Epidemic situation emergency response decision-making method and system, computer equipment and storage medium

Abstract

The invention discloses an epidemic situation emergency response decision-making method and system, computer equipment and a storage medium. The epidemic situation emergency response decision-making method comprises the steps of fitting parameters of an infectious disease kinetic model according to an L-BFGS-B optimization algorithm to generate a Hessian matrix; carrying out ill-conditioned data detection on the Hessian matrix to remove abnormal data; and sampling the Hessian matrix after the ill-conditioned data is removed according to an adaptive MCMC algorithm, and inputting the sampled parameters in the Hessian matrix into an infectious disease dynamic model to generate a prediction curve bundle. The method and the device are used for quickly outputting the epidemic emergency response decision.

Description

Technical Field

[0001] This application relates to the technical field of large model data analysis, and in particular to decision-making methods, systems, computer equipment, and storage media for epidemic emergency response. Background Technology

[0002] Infectious disease dynamics models (such as SIR, SIRS, and SEIR models) are widely used for epidemic prediction. Uncertainty quantification of parameters is necessary in infectious disease dynamics models. Existing techniques for parameter uncertainty quantification include the following methods: (1) The complete Bayesian MCMC method, based on the Fisher information matrix theory, uses the Markov chain Monte Carlo method to infer the posterior distribution of parameters; (2) Approximate Bayesian computation method, using simulation experiments and cross-validation; (3) Neural network acceleration method, using deep learning to accelerate parameter inference.

[0003] However, existing technologies have several problems. The first problem is the excessive computation time, which cannot meet the timeliness requirements of emergency response. A single analysis in the complete Bayesian MCMC method typically takes 3-10 minutes; comparing 10 alternatives takes 30-100 minutes. Combined with data processing time, this makes it impossible to complete the evaluation of decision-making options within 2 hours. Approximate Bayesian computation takes 5-15 minutes per analysis and requires cross-validation and multiple iterations, also exceeding the time limit. While neural network methods offer fast inference, they require several hours of retraining for each new epidemic data point, lacking versatility and unsuitable for emergency response to sudden outbreaks. The second problem is the need for specialized statistical knowledge, creating a high barrier to entry for grassroots staff, who typically lack the necessary background. The third problem is the high computational resource requirements; grassroots hardware is often too basic to provide the necessary resources, and cloud computing services pose data security risks. The fourth problem is the lack of engineered automated processing mechanisms. Encountering singular or excessively large data sets easily leads to errors and exits, making it impossible for non-professional users to use independently, resulting in a disjointed data-to-decision process. The fifth problem is the lack of a complete system for decision-making. Parameter estimation, uncertainty quantification, and intervention assessment are all independent processes, requiring researchers to manually combine each step, making it difficult to automate the generation of risk assessments and decision recommendations. Summary of the Invention

[0004] To overcome one or more technical problems existing in the prior art, this application provides an epidemic emergency response decision-making method, system, computer equipment, and storage medium, which can realize the rapid output of epidemic emergency response decisions.

[0005] The technical solution of this application is as follows.

[0006] In the first aspect, this application provides a decision-making method for emergency response to an epidemic, including the following steps: The Hessian matrix is ​​generated by fitting the parameters of the infectious disease dynamics model using the L-BFGS-B optimization algorithm. The Hessian matrix is ​​subjected to ill-conditioned data detection to remove outliers; The Hessian matrix after removing pathological data is sampled using an adaptive MCMC algorithm, and the sampled parameters in the Hessian matrix are input into the infectious disease dynamics model to generate prediction curve bundles.

[0007] The first preferred embodiment of the first aspect further includes the step of: A risk assessment is performed on the predicted curve bundle to obtain decision recommendations.

[0008] In a first preferred embodiment of the first aspect, more preferably, the step of: Under constraints, the decision recommendations are optimized according to the control objectives to obtain the final decision scheme.

[0009] The second preferred embodiment of the first aspect is to use one of the infectious disease dynamics models SI model, SIR model, SIRS model, and SEIR model.

