Intelligent analysis and early warning system of CDC network data combined with epidemiological investigation

Abstract

The present application relates to the technical field of big data analysis, in particular to a control network data intelligent analysis and early warning system fusing epidemiological investigation. It comprises an epidemiological investigation logic driving module for algorithmizing epidemiological investigation methodology and generating a structured causal diagram representing transmission relationship; a causal diagram posteriori uncertainty driving module for calculating information gain expectation value and automatically generating optimal epidemiological investigation query; a counterfactual reasoning transmission evolution module for simulating epidemic evolution trajectory under different interventions; and an intervenable early warning index calculation module for calculating intervenable gain and outputting graded early warning signals and optimal prevention and control suggestions. The present application introduces counterfactual causal inference into disease control early warning, realizes paradigm transition from passive monitoring to active driving of epidemiological investigation, and significantly improves the decision support capability and intervention guidance value of early warning.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and more specifically, to a disease control network data intelligent analysis and early warning system that integrates epidemiological surveys. Background Technology

[0002] Currently, infectious disease surveillance and early warning systems generally employ multi-point triggering mechanisms, collecting data from nodes such as medical institutions, drug retailers, and schools, and issuing early warning signals through threshold comparison or statistical anomaly detection. In recent years, some systems have begun to integrate multi-source information such as social media, meteorological data, and population flow data, and have attempted to use machine learning models for prediction. However, the core analytical logic of existing systems remains at a superficial level of "data collection—correlation analysis—threshold triggering," failing to deeply integrate the inherent logic of epidemiological investigation as a professional methodology. Epidemiological investigation is an empirical science that determines transmission relationships and causal effects through methods such as time sequence, exposure tracing, bias control, and counterfactual reasoning. Its methodological value has not yet been effectively absorbed by data systems. Therefore, it is necessary to design a disease control network data intelligent analysis and early warning system that integrates epidemiological investigation.

[0003] The existing technology has the following technical defects, specifically:

[0004] Existing systems can only detect statistical correlations in data (such as an increase in the number of cases or spatial clustering) through threshold comparison or anomaly detection. They cannot determine the causal direction of transmission relationships, nor can they answer the counterfactual question that is crucial to prevention and control decisions: "If a certain transmission chain is cut off, how will the epidemic evolve?" This results in early warning results lacking explanatory power for transmission mechanisms and guiding value for intervention measures.

[0005] Existing technologies for fusing multi-source heterogeneous data remain at a superficial level, failing to provide a deep understanding of epidemiological semantics (incubation period constraints, contact relationships, exposure windows); at the same time, the system passively waits for data to be reported, unable to proactively identify data gaps (structural voids), and even less able to drive epidemiological investigators to collect high-value evidence in a targeted manner, resulting in a delay in the early warning stage.

[0006] Faced with common quality problems in epidemiological investigation data, such as recall bias, non-response bias, and selection bias, existing technologies use hard deletion or simple statistical imputation, which neither distinguishes the type of bias nor assesses its severity. A large amount of potentially valuable data is discarded or improperly corrected, and even directional misleading is introduced, which seriously affects the credibility of early warning analysis. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent analysis and early warning system for disease control network data that integrates epidemiological surveys, in order to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention aims to provide an intelligent analysis and early warning system for disease control network data that integrates epidemiological investigation, comprising: an epidemiological investigation logic-driven module, used to receive multi-source heterogeneous data from the disease control network, and based on algorithmic modeling of epidemiological investigation methodology, to transform the implicit transmission relationships in the data into a computable counterfactual causal reasoning framework, and generate a structured causal graph representing the potential transmission relationships between cases.

[0009] The posterior uncertainty driving module of the causal graph is coupled to the epidemiological investigation logic driving module. It is used to perform active reasoning based on the posterior uncertainty of the structured causal graph, calculate the expected information gain of candidate epidemiological investigation queries, select the query that maximizes the reduction of posterior uncertainty as the optimal epidemiological investigation query, and generate instructions to drive external epidemiological investigators to collect missing data in a targeted manner.

