Big data analysis-based infectious disease transmission risk prediction method and system

Abstract

The invention relates to the technical field of infectious disease prediction, in particular to an infectious disease transmission risk prediction method and system based on big data analysis, and the method comprises the steps: firstly collecting multi-source heterogeneous data of population flow, environmental meteorology, historical epidemic situation, social media and public health resources, and then carrying out the cleaning, fusion and space-time alignment of the data, constructing a standardized database, training a dynamic propagation model based on machine learning, iteratively optimizing parameters by combining an SEIR improved model and real-time data, and finally outputting a regional risk level prediction result and a visual early warning map; according to the method, the multi-source heterogeneous data are fused, and the dynamic machine learning model is used for real-time analysis, so that complex influence factors of virus transmission can be more comprehensively captured, prediction deviation caused by single data or update delay in a traditional method is remarkably reduced, the risk early warning precision is improved, and the risk early warning efficiency is improved. And minute-level data processing and risk level updating are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infectious disease prediction, and in particular to an infectious disease transmission risk prediction method and system based on big data analysis. BACKGROUND

[0002] The existing infectious disease transmission risk prediction technology mainly relies on statistical analysis of historical epidemic data or propagation dynamics model based on fixed parameters. For example, some systems use the traditional SEIR (Susceptible-Exposed-Infected-Recovered) model, combine census data and static environmental parameters for prediction, and some methods try to integrate social media public opinion or mobile communication data, but usually only as auxiliary input, without realizing the deep collaborative analysis of multi-source heterogeneous data.

[0003] The existing method is often difficult to effectively integrate cross-domain data such as population flow, meteorological conditions and social media public opinion, so that the prediction model cannot dynamically adapt to the complex and changeable transmission environment. The traditional model may ignore the impact of sudden population migration such as holiday mass flow on virus spread, or fail to timely capture the potential role of abnormal climate conditions on virus survival rate, thereby reducing the prediction accuracy. At the same time, most systems rely too much on periodic batch data processing, cannot realize minute-level risk update, and the model parameters are fixed, making it difficult to dynamically optimize through real-time feedback. For example, if the model fails to adjust the warning level according to real-time public opinion data in the early stage of the epidemic, it may lead to a lag in prevention and control response.

[0004] Therefore, in order to solve the problems of insufficient data fusion capability and real-time deficiency, the present application provides an infectious disease transmission risk prediction method and system based on big data analysis, which realizes high-precision and low-delay risk warning through heterogeneous data spatio-temporal alignment, reinforcement learning driven parameter adaptive optimization and edge computing real-time processing. SUMMARY

[0005] In order to overcome the problems of insufficient data fusion capability and real-time deficiency, the present application provides an infectious disease transmission risk prediction method and system based on big data analysis.

[0006] The technical scheme of the present application is: an infectious disease transmission risk prediction method and system based on big data analysis, comprising the following steps: S1, collecting multi-source heterogeneous data, including population flow data, environmental meteorological data, historical epidemic data, social media public opinion data and public health resource data; S2, cleaning, fusing and spatio-temporal aligning the data, and constructing a standardized infectious disease risk analysis database; S3, training a dynamic propagation model based on machine learning algorithm, combining an improved SEIR model and iteratively optimizing parameters with real-time data; S4, output the regional risk level prediction results and the visual warning map.

[0007] As preferred, the population flow data is processed by integrating mobile terminal base station signaling data, urban public transportation card swiping records, and inter-provincial and inter-city air and rail passenger volume data, using Apache Kafka to build a real-time data pipeline for streaming processing, and using complex network theory to build a directed and weighted graph model, wherein the nodes represent administrative regions, and the edge weights are determined by the population migration scale and the traffic frequency. A gravity model is introduced to calibrate the inter-regional attraction factor, and a community discovery algorithm is used to identify high-risk transmission clusters. The activity track of potential super spreaders is tracked by combining trip chain analysis techniques. Finally, the probability distribution map of the cross-regional transmission path is output. The model is dynamically updated every 6 hours to reflect the spatio-temporal changes in population flow.

