Disease data analysis method, device, electronic device and storage medium

Abstract

The present invention provides a disease data analysis method, device, electronic device and storage medium, which relate to the field of data processing technology, including: obtaining multi-source disease data associated with a target area within a preset time period; wherein the multi-source disease data includes at least one of the following: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports and disease news information data in the target area; based on the multi-source disease data, constructing a regional graph network of the target area; after splicing the regional graph network and the multi-source fusion features of the multi-source disease data, inputting them into a disease data analysis model containing multiple expert intelligent agents, and outputting the disease analysis results of the target area.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a disease data analysis method, device, electronic equipment and storage medium. Background Art

[0002] With the continuous advancement of technology, data analysis technology has been widely used in many fields, for example, it can be used for in-depth analysis of disease-related data.

[0003] The spread of a disease typically involves a transmission cycle and a latent period, which can lead to a delay in early warning. However, effective data analysis methods can help identify the risk of disease transmission in advance, buying valuable time for disease prevention and control measures.

[0004] Therefore, how to effectively analyze disease data has become an urgent problem to be solved in the industry. Summary of the Invention

[0005] The present invention provides a disease data analysis method, device, electronic device and storage medium to solve the problem of how to effectively perform disease data analysis in the prior art.

[0006] The present invention provides a disease data analysis method, comprising:

[0007] Acquiring multi-source disease data associated with a target area within a preset time period; wherein the multi-source disease data includes at least one of the following: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data within the target area;

[0008] constructing a regional graph network of the target area based on the multi-source disease data;

[0009] After the regional graph network and the multi-source fusion features of the multi-source disease data are spliced ​​together, the spliced ​​data are input into a disease data analysis model comprising multiple expert agents, and the disease analysis results of the target region are output.

[0010] According to a disease data analysis method provided by the present invention, based on the multi-source disease data, a regional graph network of the target area is constructed, comprising:

[0011] Divide the target area into multiple sub-areas, and determine the edge weights between the sub-areas based on the number of common activity scenes of disease infected persons between the sub-areas and the total number of activity scenes of disease infected persons in the sub-areas in the epidemiological survey report of the multi-source disease data;

[0012] Different sub-regions are connected based on the edge weights between the sub-regions to construct a regional graph network of the target region.

[0013] According to a disease data analysis method provided by the present invention, after the regional graph network and the multi-source fusion features of the multi-source disease data are spliced ​​together, the spliced ​​data are input into a disease data analysis model including multiple expert agents, and the disease analysis results of the target region are output, including:

[0014] After splicing the regional graph network and the multi-source fusion features of the multi-source disease data, the splicing is input into a disease data analysis model including multiple expert agents to obtain disease occurrence probability information, disease severity information and disease description text output by each expert agent;

[0015] Determine a prediction result of the disease occurrence probability of the target area based on the disease occurrence probability information corresponding to each expert agent and the confidence level of the expert agent;

[0016] Determining a disease severity prediction result for the target area based on the disease severity information corresponding to each expert agent and the expert agent confidence level;

[0017] Determining a disease analysis result for the target area based on the disease occurrence probability prediction result and the disease severity prediction result for the target area;

[0018] The confidence of the expert agent is determined based on the quantitative certainty of the expert agent and the historical prediction accuracy.

[0019] According to a disease data analysis method provided by the present invention, the method for obtaining a disease occurrence probability prediction result includes:

[0020] Summing the disease occurrence probability information of each expert agent and the product of the expert agent confidence to obtain a first summation result;

[0021] Determining the disease occurrence probability prediction result according to the first summation result and the summation result of the confidence levels of the respective expert agents;

[0022] The method for obtaining the disease severity prediction result includes:

[0023] Summing the disease severity information of each expert agent and the product of the expert agent confidence to obtain a second summation result;

[0024] The disease severity prediction result is determined based on the sum of the second summation result and the confidence level of each expert agent.

[0025] According to a disease data analysis method provided by the present invention, the method for calculating the confidence of the expert agent includes:

[0026] Determining the historical accuracy of the expert agent based on the difference between the disease occurrence probability information output by the expert agent at different time points and the actual disease occurrence probability information;

[0027] Determining the quantitative certainty of the expert agent based on the number of uncertainty words, the total number of words, and the number of key entity words in the disease description text output by the expert agent;

[0028] The confidence of the expert agent is obtained based on the historical accuracy of the expert agent and the mean of the quantitative certainty.

[0029] According to a disease data analysis method provided by the present invention, after splicing the region graph network and the multi-source fusion features of the multi-source disease data, the splicing is input into a disease data analysis model comprising multiple expert agents, and before the step of outputting the disease analysis result of the target region, the method further includes:

[0030] The spliced ​​regional graph network sample and the multi-source fusion feature sample are used as a training sample to obtain multiple training samples;

[0031] For each training sample, the training sample is used as an environmental state and inputted into each expert agent in the disease data analysis model to obtain experience data of each expert agent, and the experience data is stored in the experience pool corresponding to each expert agent; wherein the experience data includes: agent reward value and agent action selection;

[0032] Calculating the individual advantage of each expert agent based on the agent reward value in the experience pool of each expert agent;

[0033] Calculating a loss value for each expert agent based on the individual advantage and action selection strategy ratio of each expert agent, so as to optimize the policy network parameters of the corresponding expert agent based on the loss value;

[0034] Traverse each of the training samples until the preset training conditions are met, and obtain a disease data analysis model including multiple expert agents.

[0035] According to a disease data analysis method provided by the present invention, obtaining multi-source disease data associated with a target area within a preset time period includes:

[0036] Standardizing the pharmacy sales data, school absenteeism, and hospital visit volume using a sliding window standardization method to obtain standardized pharmacy sales data, school absenteeism, and hospital visit volume data;

[0037] Call the pre-configured large language model to extract key disease entities and relationships from news information data to obtain disease news information data;

[0038] When the virus sequence data of the disease is obtained, the pre-configured large language model is called to analyze the key sites in the virus sequence data to obtain the pathogen data.