[0010] The counterfactual reasoning propagation evolution module is used to calculate the counterfactual epidemic evolution results under each set of intervention actions by simulating the application of one or more sets of hypothetical intervention actions based on the structured causal graph. The counterfactual epidemic evolution results include the case evolution trajectory under the assumption that no specified transmission event or specific prevention and control measures have occurred.

[0011] The interventionable early warning indicator calculation module, coupled with the counterfactual reasoning propagation evolution module, is used to calculate the interventionable gain based on the difference between the natural development evolution result and the counterfactual epidemic evolution result. The interventionable gain is used to quantify the expected effect of intervention actions. The interventionable early warning indicator calculation module is also used to output a graded early warning signal based on the comparison result of the interventionable gain and the dynamic threshold, and output the intervention strategy that maximizes the interventionable gain as an executable prevention and control suggestion.

[0012] As a further improvement to this technical solution, the epidemiological investigation logic driving module includes: a time-series causal discovery unit, used to infer the structured causal graph from time-series observation data collected from the disease control network based on a time-aware additive noise model. The structured causal graph includes multiple nodes and directed edges between nodes. Each node corresponds to a case or suspected case, and the direction of the directed edges represents the presumed transmission direction based on the epidemiological time sequence.

[0013] The time-aware additive noise model satisfies the following condition: for any two nodes and If node For nodes The potential source of infection, then the node Onset time satisfy:

[0014]

[0015] in, For random variables that follow an epidemiological incubation period distribution, To be independent of Additive noise, For nodes The onset time.

[0016] The temporal causality detection unit determines the result through a matching permutation test. and Are the residuals independent of If independent, then it is determined that there exists a source. arrive The causal side.

[0017] As a further improvement to this technical solution, the matching permutation test includes: constructing a permutation distribution and randomly shuffling the nodes. Multiple permutation samples were generated based on the chronological order of the onset of illness. The quantiles of the independence statistic of the original residuals in the permutation distribution were calculated. When the quantile was lower than a preset threshold, the null hypothesis of independence was rejected, thus confirming the sequence of onset. arrive The direction of its spread.

[0018] As a further improvement to this technical solution, the epidemiological investigation logic driving module also includes an epidemiological bias active identification and correction unit. The bias active identification and correction unit is used to: automatically identify the epidemiological bias type of each data item in the input dataset based on the metadata of the data source type, collection method, and survey subject characteristic dimension. The epidemiological bias type includes at least one or more of recall bias, non-response bias, selection bias, and information bias.

[0019] Based on the identified bias type and severity, a bias correction algorithm matching the bias type is automatically selected and applied to generate corrected epidemiological investigation data as the basis for constructing the structured cause-effect graph.

[0020] As a further improvement to this technical solution, the causal graph posterior uncertainty driving module includes:

[0021] The uncertainty quantification unit is used to calculate the existence probability and confidence interval of each candidate propagation edge based on the posterior distribution of the structured causal graph, and to identify subgraph regions in the posterior distribution where the uncertainty exceeds a preset threshold.

[0022] The information gain calculation unit is used to calculate the expected information gain value for candidate epidemiological queries, and the expected information gain value is determined based on the following formula:

[0023]

[0024] in For the set of all possible epidemiological investigation queries, To execute the query Evidence that may be obtained later Based on existing observational data, To spread the causal graph, This is the posterior entropy.

[0025] The graph distance decay acceleration unit is used to calculate information gain only on local subgraphs related to a candidate query, wherein the local subgraphs are composed of neighborhoods in the propagation causal graph whose distances from the nodes involved in the query do not exceed a preset number of hops, and the distances are weighted based on the reciprocal of the propagation probability.

[0026] As a further improvement to this technical solution, the counterfactual reasoning propagation evolution module encodes the structured causal graph based on a structural causal model and performs counterfactual reasoning through do-calculus. The infection time equation for each node in the structural causal model is expressed as:

[0027]

[0028] in For nodes in a structured cause-effect graph The infection time vector of all parent nodes, This is an exogenous noise variable.