[0008] As preferred, the environmental meteorological data includes: temperature, relative humidity, wind speed, precipitation probability, and PM2.5 data of the target area obtained through meteorological bureau API and Internet of Things sensor network, and the missing values of the areas not covered by the monitoring sites are filled using the Kriging spatial interpolation algorithm. Time series alignment technology is used to match the meteorological data with the case report data in the same spatio-temporal grid. The influence of environmental factors on virus transmission is quantified based on the Bayesian hierarchical model. Meteorological numerical prediction data is introduced to predict the environmental conditions for the next 7 days. The contact transmission probability parameter of the SEIR improved model is dynamically adjusted based on the virus survival rate experimental data.

[0009] As preferred, the method uses natural language processing techniques to perform sentiment analysis and keyword extraction on social media public opinion data, monitors abnormal health event reports in real time, and corrects the risk prediction results.

[0010] As preferred, the dynamic transmission model introduces a reinforcement learning mechanism, which includes: in the framework of the SEIR improved model, the basic reproduction number, the latent period length, and the proportion of asymptomatic infections are defined as adjustable parameters, the initial values are determined by fitting historical epidemic data, the daily increase in laboratory-confirmed cases, the nucleic acid test positive rate, and the close contact tracing results are used as real-time feedback signals, the parameter adjustment direction is calculated through the policy gradient algorithm, and a reward function is designed to consider the prediction error, model stability, and public health cost. The model generates a set of parameter candidates every 24 hours, which is updated to the next period after being verified by disease control experts, forming a closed-loop optimization system of "data-driven-expert intervention".

[0011] As preferred, an infectious disease transmission risk prediction system based on big data analysis includes: A multi-source data acquisition module is used to connect government open data platforms, Internet of Things sensors, and third-party data interfaces. Distributed computing engine: Real-time stream processing and batch analysis of massive data based on Spark implementation; Risk prediction model library: Integration of infectious disease dynamics model, graph neural network and time series prediction algorithm; Visual interactive terminal: Generate dynamic heat map, risk trend curve and prevention and control suggestion report.

[0012] As preferred, the system further comprises an edge computing node deployed in a key area to support localized real-time risk calculation.

[0013] As preferred, the edge computing node is equipped with a GPU acceleration module to support local real-time calculation, each node synchronizes core parameters with the center cloud platform through a 5G private network or optical fiber but independently performs risk prediction tasks, the node is built-in with a lightweight infectious disease model and receives real-time data streams of local Internet of Things devices, combined with the regional population density and building layout stored in the edge, a micro-scale transmission risk index is generated every 5 minutes, if the risk threshold is broken through, local alarm is triggered immediately and an encrypted summary is uploaded to the center system, at the same time, the federal learning mechanism with adjacent nodes is started, and the center cloud platform regularly pushes global model updates to the edge nodes.

[0014] As preferred, the visual interactive terminal supports multi-level permission management, provides a decision support interface for government departments and publishes personalized early warning information to the public.

[0015] As preferred, the system is based on the consortium chain of Hyperledger Fabric, medical institutions, CDCs and research institutions participate in consensus as nodes, uses smart contracts to control data access permissions, sensitive data is encrypted by SM4 and stored as a hash value, and the original data is distributed stored by IPFS.

[0016] The beneficial effects of the present application are:

[0017] 1. By fusing multiple source heterogeneous data and using dynamic machine learning model for real-time analysis, the complex influencing factors of virus transmission can be more comprehensively captured, thereby significantly reducing the prediction deviation caused by single data or update delay of traditional methods, thereby improving the accuracy of risk warning and realizing minute-level data processing and risk level update.

