature sequences aggregated by the GCN are input into a Long Short-Term Memory (LSTM) network, which, through a gating mechanism, effectively captures the long-term dependence and short-term fluctuations of tuberculosis incidence over time, achieving joint modeling of spatial features and temporal evolution. Finally, a multilayer perceptron (MLP) is used to further enhance the model's nonlinear fitting capability, allowing the prediction results to more accurately reflect complex spatiotemporal dynamics. Thus, the model not only reflects the epidemic development trend within a region but also identifies cross-regional transmission paths, providing high-quality spatiotemporal structural input for predicting the dynamics of tuberculosis transmission, ensuring the model's prediction accuracy and timeliness.

[0033] Example 4: like Figure 1 As shown in the figure, this embodiment provides a method for predicting tuberculosis incidence at multiple scales, which includes the following steps: Step S1: Multi-source dynamic data acquisition steps; in this step: Collect and integrate population flow, meteorological, socio-economic, and historical epidemic data of the target area and its surrounding areas to form a multi-source heterogeneous spatiotemporal dataset; Step S2: The step of constructing a spatiotemporal association network based on a regional association graph structure; in this step: Using regions as nodes and flow intensity as edges, a regional correlation graph is established, and spatial dependence and temporal evolution modeling are achieved through GCN and LSTM. Step S3: The step of training the model based on Bayesian optimization; in this step: The GCN-LSTM model is trained using historical feature data, and Bayesian optimization is used to automatically search for the optimal hyperparameter combination to improve prediction performance. Step S4: The step of outputting the multi-scale spatiotemporal prediction results; in this step: Input the latest multi-source data and output the predicted incidence rate of tuberculosis at different time scales in the future, so as to realize dynamic monitoring and trend early warning.

[0034] In step S1: Multi-source heterogeneous data are collected at the target research scale and integrated to generate a multi-source heterogeneous spatiotemporal dataset, thereby comprehensively reflecting the spatiotemporal driving factors of tuberculosis transmission. Further, the feature set of the multi-source spatiotemporal data includes, but is not limited to: Inter-regional population mobility intensity data: By collecting inter-regional population migration data in real time, we can depict the scale and frequency of population migration between different regions and assess the potential transmission channels and intensity of cross-regional infection risks; Meteorological environmental parameter data: including time-series meteorological data such as daily precipitation, average temperature, and humidity, used to analyze the impact of natural environmental changes on tuberculosis transmission and to combine climate change analysis on the impact on tuberculosis transmission pathways; Socioeconomic indicators include resident population size, population density, regional GDP, and allocation of medical resources, to assess the impact of socioeconomic development levels on disease prevalence characteristics; Historical epidemic data: This includes the trend of tuberculosis incidence in the study area over a previous period, used as a benchmark for model training and prediction, thereby ensuring that the model can make effective predictions based on historical data.

[0035] After collecting the above data, it was organized into a multi-source spatiotemporal data feature set, which served as input for subsequent modeling and analysis. The collection and integration of multi-source heterogeneous data comprehensively reflects the spatiotemporal driving factors of tuberculosis transmission, providing complete basic data support for subsequent modeling. The collected data can reveal the potential pathways and influencing factors of tuberculosis transmission from different dimensions, ensuring the accuracy of subsequent models across multiple spatial and temporal scales. Furthermore, it can capture the combined effects of different factors on tuberculosis transmission, ensuring the comprehensiveness and accuracy of tuberculosis incidence prediction.

[0036] Step S2 specifically includes the following steps: S21, Construct a region association graph: Based on the geographical location and population flow data of the target study area, a spatiotemporal relationship map between regions is constructed: Each study area is treated as a node in the graph. The node attributes include multi-source feature data of the area, such as population density, meteorological and environmental parameters, socioeconomic indicators and historical epidemic data, which are used to characterize the features of the area.

[0037] Using the intensity of population flow and geographical proximity between regions as the basis for connecting nodes, a regional association structure with spatiotemporal semantics is formed. For example, if two regions have a high frequency of population flow or are spatially adjacent, connecting edges are established in the graph to reflect potential propagation paths.

