Prediction method and system for epidemic condition of respiratory infectious disease
By combining the prediction methods of case data and keyword search data, the problem of low accuracy caused by traditional models relying on historical case data is solved, and a more accurate prediction of the epidemic situation of respiratory infectious diseases is achieved.
Patent Information
- Application Number
- CN202411358797.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2025-03-28
Smart Images
Figure CN119724609A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of respiratory infectious disease data analysis and processing, and in particular to a method and system for predicting the epidemic status of respiratory infectious diseases. Background Art
[0002] Research and early intervention of respiratory infectious diseases are of great significance for ensuring public health security, reducing the disease burden, improving disease prevention and control and management capabilities, promoting vaccine and drug research and development, and responding to new infectious diseases that may emerge in the future. As such, many studies have been committed to proposing innovative mathematical models to more accurately predict disease transmission, recovery and death in various countries, so that decision makers can formulate prevention and control measures more scientifically and rationally.
[0003] In the mathematical modeling of infectious diseases, compartmental models are a commonly used modeling technique. The SIR (Susceptible-Infected-Recovered) model, as the most classic compartmental model, has been widely used to predict transmission since it was proposed. The SIR model divides the population into three categories: susceptible, infected, and recovered. The model uses differential equations to describe the flow between these three groups of people, and then predicts the spread of the disease. Similarly, the SEIR (Susceptible-Exposed-Infected-Recovered) model adds the "exposed" category to the SIR model, which describes the state of the infected person who has come into contact with the pathogen before the onset of symptoms but has not yet developed the disease.
[0004] However, traditional infectious disease prediction models such as the SIR model and SEIR model mainly rely on historical case data to predict future disease conditions, which has certain limitations. Because the spread of the disease is affected by many factors, such as population mobility, community transmission, and climate change, these factors may not be fully reflected by case data. Therefore, relying solely on case data to predict the prevalence of respiratory infectious diseases may ignore some important influencing factors, thereby affecting the accuracy and reliability of the model, making the prediction results inaccurate and unreliable. Summary of the invention
[0005] The purpose of this application is to provide a method and system for predicting the epidemic status of respiratory infectious diseases, which can analyze and process respiratory infectious disease data and improve the accuracy of predicting the epidemic status of respiratory infectious diseases.
[0006] To achieve the above objectives, this application provides the following solutions:
[0007] In a first aspect, the present application provides a method for predicting the prevalence of respiratory infectious diseases, the method comprising:
[0008] Obtain case data and keyword search data of respiratory infectious diseases in the area to be tested; the case data includes the number of existing confirmed cases, the number of existing recovered people and the total number of people, the total number of people is the total permanent population of the area to be tested, the keyword search data includes the existing keyword search ratio, the existing non-searched keyword ratio and the total search volume ratio, the existing keyword search ratio refers to the ratio of the search volume of keywords related to respiratory infectious diseases to the total search volume in the same period, the existing non-searched keyword ratio refers to the ratio of the search volume of non-respiratory infectious disease keywords in the same period, the total search volume ratio refers to the ratio of the total search volume in the same period to the total search volume in previous years in the same period, keywords refer to words related to respiratory infectious diseases, and the total search volume refers to the sum of the search volumes of all words;
[0009] The case data is input into a case prediction model to obtain a case number prediction result; the case prediction model is a trained case SIR model or a case SEIR model, and the case number prediction result is the number of confirmed cases per day in a future time period predicted based on the case data;
[0010] The keyword search data is input into a keyword prediction model to obtain a keyword search volume prediction result; the keyword prediction model is a trained keyword SIR model or a keyword SEIR model, and the keyword search volume prediction result is a daily keyword search percentage predicted based on the keyword search data and in the same time period as the case number prediction result, and the daily keyword search percentage refers to the proportion of daily search volume for keywords related to respiratory infectious diseases to the search volume of related terms in previous years during the same time period;
[0011] The case number prediction result and the keyword search volume prediction result are input into the regression model to obtain the final case number prediction result; the regression model is a trained calculation model for determining the correlation between independent variables and dependent variables, the independent variables include the case prediction model, the keyword prediction model, the keyword search volume and the actual number of confirmed cases, the dependent variables include the number of confirmed cases per day in a future time period, and the final case number prediction result is the number of confirmed cases per day in a future time period predicted based on the case data and the keyword search data.