[0018] 2. The transmission risk prediction model of the present application can automatically adjust key parameters according to real-time feedback data without relying on manual intervention or fixed experience value, thereby maintaining the robustness of prediction in the rapid evolution stage of the epidemic, at the same time, combined with the distributed deployment of edge computing nodes, the system can maintain low delay response in high concurrency scenarios, ensuring localized real-time risk assessment in densely populated areas or sudden epidemic hotspot areas, thereby improving the accuracy and implementation efficiency of prevention and control measures. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A system framework schematic diagram of the present application is shown; Figure 2 A workflow schematic diagram of the present application is shown. DETAILED DESCRIPTION

[0020] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0021] The present application provides an embodiment: a method for predicting the risk of infectious disease transmission based on big data analysis, comprising the following steps: S1, collecting multi-source heterogeneous data, including population flow data, environmental meteorological data, historical epidemic data, social media public opinion data and public health resource data; S2, cleaning, fusing and spatio-temporal aligning the data, and constructing a standardized infectious disease risk analysis database; S3, training a dynamic transmission model based on a machine learning algorithm, combining an improved SEIR (Susceptible-Exposed-Infected-Recovered) model and real-time data to iteratively optimize parameters; S4, outputting regionalized risk level prediction results and visualized early warning maps.

[0022] As a preferred, the population flow data is processed by integrating mobile terminal base station signaling data, urban public transportation card swiping records and inter-provincial and inter-city air and rail passenger volume data, using Apache Kafka to build a real-time data pipeline for stream processing, and using complex network theory to build a directed and weighted graph model, wherein the nodes represent administrative regions, and the edge weights are determined by the population migration scale and the traffic frequency. A gravity model is introduced to calibrate the inter-regional attraction factor, and a community discovery algorithm is used to identify high-risk transmission clusters. Combined with the trip chain analysis technology, the activity track of potential super spreaders is tracked, and finally the probability distribution map of the cross-regional transmission path is output. The model is dynamically updated every 6 hours to reflect the spatio-temporal changes of population flow, and provides spatial diffusion parameters for the dynamic transmission model.

[0023] As preferred, the environmental meteorological data includes: obtaining the temperature, relative humidity, wind speed, precipitation probability and PM2.5 data of the target area through the meteorological bureau API and the Internet of Things sensor network, filling in the missing values of the areas covered by the monitoring stations using the Kriging spatial interpolation algorithm, and matching the meteorological data with the case report data in the same space-time grid using the time series alignment technology, quantifying the influence of environmental factors on virus transmission based on the Bayesian hierarchical model, introducing meteorological numerical prediction data to predict the environmental conditions in the next 7 days, and dynamically adjusting the contact transmission probability parameter of the SEIR improved model combined with the virus survival rate experimental data.

[0024] As preferred, the method uses natural language processing technology to perform sentiment analysis and keyword extraction on social media public opinion data, monitors abnormal health event reports in real time, and corrects the risk prediction results.

[0025] Further, through Twitter, microblog, news comments and other public text streams, unstructured text containing health keywords such as "fever" and "quarantine" is collected in real time using a distributed crawler framework, sentiment polarity classification (positive / negative / panic) and entity recognition (location, symptoms, drugs) are performed through the BERT pre-training model, and potential epidemic-related topics such as "vaccine shortage" and "clustered infection" are extracted using the LDA topic model. For sudden public opinion events such as a surge in discussions about unexplained pneumonia in a certain place, Grubbs anomaly detection algorithm is used to identify topic heat deviating from the historical baseline, and geographic coding technology is used to map the locations mentioned in the text to administrative divisions to generate a public opinion risk index.

[0026] As preferred, the dynamic transmission model introduces a reinforcement learning mechanism, which includes: in the framework of the SEIR improved model, the basic reproductive number, the length of the incubation period, and the proportion of asymptomatic infections are defined as adjustable parameters, the initial values are determined by fitting historical epidemic data, the daily increase in laboratory-confirmed cases, the nucleic acid test positive rate, and the close contact tracing results are used as real-time feedback signals, the parameter adjustment direction is calculated through the policy gradient algorithm, and the reward function is designed to consider the prediction error, model stability, and public health cost. A set of parameter candidates is generated every 24 hours, and after expert interaction verification, it is updated to the next period to form a "data-driven-expert intervention" closed-loop optimization system.