[0038] Edges between regions are defined based on factors such as population flow intensity, geographical proximity, and transportation networks. The weight of an edge can be determined by various factors, such as the frequency of population movement, physical distance, and transportation accessibility, to quantify the transmission potential and connection strength between regions. For example, if the population flow intensity between two regions is high, the edge weight between these two regions will also be large, indicating a strong transmission potential.

[0039] The constructed regional association graph can not only retain the multidimensional feature information of each region, but also explicitly model cross-regional propagation relationships, providing high-quality structured input data for subsequent graph convolutional neural network training.

[0040] S22, Introducing the graph convolutional neural network structure: Using the constructed regional association graph as input, the message passing mechanism of a graph convolutional neural network (GCN) is employed to spatially aggregate node features. This allows each regional node to simultaneously absorb its own information and that of its neighboring nodes during feature updates, capturing spatial dependencies and potential propagation paths between regions. Multi-layered GCNs can progressively fuse spatial information over a wider range, ensuring that node features encompass both internal and neighboring regional characteristics, thereby characterizing the potential propagation paths and spatial dependencies arising from cross-regional population flows.

[0041] The message passing mechanism of the graph convolutional neural network aggregates the features of neighboring nodes, allowing nodes to propagate local information throughout the graph. It learns and updates node features through multiple layers of convolutional operations. This not only captures local dependencies between nodes but also captures global structural information through multi-level aggregation, providing effective spatial modeling capabilities for spatiotemporal prediction tasks.

[0042] S23, Spatiotemporal Feature Comprehensive Modeling: The node feature sequences, aggregated in the spatial domain, are input into an LSTM for time series modeling. The LSTM, through a gating mechanism, captures the long-term dependence and short-term fluctuations of tuberculosis incidence over time, further revealing the dynamic evolution process over time, achieving spatiotemporal joint modeling of disease transmission, and forming the GCN-LSTM spatiotemporal prediction model. Simultaneously, a multilayer perceptron (MLP) is employed to enhance the nonlinear fitting capability of the spatiotemporal joint modeling, providing high-precision and reliable input features for predicting tuberculosis incidence at future time steps, thus improving prediction performance.

[0043] This step, by constructing a graph-based spatiotemporal correlation network, effectively integrates and represents the spatiotemporal features of multiple regions, enabling accurate modeling of transmission paths and inter-regional interactions. The introduction of a graph convolutional neural network ensures that the spatial relationships between nodes are fully expressed, capturing the transmission potential and dependencies of different regions. Simultaneously, the node feature sequences aggregated by the GCN are input into a Long Short-Term Memory (LSTM) network, which, through a gating mechanism, effectively captures the long-term dependence and short-term fluctuations of tuberculosis incidence over time, achieving joint modeling of spatial features and temporal evolution. Finally, a multilayer perceptron (MLP) is used to further enhance the model's nonlinear fitting capability, allowing the prediction results to more accurately reflect complex spatiotemporal dynamics. Thus, the model not only reflects the epidemic development trend within a region but also identifies cross-regional transmission paths, providing high-quality spatiotemporal structural input for predicting the dynamics of tuberculosis transmission, ensuring the model's prediction accuracy and timeliness.

[0044] Step S3 specifically includes the following steps: Step S31, train the GCN-LSTM spatiotemporal prediction model: In the training phase of the GCN-LSTM spatiotemporal prediction model, historical multi-source feature data, including population mobility intensity, meteorological environmental parameters, socioeconomic indicators, and historical epidemic data, are first used as input features, while the tuberculosis incidence rate for the corresponding period is used as label data. The input data undergoes preprocessing to ensure it is suitable for training the neural network model. Subsequently, the network weights of the model are optimized using the backpropagation algorithm. During training, the GCN-LSTM model progressively updates the weights of the graph convolutional neural network, LSTM, and multilayer perceptron based on the error between the predicted results and the actual incidence rate, continuously reducing the prediction error and thus improving the model's ability to fit the spatiotemporal transmission characteristics of tuberculosis.