[0010] The third preferred embodiment of the first aspect is that the step of detecting and removing abnormal data from pathological data includes: Identify ill-conditioned matrices by detecting the condition number of the Hessian matrix; Singular value decomposition is performed on the ill-conditioned matrix to identify and remove collinear parameter pairs; The Hessian matrix is ​​constructed using an adaptive regularization coefficient algorithm. The parametric covariance matrix is ​​obtained by calculating the inverse of the Hessian matrix. Extract the standard error of the parameters from the diagonal elements of the parameter covariance matrix, calculate the confidence interval of the parameters of the Hessian matrix based on the standard error of the parameters, and delete the parameters in the Hessian matrix that fall outside the confidence interval of the parameters.

[0011] A preferred embodiment of the third preferred implementation is that sampling according to the adaptive MCMC algorithm includes: The number of samples for each type of parameter in the Hessian matrix after removing ill-conditioned data is dynamically determined based on the magnitude of the parameter standard error. Based on the determined number of samples and the principle of normal approximation, a multivariate normal distribution algorithm is used to randomly sample the Hessian matrix after removing ill-conditioned data.

[0012] A fourth preferred embodiment of the first aspect is that, when generating the prediction curve bundle, the quantile of each prediction curve output by the infectious disease dynamics model is calculated, and the prediction curves whose quantiles fall within the 95% confidence interval are combined into a prediction curve bundle.

[0013] Secondly, this application provides an epidemic emergency response decision system, including an infectious disease dynamics module, a Hessian matrix generation module, an anomaly handling module, and an adaptive MCMC algorithm sampling module; The Hessian matrix generation module is used to generate a Hessian matrix based on the parameters of the infectious disease dynamics model fitted by the L-BFGS-B optimization algorithm. The anomaly handling module is used to detect and remove abnormal data from the Hessian matrix. The adaptive MCMC algorithm sampling module is used to sample the Hessian matrix after removing ill-conditioned data according to the adaptive MCMC algorithm; The infectious disease dynamics module is used to input the sampled parameters from the Hessian matrix into the infectious disease dynamics model to generate a prediction curve bundle.

[0014] Thirdly, this application provides a computer device, including a memory and a processor; the memory is provided with a computer program; when the computer program is executed by the processor, it is used to implement the epidemic emergency response decision-making method described in the first aspect.

[0015] Fourthly, this application provides a storage medium storing computer instructions for execution by a processor to implement the epidemic emergency response decision-making method described in the first aspect.

[0016] Compared to existing technologies, the advantages of this application are as follows: Compared to traditional methods, it has the advantage of zero additional computational cost and short computation time, which can meet the timeliness requirements of emergency response. By performing singularity detection, it avoids the problem that traditional methods are prone to errors and exit when encountering singular or excessively large data conditions. It can also be used by non-professional users, with low requirements for computing resources. It can be used by grassroots epidemic emergency response units with ordinary hardware conditions. When facing decision-making, parameter estimation, uncertainty quantification, and intervention assessment work together to automatically generate risk assessment and decision-making plans. Attached Figure Description

[0017] Figure 1 This is a flowchart of one of the emergency response decision-making methods for an epidemic in this application.

[0018] Figure 2 For implementation Figure 1 A structural diagram of China's COVID-19 emergency response decision-making system. Detailed Implementation

[0019] Referring to the illustrations, the principles of this application are illustrated by way of example implementation in a suitable operating environment. The following description is based on the specific embodiments of this application as illustrated, and should not be construed as limiting other specific embodiments not detailed herein.

[0020] As used herein, the terms “component,” “module,” “system,” “interface,” “process,” etc., generally refer to computer-related entities: hardware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable application, an executing thread, a program, and / or a computer. As illustrated, both an application running on a controller and the controller itself can be components. One or more components may reside within an executing process and / or thread, and components may be located on a single computer and / or distributed across two or more computers.

[0021] Furthermore, the claimed subject matter can be implemented as a method, apparatus, or article of manufacture that uses standard programming and / or engineering techniques to generate software, firmware, hardware, or any combination thereof to control a computer to implement the disclosed subject matter.

[0022] As used herein, the term "preferred" is meant as an example, illustration, or illustration. Any aspect or design described herein as "preferred" need not be construed as being more advantageous than other aspects or designs. Rather, the use of the term "preferred" is intended to present the concept in a specific manner. As used in this application, the term "or" is intended to mean an inclusive "or" rather than an exclusionary "or." That is, unless otherwise specified or clear from the context, "X uses A or B" naturally includes either of the permutations. That is, if X uses A; X uses B; or X uses both A and B, then "X uses A or B" is satisfied in any of the foregoing examples.