[0039] According to a disease data analysis method provided by the present invention, the multi-source fusion features include: numerical features of numerical data, text features of text data and pathogenic data features of pathogenic data;

[0040] The numerical data includes at least one of the following: pharmacy sales data, school absence data, and hospital visit data;

[0041] The text data includes at least one of the following: epidemiological survey reports and disease news information data.

[0042] The present invention provides a disease data analysis device, comprising:

[0043] an acquisition module, configured to acquire multi-source disease data associated with a target area within a preset time period; wherein the multi-source disease data includes at least one of the following: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data within the target area;

[0044] a processing module, configured to construct a regional map network of the target area based on the positioning information of the multi-source disease data;

[0045] An analysis module is used to splice the regional graph network and the multi-source fusion features of the multi-source disease data, input them into a disease data analysis model containing multiple expert agents, and output the disease analysis results of the target area.

[0046] The present invention also provides a disease data analysis device, which is further used for:

[0047] Divide the target area into multiple sub-areas, and determine the edge weights between the sub-areas based on the number of common activity scenes of disease infected persons between the sub-areas and the total number of activity scenes of disease infected persons in the sub-areas in the epidemiological survey report of the multi-source disease data;

[0048] Different sub-regions are connected based on the edge weights between the sub-regions to construct a regional graph network of the target region.

[0049] The present invention also provides a disease data analysis device, which is further used for:

[0050] After splicing the regional graph network and the multi-source fusion features of the multi-source disease data, the splicing is input into a disease data analysis model including multiple expert agents to obtain disease occurrence probability information, disease severity information and disease description text output by each expert agent;

[0051] Determine a prediction result of the disease occurrence probability of the target area based on the disease occurrence probability information corresponding to each expert agent and the confidence level of the expert agent;

[0052] Determining a disease severity prediction result for the target area based on the disease severity information corresponding to each expert agent and the expert agent confidence level;

[0053] Determining a disease analysis result for the target area based on the disease occurrence probability prediction result and the disease severity prediction result for the target area;

[0054] The confidence of the expert agent is determined based on the quantitative certainty of the expert agent and the historical prediction accuracy.

[0055] The present invention also provides a disease data analysis device, which is further used for:

[0056] Summing the disease occurrence probability information of each expert agent and the product of the expert agent confidence to obtain a first summation result;

[0057] Determining the disease occurrence probability prediction result according to the first summation result and the summation result of the confidence levels of the respective expert agents;

[0058] The method for obtaining the disease severity prediction result includes:

[0059] Summing the disease severity information of each expert agent and the product of the expert agent confidence to obtain a second summation result;

[0060] The disease severity prediction result is determined based on the sum of the second summation result and the confidence level of each expert agent.

[0061] The present invention also provides a disease data analysis device, which is further used for:

[0062] Determining the historical accuracy of the expert agent based on the difference between the disease occurrence probability information output by the expert agent at different time points and the actual disease occurrence probability information;

[0063] Determining the quantitative certainty of the expert agent based on the number of uncertainty words, the total number of words, and the number of key entity words in the disease description text output by the expert agent;

[0064] The confidence of the expert agent is obtained based on the historical accuracy of the expert agent and the mean of the quantitative certainty.

[0065] The present invention also provides a disease data analysis device, which is further used for:

[0066] The spliced ​​regional graph network sample and the multi-source fusion feature sample are used as a training sample to obtain multiple training samples;

[0067] For each training sample, the training sample is used as an environmental state and inputted into each expert agent in the disease data analysis model to obtain experience data of each expert agent, and the experience data is stored in the experience pool corresponding to each expert agent; wherein the experience data includes: agent reward value and agent action selection;

[0068] Calculating the individual advantage of each expert agent based on the agent reward value in the experience pool of each expert agent;

[0069] Calculating a loss value for each expert agent based on the individual advantage and action selection strategy ratio of each expert agent, so as to optimize the policy network parameters of the corresponding expert agent based on the loss value;

[0070] Traverse each of the training samples until the preset training conditions are met, and obtain a disease data analysis model including multiple expert agents.

[0071] The present invention also provides a disease data analysis device, which is further used for:

[0072] Standardizing the pharmacy sales data, school absenteeism, and hospital visit volume using a sliding window standardization method to obtain standardized pharmacy sales data, school absenteeism, and hospital visit volume data;

[0073] Call the pre-configured large language model to extract key disease entities and relationships from news information data to obtain disease news information data;

[0074] When the virus sequence data of the disease is obtained, the pre-configured large language model is called to analyze the key sites in the virus sequence data to obtain the pathogen data.

[0075] According to a disease data analysis device provided by the present invention, the multi-source fusion features include: numerical features of numerical data, text features of text data and pathogenic data features of pathogenic data;

[0076] The numerical data includes at least one of the following: pharmacy sales data, school absence data, and hospital visit data;

[0077] The text data includes at least one of the following: epidemiological survey reports and disease news information data.

[0078] The present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the disease data analysis method described above is implemented.

[0079] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described disease data analysis methods.

[0080] The present invention also provides a computer program product, comprising a computer program, which implements any of the above-mentioned disease data analysis methods when executed by a processor.

[0081] The disease data analysis method, device, electronic device, and storage medium provided by the present invention can comprehensively reflect the disease-related dynamics in the target area by collecting multi-source disease data such as pharmacy sales data, school absenteeism, and hospital visit volume. The cross-modal attention mechanism can effectively fuse numerical data and text data. By learning the correlation between data of different modalities, the model can more accurately capture the key characteristics of disease transmission and improve the accuracy of data analysis. The constructed regional map network uses geographical regions as nodes and can intuitively display the spread of diseases in different regions. The disease data analysis model containing multiple expert agents can analyze disease data from different angles and ultimately generate accurate and effective disease data analysis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0083] Figure 1 1 is a flow chart of the disease data analysis method provided by the present invention;

[0084] Figure 2 A schematic diagram of the structure of the disease data analysis device provided by the present invention;

[0085] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0086] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0087] Figure 1 It is a flow chart of the disease data analysis method provided by the present invention, such as Figure 1 As shown, the method includes the following:

[0088] Step 110: Acquire multi-source disease data associated with the target area within a preset time period; wherein the multi-source disease data includes at least one of the following: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data within the target area;

[0089] In the present invention, the preset time period refers to a time range determined according to specific analysis requirements and is used to collect disease data related to the target area.