[0045] Step S32, Hyperparameter optimization: To further improve the model's predictive performance and generalization ability, a Bayesian optimization algorithm is used to automatically tune the model's key hyperparameters. These hyperparameters include the learning rate, the number of layers in the graph convolutional neural network, the number of hidden units in the LSTM, the regularization coefficient, and the time window length. Bayesian optimization efficiently explores the optimal hyperparameter configuration by sampling the parameter space and evaluating the model's performance under different combinations, thus avoiding the inefficiency and subjectivity of manual tuning. Through this step, the model can obtain the optimal parameter combination, providing a solid foundation for subsequent predictions.

[0046] Step S33, evaluate and save the model: After hyperparameter tuning, a cross-validation strategy was employed to train and evaluate the model multiple times, determining the optimal configuration, validating the model's robustness, and preventing overfitting. By dividing the training data into multiple subsets, one subset was used as the validation set in rotation, while the remaining subsets were used for training, ensuring the model's stability and accuracy across different data subsets. Finally, the best-performing hyperparameter combination was selected based on the cross-validation results, and the optimal model parameters were saved. After training, the model can quickly predict tuberculosis incidence rates in future periods and supports real-time updates, dynamically adjusting predictions based on the latest data to ensure timeliness and accuracy, providing public health departments with efficient spatiotemporal monitoring and risk warning capabilities.

[0047] Example 5: like Figure 1 As shown in the figure, this embodiment provides a method for predicting tuberculosis incidence at multiple scales, which includes the following steps: Step S1: Multi-source dynamic data acquisition steps; in this step: Collect and integrate population flow, meteorological, socio-economic, and historical epidemic data of the target area and its surrounding areas to form a multi-source heterogeneous spatiotemporal dataset; Step S2: The step of constructing a spatiotemporal association network based on a regional association graph structure; in this step: Using regions as nodes and flow intensity as edges, a regional correlation graph is established, and spatial dependence and temporal evolution modeling are achieved through GCN and LSTM. Step S3: The step of training the model based on Bayesian optimization; in this step: The GCN-LSTM model is trained using historical feature data, and Bayesian optimization is used to automatically search for the optimal hyperparameter combination to improve prediction performance. Step S4: The step of outputting the multi-scale spatiotemporal prediction results; in this step: Input the latest multi-source data and output the predicted incidence rate of tuberculosis at different time scales in the future, so as to realize dynamic monitoring and trend early warning.

[0048] In step S1: Multi-source heterogeneous data are collected at the target research scale and integrated to generate a multi-source heterogeneous spatiotemporal dataset, thereby comprehensively reflecting the spatiotemporal driving factors of tuberculosis transmission. Further, the feature set of the multi-source spatiotemporal data includes, but is not limited to: Inter-regional population mobility intensity data: By collecting inter-regional population migration data in real time, we can depict the scale and frequency of population migration between different regions and assess the potential transmission channels and intensity of cross-regional infection risks; Meteorological environmental parameter data: including time-series meteorological data such as daily precipitation, average temperature, and humidity, used to analyze the impact of natural environmental changes on tuberculosis transmission and to combine climate change analysis on the impact on tuberculosis transmission pathways; Socioeconomic indicators include resident population size, population density, regional GDP, and allocation of medical resources, to assess the impact of socioeconomic development levels on disease prevalence characteristics; Historical epidemic data: This includes the trend of tuberculosis incidence in the study area over a previous period, used as a benchmark for model training and prediction, thereby ensuring that the model can make effective predictions based on historical data.

[0049] After collecting the above data, it was organized into a multi-source spatiotemporal data feature set, which served as input for subsequent modeling and analysis. The collection and integration of multi-source heterogeneous data comprehensively reflects the spatiotemporal driving factors of tuberculosis transmission, providing complete basic data support for subsequent modeling. The collected data can reveal the potential pathways and influencing factors of tuberculosis transmission from different dimensions, ensuring the accuracy of subsequent models across multiple spatial and temporal scales. Furthermore, it can capture the combined effects of different factors on tuberculosis transmission, ensuring the comprehensiveness and accuracy of tuberculosis incidence prediction.