[0023] Furthermore, although specific features of this disclosure have been disclosed with respect to only one of several implementations, such features may be combined with one or more other features of other implementations that may be desirable and advantageous for a given or particular application. Moreover, with regard to the use of the terms “comprising,” “having,” “containing,” or variations thereof in the Detailed Description or claims, such terms are intended to be included in a manner similar to the term “comprising.”

[0024] The first embodiment of this application is described below.

[0025] Please refer to Figure 1The epidemic emergency response decision-making method proposed in this application adopts one of the following infectious disease dynamics models: SI model, SIR model, SIRS model, and SEIR model, and includes the following steps.

[0026] S1. Generate the Hessian matrix by fitting the parameters of the infectious disease dynamics model using the L-BFGS-B optimization algorithm.

[0027] In this embodiment, the specific process of step S1 includes: S11. Obtain the epidemic observation data required for the corresponding infectious disease dynamics model as parameters; S12. Call the L-BFGS-B optimization algorithm to fit the parameters. The result of the L-BFGS-B optimization algorithm is the corresponding Hessian matrix.

[0028] S2. Perform pathological data detection and remove abnormal data from the generated Hessian matrix.

[0029] In this embodiment, the specific process of step S2 includes: S21. Detect the condition number of the Hessian matrix to identify ill-conditioned matrices. A matrix with ill-conditioned properties is called an ill-conditioned matrix. Ill-conditioned properties are usually related to the condition number of the matrix. The condition number is the ratio of the maximum singular value to the minimum singular value of the matrix. If the condition number of a matrix is ​​very large, then the matrix is ​​considered ill-conditioned. In this step S21, the corresponding parameters are judged to be ill-conditioned data based on the set threshold of the condition number. The set threshold can be adaptively set based on the amount of observed epidemic data and the number of parameters of the infectious disease dynamics model to be used. S22. Perform singular value decomposition (SVD) on the ill-conditioned matrix to identify and remove collinear parameter pairs; S23. Construct a regularized Hessian matrix according to the adaptive regularization coefficient algorithm; In this step S23, the automatic estimation method of the regularization coefficient can be designed based on the minimum singular value. S24. Calculate the inverse of the Hessian matrix to obtain the parametric covariance matrix; S25. Extract the standard error of the parameters from the diagonal elements of the parameter covariance matrix, calculate the confidence interval of the parameters in the Hessian matrix based on the standard error of the parameters, and delete the parameters in the Hessian matrix that fall outside the confidence interval.

[0030] S3. Sample the Hessian matrix after removing pathological data using the adaptive MCMC algorithm, and input the sampled parameters from the Hessian matrix into the infectious disease dynamics model to generate prediction curve bundles. The adaptive MCMC algorithm is a Bayesian optimization method, also known as the Adaptive-MCMC algorithm.

[0031] In this embodiment, step S3, sampling according to the adaptive MCMC algorithm, includes: dynamically determining the sampling quantity of each type of parameter in the Hessian matrix after removing ill-conditioned data based on the magnitude of the parameter standard error; and randomly sampling the Hessian matrix after removing ill-conditioned data using a multivariate normal distribution algorithm based on the determined sampling quantity and the normal approximation principle. The method for determining the sampling quantity can be designed based on the hierarchical criteria of parameter relative uncertainty, a three-level sampling quantity strategy, and the idea of ​​dynamically adjusting based on the target confidence interval width.

[0032] In this embodiment, when generating the prediction curve bundle in step S3, the quantile of each prediction curve output by the infectious disease dynamics model is calculated, and the prediction curves whose quantiles fall within the 95% confidence interval are combined into a prediction curve bundle.

[0033] S4. Conduct a risk assessment on the generated prediction curve bundle to obtain decision recommendations.

[0034] In step S4, the risk level of the prediction curve bundle is automatically assessed based on the width of the set effect confidence interval, realizing automation from data analysis to decision recommendations without the need for professional statistical knowledge; the risk level can be calculated based on the relative uncertainty formula and three levels of risk classification criteria can be set.

[0035] S5. Under constraints, the decision recommendations are optimized according to the control objectives to obtain the final decision scheme.

[0036] In step S5, the control objective is mathematically modeled by setting a constraint condition with a 95% confidence interval lower bound. The control objective can be a weighted combination of cost function and effect function, or a combination of grid search and heuristic optimization algorithm. The optimal control objective can also be a robustness score.