[0029] As a further improvement to this technical solution, the counterfactual reasoning propagation evolution module includes: a cause-finding unit, used to update each exogenous noise variable based on the observed data. The posterior distribution.

[0030] Action units are used to apply do-operations to structured causal graphs, forcing the state or infection time of specified nodes to be set according to preset intervention strategies.

[0031] The prediction unit is used to obtain counterfactual epidemic evolution results through discrete event simulation based on the updated exogenous noise distribution and the causal graph after intervention. In the discrete event simulation, the probability of the event occurrence is determined by the causal mechanism in the structural causal equation.

[0032] As a further improvement to this technical solution, the formula for calculating the interventionable gain is as follows:

[0033]

[0034] in The dynamic threshold is the expected total number of cases or a preset public health indicator value within a preset time window; the dynamic threshold is dynamically calculated based on the quantiles of a historical interveneable gain sequence. When the interveneable gain exceeds the dynamic threshold, the interveneable early warning indicator calculation module outputs a high-level early warning signal.

[0035] As a further improvement to this technical solution, the epidemiological investigation logic driving module also includes an online causal structure update unit, which updates the posterior distribution of the structured causal graph based on Bayesian variational inference or Markov chain Monte Carlo method each time a new batch of disease control network data is received, and passes the updated posterior uncertainty to the causal graph posterior uncertainty driving module to form a recursive reinforcement loop for causal graph discovery and active epidemiological investigation query.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] This invention, through structural causal modeling and do-calculus, enables the system to simulate counterfactual scenarios of "how the epidemic will evolve if a certain intervention is applied," quantifies the intervened gains, and directly answers "what action is most effective and what are the expected benefits," thus giving the early warning results causal explanatory power and action guidance value, filling the technical gap in existing systems that cannot support intervention decisions.

[0038] This invention automatically identifies the most valuable evidence gaps for priority investigation by quantifying posterior uncertainty and maximizing information gain, and proactively generates optimal epidemiological investigation queries to drive targeted data collection by epidemiological investigators. The epidemiological investigation feedback data further updates the causal graph, forming a closed-loop recursive reinforcement loop of "reasoning-driving-updating". Compared with traditional passive threshold triggering systems, it can detect hidden transmission chains earlier and reduce the false positive rate.

[0039] This invention automatically matches targeted algorithms such as weighted likelihood, inverse probability weighting, and Heckman two-stage correction based on the type and severity of bias (recall bias, non-response bias, selection bias, and information bias) to perform personalized correction on each data point. This avoids information loss and misleading information caused by "one-size-fits-all" processing and significantly improves the usability and reliability of conclusions in epidemiological investigation data in early warning analysis. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a schematic diagram of the system structure connection of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Example: Please refer to Figure 1 As shown, a disease control network data intelligent analysis and early warning system integrating epidemiological investigation is provided, including: an epidemiological investigation logic-driven module, which is used to receive multi-source heterogeneous data from the disease control network, and based on algorithmic modeling of the epidemiological investigation methodology, transform the implicit transmission relationships in the data into a computable counterfactual causal reasoning framework, and generate a structured causal graph representing the potential transmission relationships between cases.

[0044] The algorithmic modeling of the epidemiological investigation methodology specifically refers to: transforming the causal inference logic followed by epidemiological investigators in their field work—including inferring the direction of transmission based on time sequence (the onset time of an infected person must be later than the exposure time of their source of infection), verifying the transmission chain using the constraints of the incubation period distribution (automatically excluding hypothesized transmission relationships if they violate the known incubation period distribution), reducing bias through cross-verification of multi-source information (simultaneously comparing multiple evidence dimensions such as symptom onset sequence, self-reported contact history, and laboratory test timeline), and actively tracing data gaps to eliminate uncertainty—into a computable formal framework. This framework uses a structural causal model as its underlying expression, modeling the infection time of each case as a function of the infection time of the potential source of infection, the random variable of the incubation period, and independent noise. It also introduces a matching permutation test to infer the existence and direction of causal edges from the observed data, thereby enabling the system to "simulate" the reasoning process of epidemiological investigation experts at the algorithmic level. This automatically compiles the original multi-source heterogeneous epidemiological investigation data into a structured causal graph representing the potential transmission relationships between individuals, providing a computational basis consistent with epidemiological logic for subsequent counterfactual simulations and proactive inquiries.