[0027] Please refer to Figure 1 , further, an infectious disease transmission risk prediction system based on big data analysis, comprising: A multi-source data acquisition module is used to connect to the government open data platform, the Internet of Things sensor, and the third-party data interface. A distributed computing engine is used to realize real-time stream processing and batch analysis of massive data based on Spark. Risk prediction model library: integrating epidemic dynamics model, graph neural network and time series prediction algorithm; Visual interactive terminal: generating dynamic heat map, risk trend curve and prevention and control suggestion report.

[0028] The system also includes edge computing nodes deployed in key areas to support localized real-time risk calculation.

[0029] Further, the deployment and operation mechanism of the edge computing nodes specifically includes: deploying edge servers in high-density population areas such as airports, train stations, large communities and hospitals, and equipping them with GPU acceleration modules to support local real-time calculation. Each node synchronizes core parameters with the central cloud platform through 5G private network or optical fiber but independently performs risk prediction tasks, thereby reducing response lag caused by network delay. The node is built-in with a lightweight infectious disease model, receives real-time data streams from local Internet of Things devices such as infrared temperature measurement cameras and air quality sensors, and combines static data such as population density and building layout in the area stored in the edge to generate a 500-meter grid transmission risk index every 5 minutes. If the risk threshold is broken (such as R0>1.5), local alarm is triggered immediately and encrypted summary is uploaded to the central system, while the federal learning mechanism with adjacent nodes is started to exchange model gradients instead of raw data to protect privacy. The central cloud platform pushes global model updates to edge nodes every morning to ensure the consistency of distributed calculation and centralized training.

[0030] The visual interactive terminal supports multi-level permission management, provides decision support interface for government departments, and publishes personalized early warning information to the public.

[0031] Further, the multi-level permission management and information release function of the visual interactive terminal is specifically implemented as follows: for government departments such as CDC and emergency management bureau, a decision cockpit interface based on WebGIS is provided, integrating risk heat map (classified by red / orange / yellow / blue four colors), medical resource pressure dashboard (ICU bed occupancy rate, nucleic acid testing queue length) and simulation tool, supporting multi-dimensional filtering and highlighting key monitoring area labeling of time sliding axis and administrative division drilling, for the public, through WeChat applet, SMS push and electronic fence technology, hierarchical early warning is released, such as sending personalized protection suggestions to high-risk area residents, such as avoiding peak travel or navigating to nearby nucleic acid testing point; and embedding anonymous travel comparison function (prompting the degree of coincidence with confirmed cases trajectory); permission control adopts RBAC (role-based access control) model, government administrators can view raw data and prediction details, community workers can only access district aggregated results, and the public is limited to receive desensitized information; all visual outputs meet WCAG 2.1 accessibility standards, including color blindness adaptation mode and voice broadcast option.

[0032] The system is based on a consortium blockchain of Hyperledger Fabric. Medical institutions, disease control centers and research institutions participate in consensus as nodes. Smart contracts are used to control data access permissions. Sensitive data is encrypted with SM4 and then stored on the blockchain as hash values. The original data is distributed and stored through IPFS.

[0033] Furthermore, data traceability and shared auditing are achieved through blockchain technology. The specific architecture of the blockchain data traceability and shared auditing is as follows: A Hyperledger Fabric consortium blockchain framework is adopted, with node members including top-tier hospitals, disease control centers, big data bureaus, and authorized research institutions. Each institution deploys Peer nodes locally and verifies their identity through MSP (Member Service Provider). For example, epidemic data such as case reports and close contact lists are encrypted using the national cryptographic SM4 algorithm when uploaded to the blockchain, and a Merkle tree digest is generated and stored on the chain. The original data hash value and access logs are automatically recorded through smart contracts, ensuring that any data modification behavior, such as a disease control center correcting the location of a case, is traceable and tamper-proof. Data sharing adopts a "permissioned + token incentive" mechanism. The requester needs to stake digital tokens and submit proof of purpose. After being approved by a majority vote of the nodes, a temporary access key is obtained. After use, the system automatically destroys the key and returns the staked tokens. Large-scale spatiotemporal trajectory data is stored off-chain via IPFS, while only its content identifier (CID) is stored on the chain, balancing storage efficiency and auditing requirements.