[0012] In a second aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for predicting the epidemic status of respiratory infectious diseases described in the first aspect.
[0013] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting the epidemic status of respiratory infectious diseases described in the first aspect.
[0014] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the method for predicting the epidemic status of respiratory infectious diseases described in the first aspect.
[0015] According to the specific embodiments provided in this application, this application discloses the following technical effects:
[0016] The present application provides a method and system for predicting the epidemic situation of respiratory infectious diseases, by obtaining case data and keyword search data of respiratory infectious diseases, and then inputting the case data and keyword search data into the case prediction model and the keyword prediction model respectively, so as to preliminarily obtain the case number prediction results and the keyword search volume prediction results; then the case number prediction results and the keyword search volume prediction results are used as the input of the regression model, and input into the regression model together to predict the final case number prediction results reflecting the correlation between the case data and the keyword search data. Therefore, the present application starts from the perspective of the correlation between the case number and the keyword search volume of respiratory infectious diseases, and analyzes the epidemic situation of respiratory infectious diseases by combining case data and keyword search data, thereby taking into account the two dimensions of case data and keyword search data, which can effectively improve the accuracy of predicting the epidemic situation of respiratory infectious diseases, no longer relying solely on case data, and effectively solves the problem that the current traditional infectious disease dynamics models such as the SIR model and the SEIR model are too dependent on historical case data, which have great limitations and lead to low model accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0018] Figure 1A flowchart of a method for predicting the prevalence of respiratory infectious diseases provided in one embodiment of the present application.
[0019] Figure 2 This is an overall experimental flow chart of a method for predicting the prevalence of respiratory infectious diseases provided in one embodiment of the present application.
[0020] Figure 3 A schematic diagram of a data set division method provided in an embodiment of the present application.
[0021] Figure 4 A schematic diagram of the fitting of the first case prediction SIR model training set provided in one embodiment of the present application.
[0022] Figure 5 This is a schematic diagram of the prediction of future influenza trends by the first case prediction SIR model provided in one embodiment of the present application.
[0023] Figure 6 A schematic diagram of the prediction of future keyword proportions by the first “fever” keyword SIR model provided in an embodiment of the present application.
[0024] Figure 7 A schematic diagram of the prediction of future keyword proportions by the first "pneumonia" keyword SIR model provided in an embodiment of the present application.
[0025] Figure 8 A schematic diagram of the prediction of future keyword proportions by the first “sum” keyword SIR model provided in an embodiment of the present application.
[0026] Fig. 9 A schematic diagram of the prediction of future keyword proportions by the first “body temperature” keyword SIR model provided in an embodiment of the present application.
[0027] Fig.10 A schematic diagram of the prediction of future keyword proportions by the first “antipyretic drug” keyword SIR model provided in one embodiment of the present application.
[0028] Fig.11 A schematic diagram of the prediction of future keyword proportions by the first “flu” keyword SIR model provided in an embodiment of the present application.
[0029] Fig.12 A schematic diagram of the prediction of future keyword proportions by the first “pharyngitis” keyword SIR model provided in an embodiment of the present application.
[0030] Fig.13 A schematic diagram of the fitting of the SEIR model training set for the first case prediction provided in one embodiment of the present application.
[0031] Fig.14 A schematic diagram of the prediction of future influenza trends by the first case prediction SEIR model provided in one embodiment of the present application.
[0032] Fig.15 A schematic diagram of the prediction of future keyword proportions by the first “fever” keyword SEIR model provided in an embodiment of the present application.
[0033] Fig.16 A schematic diagram of the prediction of future keyword proportions by the first SEIR model of the keyword "pneumonia" provided in one embodiment of the present application.
[0034] Fig.17 A schematic diagram of the prediction of future keyword proportions by the first “sum” keyword SEIR model provided in an embodiment of the present application.