[0037] Compared to existing technologies, the emergency response decision-making method for epidemics proposed in this application has the following advantages: The Hessian matrix is ​​a computational tool inherent in the L-BFGS-B optimization algorithm. Compared to the traditional Bayesian MCMC, it has the advantage of zero additional computational cost and short computation time, which can meet the timeliness requirements of emergency response. Singularity detection on the Hessian matrix avoids the problem of traditional methods easily crashing and exiting when encountering singular data or data with excessively large condition numbers. It can also be used by non-professional users, with low requirements for computing resources, and can be used by grassroots epidemic emergency response units with ordinary hardware conditions. When facing decision-making, parameter estimation, uncertainty quantification, and intervention assessment work together to automatically generate risk assessments and decision-making plans.

[0038] The second embodiment of this application is described below.

[0039] Please refer to Figure 2 The epidemic emergency response decision-making system of this application includes an infectious disease dynamics module, a Hessian matrix generation module, an anomaly handling module, an adaptive MCMC algorithm sampling module, a risk assessment module, and a decision optimization module.

[0040] The Hessian matrix generation module is used to generate a Hessian matrix based on the parameters of an infectious disease dynamics model fitted using the L-BFGS-B optimization algorithm. The specific processing steps of the Hessian matrix generation module are as follows: Obtain the epidemic observation data required for the corresponding infectious disease dynamics model as parameters; call the L-BFGS-B optimization algorithm to fit the parameters; and output the corresponding Hessian matrix as the result of the L-BFGS-B optimization algorithm.

[0041] The anomaly handling module is used to detect and remove outliers from the Hessian matrix. The specific processing steps are as follows: The condition number of the Hessian matrix is ​​detected to identify ill-conditioned matrices; singular value decomposition is performed on the ill-conditioned matrix to identify and remove collinear parameter pairs; a regularized Hessian matrix is ​​constructed using an adaptive regularization coefficient algorithm; the inverse of the Hessian matrix is ​​calculated to obtain the parameter covariance matrix; the parameter standard error is extracted from the diagonal elements of the parameter covariance matrix; the parameter confidence interval of the Hessian matrix is ​​calculated; and parameters in the Hessian matrix that fall outside the parameter confidence interval are deleted.

[0042] The adaptive MCMC algorithm sampling module is used to sample the Hessian matrix after removing ill-conditioned data according to the adaptive MCMC algorithm. The processing of the adaptive MCMC algorithm sampling module includes: dynamically determining the number of samples for each type of parameter in the Hessian matrix after removing ill-conditioned data based on the magnitude of the parameter standard error; and randomly sampling the Hessian matrix after removing ill-conditioned data using a multivariate normal distribution algorithm based on the determined number of samples and the normal approximation principle.

[0043] The Infectious Disease Dynamics module is used to input the sampled parameters from the Hessian matrix into the infectious disease dynamics model to generate prediction curve bundles. The infectious disease dynamics model used in the module is one of the SI model, SIR model, SIRS model, or SEIR model.

[0044] The risk assessment module is used to assess the risks of the generated prediction curves and obtain decision recommendations. The module's processing methods include: automatically assessing the risk level of the prediction curves based on the width of a set effect confidence interval, thus automating the process from data analysis to decision recommendations. The risk level can be calculated using a relative uncertainty formula and a three-level risk classification standard can be set.

[0045] The decision optimization module is used to optimize decision suggestions based on control objectives under constraints to obtain the final decision scheme. The processing methods of the decision optimization module include: setting constraints with a 95% confidence interval lower bound as the control objective, mathematically modeling the control objective, and specifically employing a weighted combination of cost and effect functions, or a combination of grid search and heuristic optimization algorithms. The optimal solution's control objective can also be a robustness score.

[0046] Compared to existing technologies, the emergency response decision-making system for the epidemic in this embodiment has the following advantages: The Hessian matrix is ​​a computational tool inherent in the L-BFGS-B optimization algorithm. Compared to the traditional Bayesian MCMC, it has the advantage of zero additional computational cost and short computation time, which can meet the timeliness requirements of emergency response. Singularity detection on the Hessian matrix avoids the problem of traditional methods easily crashing and exiting when encountering singular data or data with excessively large condition numbers. It can also be used by non-professional users, with low requirements for computing resources, and can be used by grassroots epidemic emergency response units with ordinary hardware conditions. When facing decision-making, parameter estimation, uncertainty quantification, and intervention assessment work together to automatically generate risk assessments and decision-making plans.