[0034] Please see Figure 2 Furthermore, the specific workflow of this invention will be described below: First, a distributed data crawling engine is used to collect multi-source heterogeneous data in real time, including mobile communication signaling data, public transportation card swipe records, air / railway passenger data, meteorological monitoring data, government-published epidemic reports, social media texts, and medical institution resource data. Apache NiFi is used to build a data pipeline for preliminary cleaning, and a spatiotemporal alignment algorithm is used to unify the data from different sources into a standardized spatiotemporal coordinate system, such as a 1km×1km grid with hourly time slices. Finally, a structured infectious disease risk analysis database is formed, providing high-quality input for subsequent modeling.

[0035] Based on the pre-processed data, the system initializes an improved SEIR dynamic model framework, introducing a reinforcement learning agent to dynamically regulate core parameters: including the basic reproduction number R0 (reflecting the virus transmission force), the inverse of the incubation period length σ, the proportion of asymptomatic infections ρ, and the environmental decay factor (quantifying the impact of weather on virus survival rate). During the model training phase, the parameters are first initialized using historical epidemic data through maximum likelihood estimation. Subsequently, after connecting to real-time data streams, parameter updates are performed every 6 hours. The residual error between the latest reported cases and the model's predicted values serves as the reward signal, guiding the parameter search direction through the policy gradient algorithm. Meanwhile, federated learning techniques are employed to aggregate local training results from edge nodes, such as the sensitivity coefficients of passenger density in airport area models, ensuring that the model can adapt to global trends while capturing regional specificity.

[0036] The trained model performs multi-scenario predictions of the transmission risk in a target region, such as a provincial administrative area, for the next 7 days: under the assumption that the current prevention and control policies remain unchanged, it outputs daily new case interval estimates, and in the intervention scenario of assuming the implementation of lockdown or increased detection efforts, it simulates the changes in the transmission curve. The prediction results are presented through three layers of visualization: on the macro level, a provincial risk heat map is generated (four-color classification of red / orange / yellow / blue), on the meso level, city-level medical resource pressure indices are marked (such as the projected gap in ICU beds), and on the micro level, high-risk points are labeled on the electronic map (such as the market with positive environmental testing). All visualizations are accelerated by WebGL rendering, supporting decision-makers in dragging the time axis to view prediction trends or clicking on regions to drill down into detailed data such as the infection probability of people over 60 years old in a certain district.

[0037] Edge nodes deployed in high-risk areas such as international airports continuously receive local Internet of Things device data (such as body temperature screening instruments and crowd density cameras), run lightweight LSTM models for 15-minute interval microgrid (200m x 200m) risk calculations, and if an anomaly is detected (such as a terminal building R0 instantaneous value > 2.0), trigger a three-level response: the first level responds to the on-site broadcast evacuation prompt and starts close contact tracing, the second level uploads encrypted warnings to the central platform to apply resource allocation, and the third level updates global model parameters through federated learning. Meanwhile, the public side APP receives personalized warnings (such as "You are in a mall where 3 close contacts have been detected within 1 hour"), and users can report additional data anonymously after confirming their itinerary, such as "wearing a mask at the time", forming a "monitoring-warning-feedback" closed-loop optimization.