[0050] Step S2 specifically includes the following steps: S21, Construct a region association graph: Based on the geographical location and population flow data of the target study area, a spatiotemporal relationship map between regions is constructed: Each study area is treated as a node in the graph. The node attributes include multi-source feature data of the area, such as population density, meteorological and environmental parameters, socioeconomic indicators and historical epidemic data, which are used to characterize the features of the area.

[0051] Using the intensity of population flow and geographical proximity between regions as the basis for connecting nodes, a regional association structure with spatiotemporal semantics is formed. For example, if two regions have a high frequency of population flow or are spatially adjacent, connecting edges are established in the graph to reflect potential propagation paths.

[0052] Edges between regions are defined based on factors such as population flow intensity, geographical proximity, and transportation networks. The weight of an edge can be determined by various factors, such as the frequency of population movement, physical distance, and transportation accessibility, to quantify the transmission potential and connection strength between regions. For example, if the population flow intensity between two regions is high, the edge weight between these two regions will also be large, indicating a strong transmission potential.

[0053] The constructed regional association graph can not only retain the multidimensional feature information of each region, but also explicitly model cross-regional propagation relationships, providing high-quality structured input data for subsequent graph convolutional neural network training.

[0054] S22, Introducing the graph convolutional neural network structure: Using the constructed regional association graph as input, the message passing mechanism of a graph convolutional neural network (GCN) is employed to spatially aggregate node features. This allows each regional node to simultaneously absorb its own information and that of its neighboring nodes during feature updates, capturing spatial dependencies and potential propagation paths between regions. Multi-layered GCNs can progressively fuse spatial information over a wider range, ensuring that node features encompass both internal and neighboring regional characteristics, thereby characterizing the potential propagation paths and spatial dependencies arising from cross-regional population flows.

[0055] The message passing mechanism of the graph convolutional neural network aggregates the features of neighboring nodes, allowing nodes to propagate local information throughout the graph. It learns and updates node features through multiple layers of convolutional operations. This not only captures local dependencies between nodes but also captures global structural information through multi-level aggregation, providing effective spatial modeling capabilities for spatiotemporal prediction tasks.

[0056] S23, Spatiotemporal Feature Comprehensive Modeling: The node feature sequences, aggregated in the spatial domain, are input into an LSTM for time series modeling. The LSTM, through a gating mechanism, captures the long-term dependence and short-term fluctuations of tuberculosis incidence over time, further revealing the dynamic evolution process over time, achieving spatiotemporal joint modeling of disease transmission, and forming the GCN-LSTM spatiotemporal prediction model. Simultaneously, a multilayer perceptron (MLP) is employed to enhance the nonlinear fitting capability of the spatiotemporal joint modeling, providing high-precision and reliable input features for predicting tuberculosis incidence at future time steps, thus improving prediction performance.

[0057] This step, by constructing a graph-based spatiotemporal correlation network, effectively integrates and represents the spatiotemporal features of multiple regions, enabling accurate modeling of transmission paths and inter-regional interactions. The introduction of a graph convolutional neural network ensures that the spatial relationships between nodes are fully expressed, capturing the transmission potential and dependencies of different regions. Simultaneously, the node feature sequences aggregated by the GCN are input into a Long Short-Term Memory (LSTM) network, which, through a gating mechanism, effectively captures the long-term dependence and short-term fluctuations of tuberculosis incidence over time, achieving joint modeling of spatial features and temporal evolution. Finally, a multilayer perceptron (MLP) is used to further enhance the model's nonlinear fitting capability, allowing the prediction results to more accurately reflect complex spatiotemporal dynamics. Thus, the model not only reflects the epidemic development trend within a region but also identifies cross-regional transmission paths, providing high-quality spatiotemporal structural input for predicting the dynamics of tuberculosis transmission, ensuring the model's prediction accuracy and timeliness.