[0045] In one specific embodiment, the epidemiological investigation logic driving module includes: a time-series causal discovery unit, used to infer the structured causal graph from time-series observation data collected from the disease control network based on a time-aware additive noise model. The structured causal graph includes multiple nodes and directed edges between nodes. Each node corresponds to a case or suspected case, and the direction of the directed edges represents the presumed transmission direction based on the epidemiological time sequence.

[0046] The time-aware additive noise model satisfies the following condition: for any two nodes and If node For nodes The potential source of infection, then the node Onset time satisfy:

[0047]

[0048] in, For random variables that follow an epidemiological incubation period distribution, To be independent of Additive noise, For nodes The onset time.

[0049] The temporal causality detection unit determines the result through a matching permutation test. and Are the residuals independent of If independent, then it is determined that there exists a source. arrive The causal side.

[0050] In one specific embodiment, the matching permutation test includes: constructing a permutation distribution by randomly shuffling the nodes. Multiple permutation samples were generated based on the chronological order of the onset of illness. The quantiles of the independence statistic of the original residuals in the permutation distribution were calculated. When the quantile was lower than a preset threshold, the null hypothesis of independence was rejected, thus confirming the sequence of onset. arrive The direction of its spread.

[0051] In one specific embodiment, the epidemiological investigation logic driving module further includes an epidemiological bias active identification and correction unit, which is used to: automatically identify the epidemiological bias type of each data item in the input dataset based on the metadata of the data source type, collection method, and survey subject characteristic dimension. The epidemiological bias type includes at least one or more of recall bias, non-response bias, selection bias, and information bias.

[0052] Based on the identified bias type and severity, a bias correction algorithm matching the bias type is automatically selected and applied to generate corrected epidemiological investigation data as the basis for constructing the structured cause-effect graph.

[0053] The system automatically identifies and corrects epidemiological bias using the following steps: First, it extracts metadata feature vectors for each epidemiological survey record, including data source type (e.g., hospital reporting, telephone survey, on-site interview), collection method (face-to-face / remote, structured questionnaire / free text), demographic attributes of the survey participants (age, occupation, symptom severity), and response completeness indicators. Then, based on a pre-built bias classification decision tree or lightweight classifier (e.g., random forest), these metadata are mapped to preset bias types—recall bias is typically associated with long retrospective windows and unstructured responses, while no-response bias is associated with specific individuals. Low response rates in the cohort are associated with selection bias, non-random sampling patterns are associated with selection bias, and inconsistencies in measurement tools are associated with information bias. Simultaneously, the bias impact weight of each data item is quantified using a bias severity scoring function (e.g., recall bias severity is exponentially positively correlated with the number of days of retracement, and non-response bias severity is proportional to the product of the non-response rate and the difference in baseline risk among the population). Finally, based on the identified bias type, the corresponding correction algorithm library is invoked—recall bias is corrected using weighted likelihood or calibrated regression; non-response bias is corrected using inverse probability weighting or multiple imputation models (propensity score is calculated based on observable covariates); selection bias is corrected using Heckman two-stage correction or reweighting; and information bias is corrected using a measurement error model based on a validation subset. The correction strength (e.g., number of imputations, weight cutoff values) is dynamically adjusted based on the bias severity. All corrected data fields are output as a unified format epidemiological dataset, serving as the basic input for subsequent structured causal graph construction.