[0035] Fig.18 A schematic diagram of the prediction of future keyword proportions by the first “body temperature” keyword SEIR model provided in an embodiment of the present application.
[0036] Fig.19 A schematic diagram of the prediction of future keyword proportions by the first “antipyretic drug” keyword SEIR model provided in an embodiment of the present application.
[0037] Fig. 20 A schematic diagram of the prediction of future keyword proportions by the first “flu” keyword SEIR model provided in an embodiment of the present application.
[0038] Fig.21 A schematic diagram of the prediction of future keyword proportions by the first “pharyngitis” keyword SEIR model provided in an embodiment of the present application.
[0039] Fig. 22 A schematic diagram of the prediction results of the regression model provided in one embodiment of the present application on a validation set.
[0040] Fig.23 A schematic diagram comparing the prediction results of the model of the present invention and the traditional SIR model on a test set provided in one embodiment of the present application.
[0041] Fig.24 A schematic diagram comparing the influenza trend prediction results of the model of the present invention and the traditional SIR model provided in one embodiment of the present application.
[0042] Fig.25 A schematic diagram comparing the prediction results of the model of the present invention and the traditional SEIR model on a test set provided in one embodiment of the present application.
[0043] Fig.26A schematic diagram comparing the influenza trend prediction results of the model of the present invention and the traditional SEIR model provided in one embodiment of the present application. DETAILED DESCRIPTION
[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0045] like Figure 1 As shown, this embodiment proposes a method for predicting the prevalence of respiratory infectious diseases, and the method for predicting the prevalence of respiratory infectious diseases specifically includes the following steps:
[0046] Step S1, obtaining case data and keyword search data of respiratory infectious diseases in the area to be tested. Among them, the case data includes the number of existing confirmed cases, the number of existing recovered people and the total number of people, etc. The total number of people is the total permanent population of the area to be tested, and the keyword search data includes the existing keyword search ratio, the existing unsearched keyword ratio and the total search volume ratio, etc. The existing keyword search ratio refers to the proportion of the search volume of keywords related to respiratory infectious diseases to the total search volume in the same period, the existing unsearched keyword ratio refers to the proportion of the search volume of non-respiratory infectious disease keywords in the same period, the total search volume ratio refers to the proportion of the total search volume in the same period to the total search volume in previous years in the same period, the keyword refers to the vocabulary related to respiratory infectious diseases, and the total search volume refers to the sum of the search volumes of all vocabulary.
[0047] Step S2: input the case data into the case prediction model to obtain a case number prediction result, wherein the case prediction model is a trained case SIR model or case SEIR model, and the case number prediction result is the number of confirmed cases per day in a future time period predicted based on the case data.
[0048] Step S3: input the keyword search data into the keyword prediction model to obtain the keyword search volume prediction result. The keyword prediction model is a trained keyword SIR model or keyword SEIR model, and the keyword search volume prediction result is the daily keyword search percentage predicted based on the keyword search data and in the same time period as the case number prediction result, and the daily keyword search percentage refers to the proportion of the daily search volume of keywords related to respiratory infectious diseases to the search volume of related terms in previous years during the same period.
[0049] Step S4, input the case number prediction result and the keyword search volume prediction result into the regression model to obtain the final case number prediction result. Among them, the regression model is a trained calculation model for determining the correlation between independent variables and dependent variables, and the independent variables include the case prediction model, the keyword prediction model, the keyword search volume and the actual number of confirmed cases, and the dependent variable includes the number of confirmed cases per day in the future time period. The focus of the regression model is to determine the correlation between the keyword search volume and the number of confirmed cases. The final case number prediction result is the number of confirmed cases per day in the future time period predicted based on the case data and the keyword search data.
[0050] In this embodiment, step S1 obtains case data and keyword search data of respiratory infectious diseases in the area to be tested, specifically including:
[0051] Step S11: Acquire case data using a web crawler.
[0052] Step S12: using a search engine, such as Baidu search engine, etc., to obtain keyword search data in units of days.