[0047] The third embodiment of this application is described below.

[0048] The computer device of this application includes a memory and a processor; the memory contains a computer program; when the computer program is executed by the processor, it is used to implement the epidemic emergency response decision-making method of the first embodiment.

[0049] The fourth embodiment of this application is described below.

[0050] The storage medium of this application stores computer instructions that are executed by a processor to implement the epidemic emergency response decision-making method of the first embodiment.

[0051] Although the present application has disclosed the preferred embodiments above, the above preferred embodiments are not intended to limit the present application. Those skilled in the art can make various modifications and alterations without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be determined by the scope defined in the claims.

Claims

1. A decision-making method for emergency response to an epidemic, characterized in that, Including the following steps: The Hessian matrix is ​​generated by fitting the parameters of the infectious disease dynamics model using the L-BFGS-B optimization algorithm. The Hessian matrix is ​​subjected to ill-conditioned data detection to remove outliers; The Hessian matrix after removing pathological data is sampled using an adaptive MCMC algorithm, and the sampled parameters in the Hessian matrix are input into the infectious disease dynamics model to generate prediction curve bundles.

2. The emergency response decision-making method for an epidemic as described in claim 1, characterized in that, It also includes the following steps: A risk assessment is performed on the predicted curve bundle to obtain decision recommendations.

3. The emergency response decision-making method for an epidemic as described in claim 2, characterized in that, It also includes the following steps: Under constraints, the decision recommendations are optimized according to the control objectives to obtain the final decision scheme.

4. The emergency response decision-making method for an epidemic as described in claim 1, characterized in that, One of the following infectious disease dynamics models is used: SI model, SIR model, SIRS model, or SEIR model.

5. The emergency response decision-making method for an epidemic as described in claim 1, characterized in that, The steps for detecting and removing outliers from pathological data include: Identify ill-conditioned matrices by detecting the condition number of the Hessian matrix; Singular value decomposition is performed on the ill-conditioned matrix to identify and remove collinear parameter pairs; The Hessian matrix is ​​constructed using an adaptive regularization coefficient algorithm. The parametric covariance matrix is ​​obtained by calculating the inverse of the Hessian matrix. Extract the standard error of the parameters from the diagonal elements of the parameter covariance matrix, calculate the confidence interval of the parameters of the Hessian matrix based on the standard error of the parameters, and delete the parameters in the Hessian matrix that fall outside the confidence interval of the parameters.

6. The emergency response decision-making method for an epidemic as described in claim 5, characterized in that, Sampling based on the adaptive MCMC algorithm includes: The number of samples for each type of parameter in the Hessian matrix after removing ill-conditioned data is dynamically determined based on the magnitude of the parameter standard error. Based on the determined number of samples and the principle of normal approximation, a multivariate normal distribution algorithm is used to randomly sample the Hessian matrix after removing ill-conditioned data.

7. The emergency response decision-making method for an epidemic as described in claim 1, characterized in that, When generating the prediction curve bundle, the quantile of each prediction curve output by the infectious disease dynamics model is calculated, and the prediction curves whose quantiles fall within the 95% confidence interval are combined into a prediction curve bundle.

8. An emergency response decision-making system for an epidemic, characterized in that, It includes an infectious disease dynamics module, a Hessian matrix generation module, an anomaly handling module, and an adaptive MCMC algorithm sampling module; The Hessian matrix generation module is used to generate a Hessian matrix based on the parameters of the infectious disease dynamics model fitted by the L-BFGS-B optimization algorithm. The anomaly handling module is used to detect and remove abnormal data from the Hessian matrix. The adaptive MCMC algorithm sampling module is used to sample the Hessian matrix after removing ill-conditioned data according to the adaptive MCMC algorithm; The infectious disease dynamics module is used to input the sampled parameters from the Hessian matrix into the infectious disease dynamics model to generate a prediction curve bundle.

9. A computer device, characterized in that, It includes a memory and a processor; the memory is equipped with a computer program; when the computer program is executed by the processor, it is used to implement the epidemic emergency response decision-making method according to any one of claims 1-7.

10. A storage medium, characterized in that, The storage medium stores computer instructions that are executed by a processor to implement the emergency response decision-making method for an epidemic as described in any one of claims 1-7.

Images