[0090] The pre-set time period is determined based on the time of the early warning signal if an epidemiological investigation has not yet been conducted. If an epidemiological investigation has already been conducted, the pre-set time period is based on the "time of the first case of disease infection" in the epidemiological investigation report. The pre-set time period can specifically refer to a window period from the pre-set date to ensure consistency in the data timeline.

[0091] In the present invention, the target area may refer to a specific geographical scope for disease analysis, such as a city, province, country, etc.

[0092] In the present invention, multi-source disease data includes various types of data such as pharmacy sales data, school absences, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data.

[0093] More specifically, pharmacy sales data refers to the drug sales records of pharmacies in a target area during a specific time period. This data may include sales of disease-related drugs such as antipyretics and cough suppressants.

[0094] During a disease outbreak, sales of related medicines typically increase significantly. Monitoring pharmacy sales data can help detect early signs of disease spread.

[0095] School absences record the number of students absent and the reasons for absences, which is derived from school attendance records. The spread of disease among students will lead to an increase in the number of absences, which helps to assist in determining the spread of the disease.

[0096] Hospital visit volume data counts the number of outpatients and inpatients in various departments of the hospital, directly reflecting the disease medical burden in the region. An abnormal increase in the number of visits may indicate a disease outbreak.

[0097] Epidemiological investigation reports are written by professionals and cover disease transmission routes, case exposure history, close contact information, etc., providing key basis for disease research. Epidemiological investigation reports can be obtained from public website information.

[0098] Disease news information data includes news reports and the level of discussion about diseases on social media. It reflects public awareness of diseases and helps monitor their spread and social impact. This data, combined, helps improve the accuracy and reliability of disease analysis.

[0099] Step 120: constructing a regional graph network of the target area based on the multi-source disease data;

[0100] In the present invention, a regional graph network of a target area is constructed based on multi-source disease data, which is a graph structure that uses geographical areas as spatial units and the association relationship between regions as a connecting link.

[0101] For example, a high number of shared activity sites between two regions indicates greater mobility and contact between people, and a higher risk of disease transmission, resulting in a larger edge weight. Regional graph networks can intuitively demonstrate the spread of disease across different regions, providing strong support for identifying disease transmission pathways and locating high-risk areas.

[0102] Step 130 , after splicing the regional graph network and the multi-source fusion features of the multi-source disease data, input them into a disease data analysis model comprising multiple expert agents, and output the disease analysis results of the target region.

[0103] In this paper, the structural information of the regional graph network and multi-source fusion features are spliced ​​together to form a unified feature vector. The splicing operation can be a simple vector concatenation, which sequentially connects the node features of the regional graph network and the multi-source fusion features into a long vector. The fused feature vector contains both the structural information of the regional graph network and the key information of the multi-source fusion features, and can comprehensively describe the disease spread in the target area.

[0104] The disease data analysis model contains models of multiple expert agents, each of which focuses on different disease analysis tasks.

[0105] For example, some agents excel at predicting disease spread trends, while others focus on assessing medical resource needs. Each agent has its own strategy network and value network, enabling independent analysis and judgment based on input feature vectors.

[0106] In an optional embodiment, the disease data analysis model includes: clinical medicine experts, disease prevention and control experts, virology experts, data scientists, medical resource planning experts, etc., which can be specifically configured and selected through a pre-defined expert agent selection interface.

[0107] The disease data analysis model conducts a comprehensive analysis based on the output results of each expert agent and outputs the final disease analysis results.

[0108] In the present invention, the final disease analysis result may include a prediction result of the likelihood of disease occurrence in the target area and a prediction result of the severity of the disease.

[0109] More specifically, a disease occurrence probability prediction can be the predicted probability of a disease outbreak or spread within a target area. This is expressed as a numerical value between 0 and 1, with higher values ​​indicating a greater likelihood of disease occurrence. This helps decision-makers understand the risk level of a disease outbreak and formulate preventive and response measures in advance.

[0110] The disease severity prediction results assess the potential severity of a disease outbreak in the target area. It is expressed as a value between 0 and 1, with higher values ​​indicating more severe disease.

[0111] This allows decision-makers to understand the potential harm of diseases, allocate medical resources rationally, and develop appropriate treatment and isolation plans. Through these specific predictions, decision-makers can conduct disease prevention and control and allocate resources more scientifically, improving their ability to respond to disease outbreaks.

[0112] In this invention, by collecting multi-source disease data such as pharmacy sales data, school absenteeism, and hospital visit volume, the disease-related dynamics in the target area can be fully reflected. The cross-modal attention mechanism can effectively fuse numerical data and text data. By learning the correlation between different modal data, the model can more accurately capture the key characteristics of disease transmission and improve the accuracy of data analysis. The constructed regional graph network uses geographical regions as nodes and can intuitively display the spread of diseases in different regions. The disease data analysis model containing multiple expert agents can analyze disease data from different angles and ultimately generate accurate and effective disease data analysis results.

[0113] Optionally, the multi-source fusion features include: numerical features of numerical data, text features of text data, and pathogenic data features of pathogenic data;

[0114] The numerical data includes at least one of the following: pharmacy sales data, school absence data, and hospital visit data;

[0115] The text data includes at least one of the following: epidemiological survey reports and disease news information data.

[0116] In the present invention, numerical data refers to information that is expressed in numerical form, is quantifiable, and can be mathematically calculated. It is commonly found in multi-source disease data such as pharmacy sales records, school absenteeism statistics, and hospital visit statistics.

[0117] For example, the sales volume of antipyretics and cough suppressants in a pharmacy on a certain day, the number of students absent from a class due to illness, the number of patients seen by a department of a hospital on that day, etc., provide quantitative indicators for disease analysis.

[0118] Text data is descriptive information recorded in text form, which is commonly found in epidemiological survey reports and disease news information.

[0119] For example, the activity trajectories and symptom descriptions of cases in investigation reports, the disease transmission dynamics and public discussion heat in news reports, etc. can use natural language processing technology to extract key entities and relationships and convert them into structured feature vectors for analysis.