[0054] The posterior uncertainty driving module of the causal graph is coupled to the epidemiological investigation logic driving module. It is used to perform active reasoning based on the posterior uncertainty of the structured causal graph, calculate the expected information gain of candidate epidemiological investigation queries, select the query that maximizes the reduction of posterior uncertainty as the optimal epidemiological investigation query, and generate instructions to drive external epidemiological investigators to collect missing data in a targeted manner.

[0055] In one specific embodiment, the causal graph posterior uncertainty driving module includes:

[0056] The uncertainty quantification unit is used to calculate the existence probability and confidence interval of each candidate propagation edge based on the posterior distribution of the structured causal graph, and to identify subgraph regions in the posterior distribution where the uncertainty exceeds a preset threshold.

[0057] Suppose a structured cause-effect graph The posterior distribution is ,in For the observed data. For any candidate directed edge in the graph. (from node) Pointing to node The following quantitative indicators are defined:

[0058] edge existence probability

[0059]

[0060] in, As an indicator function, it is approximated by posterior sampling during actual calculation: , Let N be the sample graphs drawn from the posterior distribution.

[0061] Confidence interval for the probability of existence

[0062] Bayesian highest posterior density (HPD) intervals are used to cover confidence levels. ,(generally )

[0063]

[0064] in, and satisfy:

[0065]

[0066] The posterior edge density, representing the probability of edge existence (which can be obtained from sampling), (Empirical distribution estimation), in a simplified practical implementation, can be achieved by adjusting the sampling indicator variable. Calculate quantiles:

[0067]

[0068]

[0069] Uncertainty measure of edges

[0070] Uncertainty score for defining edges:

[0071]

[0072] Range of values ,when hour (Completely uncertain); when Or 1 o'clock 0 (Completely Determined). This metric is equivalent to It does not introduce any additional parameters, but only depends on .

[0073] Determination of high uncertainty edges

[0074] Preset uncertainty threshold (For example ),like Then the edge It is marked as a high uncertainty edge.

[0075] Identification of high uncertainty subgraph regions

[0076] Define a subgraph with high uncertainty :

[0077]

[0078] Edge set:

[0079] Node set:

[0080] The information gain calculation unit is used to calculate the expected information gain value for candidate epidemiological queries, and the expected information gain value is determined based on the following formula:

[0081]

[0082] in For the set of all possible epidemiological investigation queries, To execute the query Evidence that may be obtained later Based on existing observational data, To spread the causal graph, This is the posterior entropy.

[0083] The graph distance decay acceleration unit is used to calculate information gain only on local subgraphs related to a candidate query, wherein the local subgraphs are composed of neighborhoods in the propagation causal graph whose distances from the nodes involved in the query do not exceed a preset number of hops, and the distances are weighted based on the reciprocal of the propagation probability.

[0084] The counterfactual reasoning propagation evolution module is used to calculate the counterfactual epidemic evolution results under each set of intervention actions by simulating the application of one or more sets of hypothetical intervention actions based on the structured causal graph. The counterfactual epidemic evolution results include the case evolution trajectory under the assumption that no specified transmission event or specific prevention and control measures have occurred.

[0085] In one specific embodiment, the counterfactual reasoning propagation evolution module encodes the structured causal graph based on a structural causal model and performs counterfactual reasoning through do-calculus. The infection time equation for each node in the structural causal model is expressed as:

[0086]

[0087] in For nodes in a structured cause-effect graph The infection time vector of all parent nodes, This is an exogenous noise variable.

[0088] In one specific embodiment, the counterfactual reasoning propagation evolution module includes: abductive unit, used to update each exogenous noise variable based on observation data. The posterior distribution.

[0089] Action units are used to apply do-operations to structured causal graphs, forcing the state or infection time of specified nodes to be set according to preset intervention strategies.

[0090] The prediction unit is used to obtain counterfactual epidemic evolution results through discrete event simulation based on the updated exogenous noise distribution and the causal graph after intervention. In the discrete event simulation, the probability of the event occurrence is determined by the causal mechanism in the structural causal equation.