[0058] Step S3 specifically includes the following steps: Step S31, train the GCN-LSTM spatiotemporal prediction model: In the training phase of the GCN-LSTM spatiotemporal prediction model, historical multi-source feature data, including population mobility intensity, meteorological environmental parameters, socioeconomic indicators, and historical epidemic data, are first used as input features, while the tuberculosis incidence rate for the corresponding period is used as label data. The input data undergoes preprocessing to ensure it is suitable for training the neural network model. Subsequently, the network weights of the model are optimized using the backpropagation algorithm. During training, the GCN-LSTM model progressively updates the weights of the graph convolutional neural network, LSTM, and multilayer perceptron based on the error between the predicted results and the actual incidence rate, continuously reducing the prediction error and thus improving the model's ability to fit the spatiotemporal transmission characteristics of tuberculosis.

[0059] Step S32, Hyperparameter optimization: To further improve the model's predictive performance and generalization ability, a Bayesian optimization algorithm is used to automatically tune the model's key hyperparameters. These hyperparameters include the learning rate, the number of layers in the graph convolutional neural network, the number of hidden units in the LSTM, the regularization coefficient, and the time window length. Bayesian optimization efficiently explores the optimal hyperparameter configuration by sampling the parameter space and evaluating the model's performance under different combinations, thus avoiding the inefficiency and subjectivity of manual tuning. Through this step, the model can obtain the optimal parameter combination, providing a solid foundation for subsequent predictions.

[0060] Step S33, evaluate and save the model: After hyperparameter tuning, a cross-validation strategy was employed to train and evaluate the model multiple times, determining the optimal configuration, validating the model's robustness, and preventing overfitting. By dividing the training data into multiple subsets, one subset was used as the validation set in rotation, while the remaining subsets were used for training, ensuring the model's stability and accuracy across different data subsets. Finally, the best-performing hyperparameter combination was selected based on the cross-validation results, and the optimal model parameters were saved. After training, the model can quickly predict tuberculosis incidence rates in future periods and supports real-time updates, dynamically adjusting predictions based on the latest data to ensure timeliness and accuracy, providing public health departments with efficient spatiotemporal monitoring and risk warning capabilities.

[0061] Step S4 specifically includes the following steps: Step S41, input the latest multi-source feature data: After completing model training and saving the optimal parameters, the latest collected multi-source feature data is input into the trained GCN-LSTM spatiotemporal prediction model. This multi-source feature data includes real-time population flow intensity, meteorological environmental parameters, socioeconomic indicators, and recent historical epidemic data. Before input, the data is processed and standardized according to a preset time window and spatial scale to ensure that the input data format is consistent with the data used during model training and can fully reflect the spatiotemporal characteristics and dynamic trends between regions.

[0062] Step S42, Generate model predictions: After the processed multi-source feature data is input into the GCN-LSTM model, GCN first aggregates the spatial features of each regional node, fusing the feature information of neighboring regions with its own features to reflect the spatial dependencies between regions. Subsequently, LSTM models the time series of node features, capturing the dynamic evolution of tuberculosis incidence over time. Finally, the MLP module enhances the model's nonlinear fitting capability, enabling prediction of tuberculosis incidence in future periods. The model can output predicted values ​​for each study region over multiple future time steps and can generate prediction results at different time scales (such as daily, weekly, and monthly) to adapt to different management and control scenarios.

[0063] Step S43, output the prediction result: After the model completes its predictions, the results are compiled and output to provide public health departments with spatiotemporally dynamic tuberculosis incidence data. The predictions can be used for scenarios such as epidemic trend analysis, risk warning, resource allocation optimization, and decision support. Because the model supports real-time updates, it can continuously refine the predictions by incorporating the latest multi-source data, ensuring high timeliness and accuracy, and providing scientific basis and technical support for public health prevention and control.

[0064] Example 6: like Figure 2 and Figure 3 As shown in the figure, this embodiment provides a multi-scale tuberculosis incidence prediction system, including: The data acquisition and preprocessing module 1 collects multi-source heterogeneous data from the target area and its surrounding areas, including population mobility intensity, meteorological and environmental parameters, socioeconomic indicators, and historical tuberculosis epidemic data. It then cleans, standardizes, and formats the collected data to form a high-quality multi-source spatiotemporal feature set. By integrating data from different sources and of different types, this module provides reliable and accurate input data for subsequent modeling, improving the effectiveness of model training and the accuracy of prediction results.