[0120] The pathogenic data characteristics of pathogenic data focus on the pathogen itself, such as viral gene sequences, bacterial culture results, etc., which are obtained after analyzing key sites and help understand the transmission ability and pathogenicity of pathogens.

[0121] In the present invention, multi-source fusion features integrate different types of disease data into a unified feature representation, providing a more comprehensive information basis for disease analysis.

[0122] Optionally, constructing a regional graph network of the target area based on the multi-source disease data includes:

[0123] Divide the target area into multiple sub-areas, and determine the edge weights between the sub-areas based on the number of common activity scenes of disease infected persons between the sub-areas and the total number of activity scenes of disease infected persons in the sub-areas in the epidemiological survey report of the multi-source disease data;

[0124] Different sub-regions are connected based on the edge weights between the sub-regions to construct a regional graph network of the target region.

[0125] In the present invention, a target area (such as a city) is divided into multiple sub-areas, for example, based on streets or communities.

[0126] Each sub-region is a node, and the node's feature vector includes the following indicators:

[0127] Pharmacy sales: sales of antipyretics, cough suppressants and other disease-related medicines in pharmacies within the sub-region; school absenteeism: the proportion of absent students in schools within the sub-region; number of medical visits: the number of visits to hospitals and clinics within the sub-region; medical resource utilization rate: the utilization of medical resources (such as beds, equipment, etc.) within the sub-region.

[0128] In the present invention, the number of common activity places refers to the number of common activity places between sub-area i and sub-area j. These places are potential high-risk areas for disease transmission, such as shopping malls, parks, schools, transportation hubs, etc.

[0129] Count the number of common activity venues between sub-regions i and j. These venues are potentially high-risk areas for disease transmission, such as shopping malls, parks, schools, and transportation hubs. Also, count the total number of activity venues in sub-region i.

[0130] The calculation method of edge weight is as follows:

[0131] ;

[0132] In the present invention, based on the above node definition and edge weight calculation, a regional graph network of the target area is constructed.

[0133] Each node represents a subregion, and the strength of the connection between nodes is determined by the edge weight. A graph data structure (such as an adjacency matrix or adjacency list) is used to represent the regional graph network, where nodes represent subregions and edges represent the strength of the connection between subregions.

[0134] In this paper, the regional graph network can intuitively display the spread of the disease in different subregions, helping to analyze the disease's transmission path and high-risk areas. By calculating edge weights, the regional graph network can accurately reflect the strength of the connection between subregions and ensure the proper alignment of data in the spatial dimension.

[0135] Optionally, after the regional graph network and the multi-source fusion features of the multi-source disease data are spliced ​​together, the splicing is input into a disease data analysis model including multiple expert agents, and the disease analysis results of the target region are output, including:

[0136] After splicing the regional graph network and the multi-source fusion features of the multi-source disease data, the splicing is input into a disease data analysis model including multiple expert agents to obtain disease occurrence probability information, disease severity information and disease description text output by each expert agent;

[0137] Determine a prediction result of the disease occurrence probability of the target area based on the disease occurrence probability information corresponding to each expert agent and the confidence level of the expert agent;

[0138] Determining a disease severity prediction result for the target area based on the disease severity information corresponding to each expert agent and the expert agent confidence level;

[0139] Determining a disease analysis result for the target area based on the disease occurrence probability prediction result and the disease severity prediction result for the target area;

[0140] The confidence of the expert agent is determined based on the quantitative certainty of the expert agent and the historical prediction accuracy.

[0141] In this paper, the regional graph network provides spatial information about each sub-region within the target region and their interconnectedness. By splicing, the model can understand the disease's spread path and intensity in different sub-regions, as well as the risk of spread between sub-regions.

[0142] Multi-source fusion features include key information extracted from various data sources, such as numerical data such as pharmacy sales, school absenteeism, and medical consultations, as well as textual data such as epidemiological survey reports and disease news. This information reflects different aspects of disease transmission.

[0143] Combining the regional graph network with multi-source fusion features enables the model to simultaneously utilize spatial information and multi-source features for comprehensive analysis. This helps the model more accurately grasp the patterns of disease transmission and improve the accuracy of predictions of disease likelihood and severity.

[0144] In the present invention, each expert agent outputs a disease occurrence probability value Pk based on the input feature vector, that is, the spliced ​​regional graph network and multi-source fusion features, which represents the agent k's prediction of the possibility of disease occurrence in the target area.

[0145] The confidence of each expert agent It is determined based on its quantitative certainty and historical prediction accuracy. is the historical prediction accuracy of agent k, calculated as the weighted mean absolute error.

[0146] In the present invention, according to the quantitative certainty It is the quantitative value of the evidence certainty of agent k, which is evaluated based on the word frequency of uncertainty words and the coverage of key entities in the evidence text.

[0147] Prediction results of the likelihood of disease occurrence and disease severity prediction results Perform geometric averaging to obtain a comprehensive risk assessment index Final_Probability,

[0148] In the present invention, the outbreak possibility and severity of the disease are combined through the geometric mean method to obtain a comprehensive disease risk assessment, providing stronger support for decision makers.

[0149] Optionally, the method for obtaining the disease occurrence probability prediction result includes:

[0150] Summing the disease occurrence probability information of each expert agent and the product of the expert agent confidence to obtain a first summation result;

[0151] Determining the disease occurrence probability prediction result according to the first summation result and the summation result of the confidence levels of the respective expert agents;

[0152] The method for obtaining the disease severity prediction result includes:

[0153] Summing the disease severity information of each expert agent and the product of the expert agent confidence to obtain a second summation result;

[0154] The disease severity prediction result is determined based on the sum of the second summation result and the confidence level of each expert agent.

[0155] In the present invention, the disease occurrence probability information of all expert agents is calculated The corresponding confidence The sum of the products, that is, the first summation result , summarizes the importance of each agent's prediction results, and agents with higher confidence have a greater impact on the total.

[0156] More specifically, the confidence of all expert agents is calculated The sum of , which is used to normalize the sum of the molecular parts to ensure that the final result is within a reasonable numerical range.