[0038] All data exchange and model update records are stored through the Hyperledger Fabric consortium chain: the case reports uploaded by medical institutions are stored on the chain after SM3 hash processing and time stamping, and the research institutions need to submit the purpose proof specified by the smart contract when applying for data access, and obtain a temporary key after being verified by half of the nodes. The model iteration log (such as "2023-12-0110:00 R0 from 1.2 to 1.5") is stored through the Merkle tree structure, ensuring that any parameter modification can be traced back to the original data change. This mechanism not only guarantees the trusted cooperation between the CDC, hospitals and research teams, but also meets the audit requirements of the Data Security Law for sensitive public health data.

[0039] Further, the present application provides an embodiment based on multi-population flow for early warning of epidemic outbreak: During the multi-population flow, the system accesses real-time railway ticketing data, highway ETC pass records and mobile signaling data to construct a nationwide cross-provincial population migration network. The model detects a 200% surge in single-day migration from the south to the Yangtze River Delta region, combined with the low-temperature and high-humidity weather conditions in the north, predicts that the transmission risk index of this route rises to the red level, and the system automatically pushes the warning to the northern CDC, suggesting the addition of a body temperature screening channel at the train station and sending a nucleic acid test reminder to citizens who have recently traveled from the south to the north, which can effectively delay the spread of the epidemic.

[0040] Further, the present application provides an embodiment for community transmission risk prediction in typhoon weather: A coastal city is hit by a typhoon, and meteorological data detects that continuous heavy rain causes waterlogging in some areas. The system combines historical data to find that such an environment can extend the survival time of the influenza virus by 40%. At the same time, the social media public opinion module captures multiple community groups discussing "basement water leakage leading to the aggregation of people in shelters", and the edge computing node analyzes the shelter Bluetooth probe data in real time to find that the personnel density is 2.3 times higher than the standard. After comprehensive judgment by the model, the community transmission risk is raised to orange, triggering two responses: first, push the evacuation plan for taking materials in different time periods to the community; second, dispatch mobile PCR detection vehicles to the scene.

[0041] Further, the present application provides an embodiment for community transmission risk prediction in typhoon weather: After a star concert, the system obtains 60,000 seat distribution data through the ticket system, and reconstructs the audience flow heat map combined with the signaling data of the 5G base station in the venue. The next day, the public opinion module finds that the discussion volume of the micro-blog topic "concert fever after the concert" has increased sharply, and the NLP analysis confirms that it involves keywords such as cough and fever. The system immediately starts the super-spreader identification: first, compare the medical insurance data to find that 3 audience members have medical records that coincide with the concert seat area; second, through federated learning, the adjacent city edge node model is called to confirm that the R0 value of the virus strain is 4.7, and finally the B area 23 row is locked as a high-risk area, and the audience in this area is given priority to send isolation notifications, which cuts off the transmission chain 48 hours earlier than traditional flow adjustment.

Claims

1. A method for predicting the risk of infectious disease transmission based on big data analysis, characterized in that, It includes the following steps: S1 collects multi-source heterogeneous data, including population flow data, environmental and meteorological data, historical epidemic data, social media public opinion data, and public health resource data; S2, The data is cleaned, fused, and spatiotemporally aligned to construct a standardized infectious disease risk analysis database; S3 is based on machine learning algorithms to train a dynamic propagation model, and combines the SEIR improved model with real-time data to iteratively optimize parameters. S4 outputs regional risk level prediction results and a visualized early warning map.

2. The method for predicting the risk of infectious disease transmission based on big data analysis according to claim 1, characterized in that: The population flow data is integrated by combining mobile terminal base station signaling data, urban public transportation card swipe records, and inter-provincial air and rail passenger volume data. A real-time data pipeline is built using Apache Kafka for streaming processing. A directed weighted graph model is constructed using complex network theory, where nodes represent administrative regions, and edge weights are determined by the scale of population migration and traffic frequency. A gravity model is introduced to calibrate the attraction factor between regions. A community detection algorithm is used to identify high-risk transmission clusters, and travel chain analysis technology is combined to track the activity trajectories of potential super-spreaders. Finally, a probability distribution map of cross-regional transmission paths is output. The model is dynamically updated every 6 hours to reflect the spatiotemporal changes in population flow.