[0091] In one specific embodiment, the formula for calculating the interventionable gain is:

[0092]

[0093] in The dynamic threshold is the expected total number of cases or a preset public health indicator value within a preset time window; the dynamic threshold is dynamically calculated based on the quantiles of a historical interveneable gain sequence. When the interveneable gain exceeds the dynamic threshold, the interveneable early warning indicator calculation module outputs a high-level early warning signal.

[0094] The interventionable early warning indicator calculation module, coupled with the counterfactual reasoning propagation evolution module, is used to calculate the interventionable gain based on the difference between the natural development evolution result and the counterfactual epidemic evolution result. The interventionable gain is used to quantify the expected effect of intervention actions. The interventionable early warning indicator calculation module is also used to output a graded early warning signal based on the comparison result of the interventionable gain and the dynamic threshold, and output the intervention strategy that maximizes the interventionable gain as an executable prevention and control suggestion.

[0095] In one specific embodiment, the epidemiological investigation logic driving module further includes an online causal structure update unit, which updates the posterior distribution of the structured causal graph based on Bayesian variational inference or Markov chain Monte Carlo method each time a new batch of disease control network data is received, and passes the updated posterior uncertainty to the causal graph posterior uncertainty driving module to form a recursive reinforcement loop for causal graph discovery and active epidemiological investigation query.

[0096] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A system for intelligent analysis and early warning of CDC network data integrated with epidemiological investigation, characterized in that, include: The epidemiological investigation logic-driven module is used to receive multi-source heterogeneous data from the disease control network, and based on the algorithmic modeling of the epidemiological investigation methodology, it transforms the implicit transmission relationships in the data into a computable counterfactual causal reasoning framework, generating a structured causal graph that represents the potential transmission relationships between cases. The posterior uncertainty driving module of the causal graph is coupled to the epidemiological investigation logic driving module. It is used to perform active reasoning based on the posterior uncertainty of the structured causal graph, calculate the expected information gain of candidate epidemiological investigation queries, select the query that maximizes the reduction of posterior uncertainty as the optimal epidemiological investigation query, and generate instructions to drive external epidemiological investigators to collect missing data in a targeted manner. The counterfactual reasoning propagation evolution module is used to calculate the counterfactual epidemic evolution results under each set of intervention actions by simulating the application of one or more sets of hypothetical intervention actions based on the structured causal graph. The counterfactual epidemic evolution results include the case evolution trajectory under the assumption that no specified transmission event or specific prevention and control measures have occurred. The interventionable early warning indicator calculation module, coupled with the counterfactual reasoning propagation evolution module, is used to calculate the interventionable gain based on the difference between the natural development evolution result and the counterfactual epidemic evolution result. The interventionable gain is used to quantify the expected effect of intervention actions. The interventionable early warning indicator calculation module is also used to output a graded early warning signal based on the comparison result of the interventionable gain and the dynamic threshold, and output the intervention strategy that maximizes the interventionable gain as an executable prevention and control suggestion.

2. The intelligent analysis and early warning system for disease control network data integrating epidemiological surveys as described in claim 1, characterized in that: The epidemiological investigation logic driving module includes: The temporal causal discovery unit is used to infer the structured causal graph from temporal observation data collected from the disease control network based on a time-aware additive noise model. The structured causal graph includes multiple nodes and directed edges between nodes. Each node corresponds to a case or suspected case, and the direction of the directed edges indicates the presumed transmission direction based on the epidemiological time sequence. The time-aware additive noise model satisfies the following condition: for any two nodes and If node For nodes The potential source of infection, then the node Onset time satisfy: in, For random variables that follow an epidemiological incubation period distribution, To be independent of Additive noise, For nodes The onset time; The temporal causality detection unit determines the result through a matching permutation test. and Are the residuals independent of If independent, then it is determined that there exists a source. arrive The causal side.