[0157] By dividing the numerator and denominator, we can get the predicted result of the disease occurrence probability in the target area:

[0158] ;

[0159] In the present invention, the disease severity information of all expert agents is calculated The corresponding confidence The sum of the products, that is, the second summation result , summarizes the importance of each agent's prediction results, and agents with higher confidence have a greater impact on the total.

[0160] More specifically, the confidence of all expert agents is calculated The sum of , which is used to normalize the sum of the molecular parts to ensure that the final result is within a reasonable numerical range.

[0161] By dividing the numerator and denominator, we can get the predicted result of the disease occurrence probability in the target area:

[0162] ;

[0163] In this invention, by weighting and integrating the prediction results of multiple expert agents, the expertise and perspectives of different agents can be comprehensively considered, preventing the prediction results of a single agent from being overly biased or affected by its limitations. Furthermore, agents with high confidence levels contribute more significantly to the final prediction results, allowing the prediction results of agents with high historical prediction accuracy and strong evidence certainty to be more prominently reflected in the final prediction of disease likelihood and severity. By weighting the prediction results using the confidence levels of expert agents, it is possible to filter out less accurate prediction information to a certain extent, reduce errors, and thus improve the accuracy and reliability of the entire prediction system.

[0164] Optionally, the method for calculating the confidence of the expert agent includes:

[0165] Determining the historical accuracy of the expert agent based on the difference between the disease occurrence probability information output by the expert agent at different time points and the actual disease occurrence probability information;

[0166] Determining the quantitative certainty of the expert agent based on the number of uncertainty words, the total number of words, and the number of key entity words in the disease description text output by the expert agent;

[0167] The confidence of the expert agent is obtained based on the historical accuracy of the expert agent and the mean of the quantitative certainty.

[0168] In the present invention, disease likelihood information is the probability of disease occurrence predicted by the agent, usually expressed as a value between 0 and 1. Actual disease likelihood information is the actual probability of disease occurrence derived from subsequent data or statistics, which is used to compare the prediction accuracy of the agent.

[0169] In the disease description text output by the agent, the number of uncertainty words refers to the number of times words such as "maybe" and "perhaps" appear, which reflect the uncertainty of the prediction results.

[0170] The total number of words is the total number of all words in the text and is used to calculate the proportion of uncertain words.

[0171] The number of key entity words refers to important disease-related terms in the text, such as "fever" and "cough". The number of these words can reflect the richness and relevance of text information.

[0172] By comprehensively considering the deviation between the agent's historical predictions and actual results (historical accuracy), as well as the presence of uncertain terms and key entity terms in the output text (quantified certainty), we can calculate the confidence level of each expert agent. This comprehensive evaluation method helps select more reliable and accurate agent predictions, thereby improving the predictive performance of the entire system.

[0173] Specifically, the confidence of each expert agent It is determined based on its quantitative certainty and historical prediction accuracy. is the historical prediction accuracy of agent k, calculated by weighted mean absolute error (WMAE).

[0174]

[0175] in, Expert Agent k In time t The prediction results, refers to the actual results, is the time decay weight.

[0176] In the present invention, according to the quantitative certainty It is the quantitative value of the evidence certainty of agent k, which is evaluated based on the word frequency of uncertainty words and the coverage of key entities in the evidence text.

[0177]

[0178] in, Words that express uncertainty, Indicates the number of key entities, Indicates the total number of words in the evidence text.

[0179] The confidence of each expert agent It is determined based on its quantitative certainty and historical prediction accuracy, and the formula is as follows:

[0180]

[0181] In the present invention, the mean of historical accuracy and quantitative certainty is used as the confidence information of the intelligent agent. It can provide a comprehensive credibility assessment based on the historical prediction performance and output text quality of the intelligent agent. It takes into account both the past prediction accuracy of the intelligent agent and the certainty and information richness of its output results, making the confidence information more comprehensive and reliable.

[0182] Optionally, after splicing the region graph network and the multi-source fusion features of the multi-source disease data, inputting the spliced ​​features into a disease data analysis model comprising multiple expert agents, and before outputting the disease analysis results of the target region, the method further includes:

[0183] The spliced ​​regional graph network sample and the multi-source fusion feature sample are used as a training sample to obtain multiple training samples;

[0184] For each training sample, the training sample is used as an environmental state and inputted into each expert agent in the disease data analysis model to obtain experience data of each expert agent, and the experience data is stored in the experience pool corresponding to each expert agent; wherein the experience data includes: agent reward value and agent action selection;

[0185] Calculating the individual advantage of each expert agent based on the agent reward value in the experience pool of each expert agent;

[0186] Calculating a loss value for each expert agent based on the individual advantage and action selection strategy ratio of each expert agent, so as to optimize the policy network parameters of the corresponding expert agent based on the loss value;

[0187] Traverse each of the training samples until the preset training conditions are met, and obtain a disease data analysis model including multiple expert agents.

[0188] In this paper, the regional graph network uses subregions of the target region as nodes, and node characteristics include pharmacy sales, school absenteeism, number of medical visits, and medical resource utilization. Edge weights are determined based on the ratio of the number of shared activity venues between subregions to the total number of activity venues in a single subregion.

[0189] Multi-source fusion features are generated by fusing numerical data (such as pharmacy sales data and hospital visit data) with textual data (such as epidemiological survey reports and disease news) through a cross-modal attention mechanism. The feature vector of the regional graph network and the multi-source fusion feature vector are sequentially concatenated into a unified feature vector.

[0190] In this paper, experience data includes the agent's reward value and action selection. The reward value is determined by the degree to which the agent's output matches the actual result. The action selection is the action the agent takes under a given environmental state, such as the predicted likelihood and severity of a disease.

[0191] In this paper, individual advantage refers to the measure of the advantage of an agent taking a specific action in a specific environmental state, which is obtained by calculating the difference between the agent's reward value and the long-term return expectation estimated by the value network.

[0192] In the present invention, the loss value can be calculated based on the individual advantage and action selection strategy ratio of the agent, and is used to evaluate the quality of the agent's strategy.

[0193] Policy network parameter optimization uses optimization algorithms such as gradient descent to adjust the agent's policy network parameters according to the loss value to improve the agent's decision-making ability.