3. The method for predicting the risk of infectious disease transmission based on big data analysis according to claim 1, characterized in that, The environmental meteorological data includes: temperature, relative humidity, wind speed, precipitation probability, and PM2.5 data of the target area obtained through meteorological bureau API and IoT sensor network; missing values ​​in areas with insufficient monitoring station coverage are filled using Kriging spatial interpolation algorithm; meteorological data are matched with case report data in the same spatiotemporal grid using time series alignment technology; the impact of environmental factors on virus transmission is quantified based on Bayesian hierarchical model; environmental conditions for the next 7 days are predicted using meteorological numerical forecast data; and the contact transmission probability parameters of the SEIR improved model are dynamically adjusted in combination with virus survival rate experimental data.

4. The method for predicting the risk of infectious disease transmission based on big data analysis according to claim 1, characterized in that: The method uses natural language processing technology to perform sentiment analysis and keyword extraction on social media public opinion data, and monitors abnormal health event reports in real time and corrects risk prediction results.

5. The method for predicting the risk of infectious disease transmission based on big data analysis according to claim 1, characterized in that: The dynamic propagation model introduces a reinforcement learning mechanism, which includes: under the SEIR improved model framework, the basic reproduction number, incubation period duration, and asymptomatic infection ratio are defined as adjustable parameters, with initial values ​​determined by fitting historical epidemic data. The daily newly confirmed laboratory cases, nucleic acid test positivity rate, and close contact tracing results serve as real-time feedback signals. The direction of parameter adjustment is calculated through a policy gradient algorithm. At the same time, a reward function is designed to comprehensively consider prediction error, model stability, and public health costs. The model generates a set of parameter candidates every 24 hours, which is updated to the next cycle after being verified by disease control experts, forming a closed-loop optimization system of "data-driven - expert intervention".

6. A system for predicting the risk of infectious disease transmission based on big data analysis, employing the method for predicting the risk of infectious disease transmission based on big data analysis as described in claims 1-5, characterized in that, Including: Multi-source data acquisition module: used to connect to government open data platforms, IoT sensors and third-party data interfaces; Distributed computing engine: Real-time stream processing and batch analysis of massive amounts of data based on Spark; Risk prediction model library: integrates infectious disease dynamics models, graph neural networks, and time series prediction algorithms; Visual interactive terminal: Generates dynamic heat maps, risk trend curves, and prevention and control recommendations reports.

7. The infectious disease transmission risk prediction system based on big data analysis according to claim 6, characterized in that: The system also includes edge computing nodes, deployed in key areas, to support localized real-time risk calculation.

8. The infectious disease transmission risk prediction system based on big data analysis according to claim 7, characterized in that: The edge computing nodes are equipped with GPU acceleration modules to support local real-time computing. Each node synchronizes core parameters with the central cloud platform via a 5G private network or fiber optic cable but independently performs risk prediction tasks. The nodes have built-in lightweight infectious disease models and receive real-time data streams from local IoT devices. Combining the population density and building layout in the area stored at the edge, a micro-scale transmission risk index is generated every 5 minutes. If a risk threshold is detected to be exceeded, a local alarm is immediately triggered and an encrypted digest is uploaded to the central system. At the same time, a federated learning mechanism with neighboring nodes is initiated. The central cloud platform regularly pushes global model updates to the edge nodes.

9. The infectious disease transmission risk prediction system based on big data analysis according to claim 6, characterized in that: The visual interactive terminal supports multi-level permission management, provides decision support interfaces to government departments, and releases personalized early warning information to the public.

10. The infectious disease transmission risk prediction system based on big data analysis according to claim 6, characterized in that: The system is based on a consortium blockchain of Hyperledger Fabric. Medical institutions, disease control centers and research institutions participate in consensus as nodes. Smart contracts are used to control data access permissions. Sensitive data is encrypted with SM4 and then stored on the blockchain as hash values. The original data is distributed and stored through IPFS.

Images