[0065] The second module of the spatiotemporal correlation network construction module builds a regional correlation graph based on geospatial relationships and population flow data. Nodes in the regional correlation graph represent the study area, and their attributes include multi-source feature data; edges represent connections between regions, and their weights are determined by population flow intensity or geographical proximity. This module introduces a graph convolutional neural network to achieve spatial aggregation of node information and combines it with LSTM for time series modeling, enabling the model to capture the spatial dependencies and dynamic evolution patterns between regions, thereby improving prediction accuracy and timeliness.

[0066] Model training module 3 trains a GCN-LSTM prediction model based on historical multi-source feature data and optimizes network weights using the backpropagation algorithm. This module combines Bayesian optimization to automatically tune key hyperparameters and evaluates the model's performance on different data subsets through cross-validation, ultimately saving the optimal model parameters. This module improves the model's predictive performance and generalization ability, ensuring its stability and accuracy under different time and regional data conditions, providing a reliable foundation for real-time prediction of future tuberculosis incidence rates.

[0067] The spatiotemporal prediction output module 4 inputs the latest collected multi-source feature data into the trained GCN-LSTM model to generate predicted tuberculosis incidence rates for each region at different time steps in the future, and provides prediction results at different time scales according to actual needs. This module dynamically updates the input data and outputs prediction results, enabling public health management departments to obtain real-time spatiotemporal dynamic information on tuberculosis, achieving risk warning, trend analysis, and optimal resource allocation, providing a scientific basis and technical support for the formulation of prevention and control strategies.

[0068] Furthermore, such as Figure 3As shown, the multi-scale tuberculosis incidence prediction system proposed in this invention generates predictions using multi-source heterogeneous data including socioeconomic factors, meteorological factors, and historical epidemic data. It can be further described as follows: First, the study area is spatially divided, and each regional unit is considered a node in a graph. Through node extraction and relation extraction, multi-dimensional attribute information of the region and the intensity of population flow between regions are obtained. This allows for the construction of a regional association graph structure reflecting spatial adjacency relationships, thereby characterizing the potential transmission paths and interaction intensity between different regions.

[0069] After establishing the regional association graph structure, a graph convolutional neural network is used to extract features and aggregate information from the attributes of each node. Through a message passing mechanism, information from the target region and its neighboring regions is fused to capture spatial correlations and propagation dependencies, forming a temporal node vector representation containing spatial features. Subsequently, the spatially aggregated node feature sequence is input into a long short-term memory network to learn the dynamic changes in disease incidence rates over time. The long short-term memory network effectively captures the long-term dependencies and trend characteristics of incidence rates, thereby enabling time series modeling and prediction. After feature extraction and temporal modeling, a multilayer perceptron is used to perform nonlinear mapping and regression analysis on the fused spatiotemporal features to improve the model's fitting ability and prediction accuracy for complex spatiotemporal relationships.

[0070] Ultimately, the model outputs predictions of future tuberculosis incidence rates for each study region, achieving multi-regional, multi-scale disease risk prediction. Through this process, accurate predictions of future tuberculosis incidence rates in various regions can be made, providing scientific basis and technical support for public health prevention and control strategy formulation, resource optimization, and risk early warning.

[0071] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the systems disclosed in the embodiments; relevant details can be found in the method section.

[0072] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0073] In the embodiments provided by this invention, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0074] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0075] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit.

[0076] Similarly, in the various embodiments of the present invention, each processing unit can be integrated into a functional module, or each processing unit can exist physically, or two or more processing units can be integrated into a functional module.