[0194] More specifically, in the present invention, domain expert agents can be constructed first, and each agent k (such as Epi-Agent, Pathogen-Agent) independently generates trajectories. ,in: is the environmental state faced by agent k (i.e., multi-channel fusion information at a certain moment); is the judgment result of agent k at time t (such as event level and handling suggestions); The reward of agent k at time t;

[0195] Construct an individual advantage value evaluation system to evaluate the independent strategic performance of each agent. The evaluation requires calculating the cumulative decision-making-reward and punishment value of a single agent in the entire process. The formula is: ;

[0196] in, The reward or penalty value of a certain disease at a certain time. is the discount factor (default 0.99); The value network of agent k, used to estimate the state long-term return expectations.

[0197] In the present invention, the spliced ​​regional graph network sample and the multi-source fusion feature sample are used as a training sample to obtain multiple training samples.

[0198] Each training sample is fed into the disease data analysis model as an environmental state. The agent selects an action based on its current strategy, such as predicting the likelihood and severity of a disease, and receives a reward based on how closely its output matches the actual outcome. This experience data, including agent rewards and action choices, is collected and stored in the corresponding expert agent's experience pool.

[0199] Using the rewards in the experience pool and the individual advantage formula above, we calculate the individual advantage of each expert agent. We also calculate the loss and optimize the policy network parameters: We calculate the loss for each agent based on its individual advantage and the action selection strategy ratio (i.e., the ratio of the probability of selecting an action under the current strategy to the probability of selecting an action under the previous strategy). We then use an optimization algorithm, such as gradient descent, to adjust the policy network parameters of the corresponding agent based on the loss to improve decision quality.

[0200] Repeat the above steps and traverse all training samples until the preset training conditions are met, such as the loss value convergence or the maximum number of training rounds is reached, and finally a disease data analysis model containing multiple expert agents is obtained.

[0201] In this invention, the training method effectively improves the credibility and accuracy of the agent's output. By repeatedly training with multiple training samples, the agent optimizes the policy network parameters during continuous learning, thereby achieving even better performance in disease data analysis. This approach ensures that the disease data analysis model can output reliable predictions of disease likelihood and severity based on high-quality training data and optimized policy parameters. This provides strong support for early disease detection, the development of prevention and control strategies, and the rational allocation of resources.

[0202] Optionally, obtaining multi-source disease data associated with the target area within a preset time period includes:

[0203] Standardizing the pharmacy sales data, school absenteeism, and hospital visit volume using a sliding window standardization method to obtain standardized pharmacy sales data, school absenteeism, and hospital visit volume data;

[0204] Call the pre-configured large language model to extract key disease entities and relationships from news information data to obtain disease news information data;

[0205] When the virus sequence data of the disease is obtained, the pre-configured large language model is called to analyze the key sites in the virus sequence data to obtain the pathogen data.

[0206] In this paper, sliding window normalization is a data normalization technique that normalizes data by calculating the mean and standard deviation within a window. It is used to eliminate periodic fluctuations and dimensional differences in data, making the data more stable and suitable for subsequent analysis.

[0207] The Big Language Model is a natural language processing model based on deep learning. Trained on large amounts of text data, it can understand and generate natural language text. It is used to extract key entities and relationships from text data and convert unstructured text information into structured feature vectors.

[0208] Virus sequence data analysis uses bioinformatics methods to analyze virus sequence data and identify key sites. This helps understand the virus's transmissibility and pathogenicity, providing a basis for disease prevention and control.

[0209] Specifically, we obtain pharmacy sales data, school absence data, hospital visit data, and news information data for the target area within a preset time period and use sliding window standardization:

[0210]

[0211] in, is the mean value in the window [tw, t], is the standard deviation, w is the window interval, which is related to the disease type and adapts to periodic fluctuations (such as W = 7 days for respiratory tract). For the data after standardization, Indicates t Data at the moment.

[0212] In an optional embodiment, for the time series data of school absences, the weekend effect needs to be eliminated: the average daily absence rate (Assuming 5 teaching days per week).

[0213] In this method, news data is fed into a preconfigured large language model. Leveraging the model's natural language processing capabilities, key disease entities (such as symptoms and transmission pathways) and relationships within the text are extracted. The extracted key entities and relationships are then converted into structured feature vectors.

[0214] In an alternative embodiment, viral sequence data is input into a preconfigured bioinformatics model. The model analyzes key sites in the viral sequence to identify important sites associated with disease transmission and pathogenicity. The analysis results are converted into structured pathogen data feature vectors.

[0215] In the present invention, by processing and converting multi-source disease data, high-quality input features are effectively provided for the disease data analysis model.

[0216] The disease data analysis device provided by the present invention is described below. The disease data analysis device described below and the disease data analysis method described above can be referenced to each other.

[0217] Figure 2 The structural diagram of the disease data analysis device provided by the present invention is as follows: Figure 2 As shown, including:

[0218] The acquisition module 210 is configured to acquire multi-source disease data associated with a target area within a preset time period; wherein the multi-source disease data includes at least one of the following: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data within the target area;

[0219] The processing module 220 is used to construct a regional map network of the target area based on the positioning information of the multi-source disease data;

[0220] The analysis module 230 is used to splice the regional graph network and the multi-source fusion features of the multi-source disease data, input them into a disease data analysis model containing multiple expert agents, and output the disease analysis results of the target area.

[0221] The present invention also provides a disease data analysis device, which is further used for:

[0222] Mapping the numerical data to a query space through a first weight matrix to obtain a query vector, and mapping the text data to a key space through a second weight matrix to obtain a key vector;

[0223] Determining a normalized weighted attention according to the query vector and the key vector, and performing weighted summation on the text features of the text data according to the weighted attention to obtain a fused text feature representation;

[0224] The fused text feature representation, the numerical feature representation of the numerical data, and the pathogenic data feature representation of the pathogenic data are spliced ​​together to obtain a multi-source fusion feature.