3. The intelligent analysis and early warning system for disease control network data integrating epidemiological surveys as described in claim 2, characterized in that: The matching permutation test includes: constructing a permutation distribution and randomly shuffling the nodes. Multiple permutation samples were generated based on the chronological order of the onset of illness. The quantiles of the independence statistic of the original residuals in the permutation distribution were calculated. When the quantile was lower than a preset threshold, the null hypothesis of independence was rejected, thus confirming the sequence of onset. arrive The direction of its spread.

4. The intelligent analysis and early warning system for disease control network data integrating epidemiological surveys as described in claim 1, characterized in that: The epidemiological investigation logic driving module also includes an epidemiological bias active identification and correction unit. The bias active identification and correction unit is used to: automatically identify the epidemiological bias type of each data item in the input dataset based on the metadata of the data source type, collection method, and survey object characteristic dimension. The epidemiological bias type includes at least one or more of recall bias, non-response bias, selection bias, and information bias. Based on the identified bias type and severity, a bias correction algorithm matching the bias type is automatically selected and applied to generate corrected epidemiological investigation data as the basis for constructing the structured cause-effect graph.

5. The intelligent analysis and early warning system for disease control network data integrating epidemiological surveys as described in claim 1, characterized in that: The causal graph posterior uncertainty driving module includes: An uncertainty quantification unit is used to calculate the existence probability and confidence interval of each candidate propagation edge based on the posterior distribution of the structured causal graph, and to identify subgraph regions in the posterior distribution where the uncertainty exceeds a preset threshold. The information gain calculation unit is used to calculate the expected information gain value for candidate epidemiological queries, and the expected information gain value is determined based on the following formula: in For the set of all possible epidemiological investigation queries, To execute the query Evidence that may be obtained later Based on existing observational data, To spread the causal graph, For posterior entropy; The graph distance decay acceleration unit is used to calculate information gain only on local subgraphs related to a candidate query, wherein the local subgraphs are composed of neighborhoods in the propagation causal graph whose distances from the nodes involved in the query do not exceed a preset number of hops, and the distances are weighted based on the reciprocal of the propagation probability.

6. The intelligent analysis and early warning system for disease control network data integrating epidemiological surveys as described in claim 1, characterized in that: The counterfactual reasoning propagation and evolution module encodes the structured causal graph based on a structural causal model and performs counterfactual reasoning through do-calculus. The infection time equation for each node in the structural causal model is expressed as: in For nodes in a structured cause-effect graph The infection time vector of all parent nodes, This is an exogenous noise variable.

7. The intelligent analysis and early warning system for disease control network data integrating epidemiological surveys as described in claim 1, characterized in that: The counterfactual reasoning propagation and evolution module includes: The attribution unit is used to update each exogenous noise variable based on the observed data. The posterior distribution of; Action unit, used to apply do-operations to structured causal graphs, forcing the state or infection time of specified nodes to be set according to preset intervention strategies; The prediction unit is used to obtain counterfactual epidemic evolution results through discrete event simulation based on the updated exogenous noise distribution and the causal graph after intervention. In the discrete event simulation, the probability of the event occurrence is determined by the causal mechanism in the structural causal equation.

8. The intelligent analysis and early warning system for disease control network data integrating epidemiological surveys as described in claim 1, characterized in that: The formula for calculating the interventionable gain is as follows: in The dynamic threshold is the expected total number of cases or a preset public health indicator value within a preset time window; the dynamic threshold is dynamically calculated based on the quantiles of a historical interveneable gain sequence. When the interveneable gain exceeds the dynamic threshold, the interveneable early warning indicator calculation module outputs a high-level early warning signal.

9. The intelligent analysis and early warning system for disease control network data integrating epidemiological surveys as described in claim 1, characterized in that: The epidemiological investigation logic driving module also includes an online causal structure update unit, which updates the posterior distribution of the structured causal graph based on Bayesian variational inference or Markov chain Monte Carlo method each time a new batch of disease control network data is received, and passes the updated posterior uncertainty to the causal graph posterior uncertainty driving module to form a recursive reinforcement loop for causal graph discovery and active epidemiological investigation query.

Images