[0077] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0078] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0079] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A method of multiscale prediction of tuberculosis incidence, characterized by, Comprise the following steps: Step S1: the step of multi-source dynamic data acquisition; in this step: Collect the population flow, meteorology, social economy and historical epidemic data of the target area and the surrounding area and integrate them to form a multi-source heterogeneous spatio-temporal data set; Step S2: the step of constructing a spatio-temporal correlation network based on a regional correlation graph structure; in this step: Take the region as the node and the flow intensity as the edge to establish a regional correlation graph, and realize the modeling of spatial dependence and time evolution through GCN and LSTM; Step S3: the step of model training based on Bayesian optimization; in this step: Train the GCN-LSTM model using historical feature data, automatically search for the optimal hyperparameter combination using Bayesian optimization, and improve the prediction performance; Step S4: the step of multi-scale spatio-temporal prediction output result; in this step: Input the latest multi-source data and output the prediction results of tuberculosis incidence at different time scales in the future to realize dynamic monitoring and trend early warning.

2. A method of multiscale prediction of tuberculosis incidence as claimed in claim 1, wherein, In the step S1: the multi-source spatio-temporal data feature set comprises: Inter-regional population flow intensity data: by collecting real-time inter-regional population migration data, the size and frequency of personnel migration between different regions are described, and the potential transmission channels and intensity of cross-regional transmission risk are evaluated; Meteorological environmental parameter data: time series meteorological data including daily precipitation, average temperature and humidity, which are used to analyze the influence of natural environmental changes on the spread of tuberculosis and the influence of climate change on the transmission path of tuberculosis; Social and economic index data: including permanent population size, population density, gross regional product, medical resource allocation, evaluating the influence of social and economic development level on the characteristics of disease prevalence; Historical epidemic data: covering the tuberculosis incidence trend of the study area in the past time period, which is used for model training and prediction benchmark reference, so as to ensure that the model can effectively predict based on historical data.

3. A method of multiscale prediction of tuberculosis incidence according to claim 1 or 2, characterized in that, The step S2 specifically comprises the following steps: S21, constructing a regional correlation graph: According to the geographical position and population flow data of the target research region, a spatio-temporal correlation graph between regions is constructed: Each research region is taken as a node in the graph, and the node attributes include the multi-source feature data of the region, which is used to represent the characteristics of the region; The population flow intensity and geographical proximity between regions are taken as the connection basis between nodes to form a regional correlation structure with spatio-temporal semantics; The edges between regions are defined by population flow intensity, geographical proximity and traffic network factors; the weight of the edge can be determined by different factors to quantify the transmission potential and connection strength between regions; S22, introduce graph convolutional neural network structure: Take the constructed regional correlation graph as input, and aggregate the node features in the spatial domain through the message passing mechanism of graph convolutional neural network, so that each regional node absorbs its own information and neighborhood node information in the feature updating process, and captures the spatial dependence relationship and potential transmission path between regions; multi-layer GCN can fuse larger range of spatial information layer by layer, so that the node features contain both regional internal features and neighborhood regional features, thereby describing the potential transmission path and spatial dependence brought by cross-regional population flow; S23, comprehensive modeling of spatio-temporal features: The node feature sequence aggregated in the spatial domain is input into the LSTM for time sequence modeling; the LSTM captures the long-term dependence and short-term fluctuation characteristics of the incidence of tuberculosis in the time dimension through a gating mechanism, further exhibits the dynamic evolution process in the time scale, realizes the spatio-temporal joint modeling of disease transmission, and forms a GCN-LSTM spatio-temporal prediction model; meanwhile, a multilayer perceptron is used to enhance the nonlinear fitting capability of the spatio-temporal joint modeling.

4. A method of multiscale prediction of tuberculosis incidence according to claim 3, characterized in that, The message passing mechanism of the graph convolutional neural network is to aggregate the features of neighbor nodes, propagate local information to the whole graph, and learn and update the features of nodes through multi-layer convolution operations.

5. A method of multiscale prediction of tuberculosis incidence according to claim 4, characterized in that, The step S3 specifically comprises the following steps: Step S31, training a GCN-LSTM spatio-temporal prediction model: In the training phase of the GCN-LSTM spatio-temporal prediction model, historical multi-source feature data is used as input features, and the incidence of tuberculosis in the corresponding period is used as label data; the input data is preprocessed to ensure that the data can adapt to the training of the neural network model; then, the network weights of the model are optimized through a back propagation algorithm; Step S32, hyperparameter optimization: A Bayesian optimization algorithm is used to automatically tune the key hyperparameters of the model; Step S33, evaluating and saving the model: After completing the hyperparameter tuning, the model is trained and evaluated for multiple rounds using a cross-validation strategy to determine the optimal configuration; the best hyperparameter combination is selected according to the cross-validation results, and the optimal model parameters are saved.