[0225] The present invention also provides a disease data analysis device, which is further used for:

[0226] Divide the target area into multiple sub-areas, and determine the edge weights between the sub-areas based on the number of common activity scenes of disease infected persons between the sub-areas and the total number of activity scenes of disease infected persons in the sub-areas in the epidemiological survey report of the multi-source disease data;

[0227] Different sub-regions are connected based on the edge weights between the sub-regions to construct a regional graph network of the target region.

[0228] The present invention also provides a disease data analysis device, which is further used for:

[0229] After splicing the regional graph network and the multi-source fusion features of the multi-source disease data, the splicing is input into a disease data analysis model including multiple expert agents to obtain disease occurrence probability information, disease severity information and disease description text output by each expert agent;

[0230] Determine a prediction result of the disease occurrence probability of the target area based on the disease occurrence probability information corresponding to each expert agent and the confidence level of the expert agent;

[0231] Determining a disease severity prediction result for the target area based on the disease severity information corresponding to each expert agent and the expert agent confidence level;

[0232] Determining a disease analysis result for the target area based on the disease occurrence probability prediction result and the disease severity prediction result for the target area;

[0233] The confidence of the expert agent is determined based on the quantitative certainty of the expert agent and the historical prediction accuracy.

[0234] The present invention also provides a disease data analysis device, which is further used for:

[0235] Summing the disease occurrence probability information of each expert agent and the product of the expert agent confidence to obtain a first summation result;

[0236] Determining the disease occurrence probability prediction result according to the first summation result and the summation result of the confidence levels of the respective expert agents;

[0237] The method for obtaining the disease severity prediction result includes:

[0238] Summing the disease severity information of each expert agent and the product of the expert agent confidence to obtain a second summation result;

[0239] The disease severity prediction result is determined based on the sum of the second summation result and the confidence level of each expert agent.

[0240] The present invention also provides a disease data analysis device, which is further used for:

[0241] Determining the historical accuracy of the expert agent based on the difference between the disease occurrence probability information output by the expert agent at different time points and the actual disease occurrence probability information;

[0242] Determining the quantitative certainty of the expert agent based on the number of uncertainty words, the total number of words, and the number of key entity words in the disease description text output by the expert agent;

[0243] The confidence of the expert agent is obtained based on the historical accuracy of the expert agent and the mean of the quantitative certainty.

[0244] The present invention also provides a disease data analysis device, which is further used for:

[0245] The spliced ​​regional graph network sample and the multi-source fusion feature sample are used as a training sample to obtain multiple training samples;

[0246] For each training sample, the training sample is used as an environmental state and inputted into each expert agent in the disease data analysis model to obtain experience data of each expert agent, and the experience data is stored in the experience pool corresponding to each expert agent; wherein the experience data includes: agent reward value and agent action selection;

[0247] Calculating the individual advantage of each expert agent based on the agent reward value in the experience pool of each expert agent;

[0248] Calculating a loss value for each expert agent based on the individual advantage and action selection strategy ratio of each expert agent, so as to optimize the policy network parameters of the corresponding expert agent based on the loss value;

[0249] Traverse each of the training samples until the preset training conditions are met, and obtain a disease data analysis model including multiple expert agents.

[0250] The present invention also provides a disease data analysis device, which is further used for:

[0251] Standardizing the pharmacy sales data, school absenteeism, and hospital visit volume using a sliding window standardization method to obtain standardized pharmacy sales data, school absenteeism, and hospital visit volume data;

[0252] Call the pre-configured large language model to extract key disease entities and relationships from news information data to obtain disease news information data;

[0253] When the virus sequence data of the disease is obtained, the pre-configured large language model is called to analyze the key sites in the virus sequence data to obtain the pathogen data.

[0254] In this invention, by collecting multi-source disease data such as pharmacy sales data, school absenteeism, and hospital visit volume, the disease-related dynamics in the target area can be fully reflected. The cross-modal attention mechanism can effectively fuse numerical data and text data. By learning the correlation between different modal data, the model can more accurately capture the key characteristics of disease transmission and improve the accuracy of data analysis. The constructed regional graph network uses geographical regions as nodes and can intuitively display the spread of diseases in different regions. The disease data analysis model containing multiple expert agents can analyze disease data from different angles and ultimately generate accurate and effective disease data analysis results.

[0255] Figure 3 Schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 3 As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other via the communications bus 340. The processor 310 may invoke logic instructions in the memory 330 to execute a disease data analysis method, which includes: obtaining multi-source disease data associated with a target area within a preset time period; wherein the multi-source disease data includes at least one of the following: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data within the target area;

[0256] constructing a regional graph network of the target area based on the multi-source disease data;

[0257] After the regional graph network and the multi-source fusion features of the multi-source disease data are spliced ​​together, the spliced ​​data are input into a disease data analysis model comprising multiple expert agents, and the disease analysis results of the target region are output.

[0258] Furthermore, the logic instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0259] In another aspect, the present invention further provides a computer program product, comprising a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the disease data analysis method provided by each of the above methods, the method comprising: obtaining multi-source disease data associated with a target area within a preset time period; wherein the multi-source disease data includes at least one of the following: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data within the target area;

[0260] constructing a regional graph network of the target area based on the multi-source disease data;

[0261] After the regional graph network and the multi-source fusion features of the multi-source disease data are spliced ​​together, the spliced ​​data are input into a disease data analysis model comprising multiple expert agents, and the disease analysis results of the target region are output.

[0262] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the disease data analysis method provided by each of the above methods, the method comprising: obtaining multi-source disease data associated with a target area within a preset time period; wherein the multi-source disease data includes at least one of the following: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data within the target area;

[0263] constructing a regional graph network of the target area based on the multi-source disease data;

[0264] After the regional graph network and the multi-source fusion features of the multi-source disease data are spliced ​​together, the spliced ​​data are input into a disease data analysis model comprising multiple expert agents, and the disease analysis results of the target region are output.

[0265] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0266] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0267] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A disease data analysis method, characterized in that: include: Acquire multi-source disease data associated with a target area within a preset time period; wherein the multi-source disease data includes: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data within the target area; Based on the multi-source disease data, a regional graph network of the target area is constructed; after splicing the regional graph network and the multi-source fusion features of the multi-source disease data, the spliced ​​data are input into a disease data analysis model comprising multiple expert agents, and the disease analysis results of the target area are output; The multiple expert agents include: clinical medicine experts, disease prevention and control experts, virology experts, data scientists, and medical resource planning experts; The regional graph network and the multi-source fusion features of the multi-source disease data are spliced ​​together and input into a disease data analysis model including multiple expert agents to output the disease analysis results of the target region, including: After splicing the regional graph network and the multi-source fusion features of the multi-source disease data, the splicing is input into a disease data analysis model including multiple expert agents to obtain disease occurrence probability information, disease severity information and disease description text output by each expert agent; Determine a prediction result of the disease occurrence probability of the target area based on the disease occurrence probability information corresponding to each expert agent and the confidence level of the expert agent; Determining a disease severity prediction result for the target area based on the disease severity information corresponding to each expert agent and the expert agent confidence level; Determining a disease analysis result for the target area based on the disease occurrence probability prediction result and the disease severity prediction result for the target area; The method for calculating the confidence of the expert agent includes: Determining the historical accuracy of the expert agent based on the difference between the disease occurrence probability information output by the expert agent at different time points and the actual disease occurrence probability information; Determining the quantitative certainty of the expert agent based on the number of uncertainty words, the total number of words, and the number of key entity words in the disease description text output by the expert agent; The confidence of the expert agent is obtained based on the historical accuracy of the expert agent and the mean of the quantitative certainty.

2. The disease data analysis method according to claim 1, characterized in that: Constructing a regional graph network of the target area based on the multi-source disease data, including: Divide the target area into multiple sub-areas, and determine the edge weights between the sub-areas based on the number of common activity scenes of disease infected persons between the sub-areas and the total number of activity scenes of disease infected persons in the sub-areas in the epidemiological survey report of the multi-source disease data; Different sub-regions are connected based on the edge weights between the sub-regions to construct a regional graph network of the target region.

3. The disease data analysis method according to claim 1, characterized in that: The method for obtaining the disease occurrence probability prediction result includes: Summing the disease occurrence probability information of each expert agent and the product of the expert agent confidence to obtain a first summation result; Determining the disease occurrence probability prediction result according to the first summation result and the summation result of the confidence levels of the respective expert agents; The method for obtaining the disease severity prediction result includes: Summing the disease severity information of each expert agent and the product of the expert agent confidence to obtain a second summation result; The disease severity prediction result is determined based on the sum of the second summation result and the confidence level of each expert agent.

4. The disease data analysis method according to claim 1, characterized in that: After splicing the region graph network and the multi-source fusion features of the multi-source disease data, the splicing is input into a disease data analysis model comprising a plurality of expert agents, and before the step of outputting the disease analysis result of the target region, the method further comprises: The spliced ​​regional graph network sample and the multi-source fusion feature sample are used as a training sample to obtain multiple training samples; For each training sample, the training sample is used as an environmental state and inputted into each expert agent in the disease data analysis model to obtain experience data of each expert agent, and the experience data is stored in the experience pool corresponding to each expert agent; wherein the experience data includes: agent reward value and agent action selection; Calculating the individual advantage of each expert agent based on the agent reward value in the experience pool of each expert agent; Calculating a loss value for each expert agent based on the individual advantage and action selection strategy ratio of each expert agent, so as to optimize the policy network parameters of the corresponding expert agent based on the loss value; Traverse each of the training samples until the preset training conditions are met, and obtain a disease data analysis model including multiple expert agents.

5. The disease data analysis method according to claim 1, characterized in that: The acquiring of multi-source disease data associated with the target area within a preset time period includes: Standardizing the pharmacy sales data, school absenteeism, and hospital visit volume using a sliding window standardization method to obtain standardized pharmacy sales data, school absenteeism, and hospital visit volume data; Call the pre-configured large language model to extract key disease entities and relationships from news information data to obtain disease news information data; When the virus sequence data of the disease is obtained, the pre-configured large language model is called to analyze the key sites in the virus sequence data to obtain the pathogen data.

6. The disease data analysis method according to claim 1, characterized in that: The multi-source fusion features include: numerical features of numerical data, text features of text data and pathogenic data features of pathogenic data; The numerical data includes at least one of the following: pharmacy sales data, school absence data, and hospital visit data; The text data includes at least one of the following: epidemiological survey reports and disease news information data.

7. A disease data analysis device, characterized in that: include: an acquisition module, configured to acquire multi-source disease data associated with a target area within a preset time period; wherein the multi-source disease data includes: pharmacy sales data, school absenteeism, hospital visit data, pathogen data, epidemiological survey reports, and disease news information data within the target area; a processing module, configured to construct a regional map network of the target area based on the positioning information of the multi-source disease data; An analysis module, configured to combine the regional graph network and the multi-source fusion features of the multi-source disease data, input the combined features into a disease data analysis model comprising multiple expert agents, and output a disease analysis result for the target region; Among them, the plurality of expert agents include: clinical medicine experts, disease prevention and control experts, virology experts, data scientists, and medical resource planning experts; Wherein, the device is also used for: After splicing the regional graph network and the multi-source fusion features of the multi-source disease data, the splicing is input into a disease data analysis model including multiple expert agents to obtain disease occurrence probability information, disease severity information and disease description text output by each expert agent; Determine a prediction result of the disease occurrence probability of the target area based on the disease occurrence probability information corresponding to each expert agent and the confidence level of the expert agent; Determining a disease severity prediction result for the target area based on the disease severity information corresponding to each expert agent and the expert agent confidence level; Determining a disease analysis result for the target area based on the disease occurrence probability prediction result and the disease severity prediction result for the target area; The method for calculating the confidence of the expert agent includes: Determining the historical accuracy of the expert agent based on the difference between the disease occurrence probability information output by the expert agent at different time points and the actual disease occurrence probability information; Determining the quantitative certainty of the expert agent based on the number of uncertainty words, the total number of words, and the number of key entity words in the disease description text output by the expert agent; The confidence of the expert agent is obtained based on the historical accuracy of the expert agent and the mean of the quantitative certainty.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the disease data analysis method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the disease data analysis method according to any one of claims 1 to 6 is implemented.

Images