6. A method of multiscale prediction of tuberculosis incidence according to claim 5, wherein, During the training process, the GCN-LSTM model gradually updates the weights of the graph convolutional neural network, LSTM, and multilayer perceptron based on the error between the prediction results and the true incidence.

7. The method of claim 5, wherein the method is a multiscale predictive method for tuberculosis incidence. The hyperparameters include learning rate, number of graph convolutional neural network layers, number of LSTM hidden units, regularization coefficient, and time window length; the Bayesian optimization efficiently explores the optimal hyperparameter configuration by sampling the parameter space and evaluating the model performance under different combinations.

8. A method of multiscale prediction of tuberculosis incidence according to claim 7, characterized in that, The step S4 specifically comprises: Step S41, inputting the latest multi-source feature data: After completing the model training and saving the optimal parameters, the latest collected multi-source feature data is input into the trained GCN-LSTM spatio-temporal prediction model; the data is organized and standardized according to the preset time window and spatial scale before input; Step S42, generating model prediction: After inputting the organized multi-source feature data into the GCN-LSTM model, the GCN first aggregates the spatial features of each regional node, fusing the feature information of the neighboring regions with its own features; then, the LSTM models the time sequence of the node features to capture the dynamic evolution law of the incidence of tuberculosis in the time dimension; finally, the MLP enhances the nonlinear fitting capability of the model to realize the prediction of the incidence of tuberculosis in the future period; Step S43, outputting the prediction result: After the model prediction is completed, the prediction result is organized and output.

9. A system for multiscale prediction of tuberculosis incidence, characterized by, It comprises: a data acquisition and processing module (1), a spatio-temporal correlation network construction module (2), a GCN-LSTM model training module (3), and a spatio-temporal prediction result output module (4). The data acquisition and processing module (1) collects population flow data, meteorological environment data, social economic indicators and historical epidemic data of the target area and the surrounding area, and performs cleaning, formatting and standardization processing on the data to generate a spatio-temporal feature data set; The spatio-temporal correlation network construction module (2) uses multi-source data to construct a spatio-temporal correlation graph structure between regions, with each region as a node and population flow intensity and other features as edge weights, to accurately model the spatial dependence relationship and dynamic evolution law between regions; The GCN-LSTM model training module (3) trains multi-source spatio-temporal data based on graph convolutional neural network and long short-term memory network, captures the spatial dependence between regions and the dynamic evolution trend of time series, and adjusts key hyperparameters through Bayesian optimization algorithm to improve the prediction accuracy and generalization ability of the model; The spatio-temporal prediction result output module (4) uses the trained GCN-LSTM model and the latest multi-source data to predict the tuberculosis incidence of each region at different time scales in the future, and provides decision support, disease risk warning and resource optimization services.

10. The system for multiscale prediction of tuberculosis incidence as claimed in claim 9, wherein, Using multi-source heterogeneous data including social economy, meteorological factors and historical epidemic data for prediction, the research area is spatially divided, and each regional unit is regarded as a node in the graph; through node extraction and relationship extraction, the multi-dimensional attribute information of the region and the population flow intensity between regions are obtained, and a regional correlation graph structure reflecting the spatial adjacency relationship is constructed; After the establishment of the regional correlation graph structure, the graph convolutional neural network is used to extract features and aggregate information of each node; through the message passing mechanism, the information of the target region and its adjacent regions is fused to capture the correlation and propagation dependence relationship in the spatial dimension, forming a time series node vector representation containing spatial features; the node feature sequence aggregated in space is input into the long short-term memory network to learn the dynamic change law of the disease incidence in the time dimension; after feature extraction and time series modeling, the multi-layer perceptron is used to perform nonlinear mapping and regression analysis on the fused spatio-temporal features; the model outputs the tuberculosis incidence prediction of each research region in the future period.

Images