[0053] In this embodiment, the selection of keywords generally includes the following three categories:
[0054] 1) Names, aliases, common names, etc. of respiratory infectious diseases, such as influenza.
[0055] 2) Symptoms of respiratory infectious diseases, such as fever, pharyngitis, pneumonia, etc.
[0056] 3) Symptomatic drugs for respiratory infectious diseases, such as ibuprofen, antipyretics, etc.
[0057] In this embodiment, the respiratory infectious disease may be influenza or other respiratory infectious diseases. Taking influenza as an example, when the respiratory infectious disease to be analyzed is influenza, the keywords corresponding to influenza may include influenza, flu, fever, pharyngitis, pneumonia, body temperature, and antipyretic drugs.
[0058] With respect to the classic infectious disease prediction method based on the SIR model and the SEIR model, this embodiment will not rely solely on case data for prediction, but will combine keyword search data to improve the accuracy of the analysis of the epidemic situation of respiratory infectious diseases. The SIR model and the SEIR model are constructed using keyword search data, and the correlation between the search volume of different keywords and the spread of influenza is analyzed, and the effectiveness of the model is verified through comparative analysis. A regression model is established, and the prediction results of the keyword SIR model and the SEIR model are used as features for prediction, and the performance of the model is compared and analyzed and evaluated to test the effectiveness and generalization of the regression model. The limitations and room for improvement of the SIR model and the SEIR model are explored in depth, and targeted improvement measures are proposed to improve the accuracy and generalization of model predictions.
[0059] This embodiment verifies the effectiveness of the model through experiments, and conducts comparative analysis with the traditional SIR model and SEIR model to evaluate the performance of the improved model. In order to improve the prediction accuracy of the SIR model and the SEIR model to make them more in line with the actual situation, big data technology is introduced. On the basis of using case data to construct the classic SIR and SEIR models, the keyword search data related to influenza is used to construct the SIR model and SEIR model for each keyword to predict the changing trend of the keyword search volume in the future. At the same time, a regression model is constructed using keyword search data and case data, and a mapping relationship between keyword search volume and number of cases is established. Finally, the trend prediction results of the case number SIR model, the keyword SIR model and the regression model are combined to obtain a prediction of the number of cases in the future. In addition, this embodiment also replaces the SIR model with the SEIR model to conduct the same experiment. The overall experimental process is as follows. Figure 2 shown.
[0060] In order to make the technical solution of this embodiment clearer, the specific implementation process of this embodiment is described below by way of example, which mainly includes the following contents:
[0061] (1) Obtain case data and keyword search data of respiratory infectious diseases.
[0062] In this embodiment, the case data reported by the CDC is obtained by means of a web crawler. The daily current confirmed number of respiratory infectious diseases is calculated by the following formula: daily current confirmed number = cumulative confirmed number - cumulative cured number - cumulative death number. The search data uses the keyword search data in units of days provided by the Baidu search engine.
[0063] (2) Construct case prediction SIR model and SEIR model.
[0064] In this embodiment, the case prediction model refers to a model for predicting the number of cases infected with respiratory infectious diseases. For the SIR model and the SEIR model, this embodiment constructs two versions of the case prediction model. The first version of the case prediction model includes a first case prediction SIR model and a first case prediction SEIR model, and the second version of the case prediction model includes a second case prediction SIR model and a second case prediction SEIR model. The data set is divided as follows Figure 3 As shown. The first version of the case prediction model is used to train the regression model of keywords and case numbers, and the second version of the case prediction model is used to directly predict the number of confirmed cases per day in the future and serve as the input of the prediction results of the case prediction SIR model in the regression model (i.e., the case number prediction results in step S1). The SEIR model uses the same data set as the SIR model for training and testing.
[0065] In this embodiment, the training data of the first version of the case prediction model uses the daily existing confirmed case data from 2022 / 11 / 01 to 2022 / 11 / 17 (a total of 17 days). The total population N is defined as the number of permanent residents in the area to be tested at the end of 2022. The initial value of the susceptible person S is the total population N minus the initial value of the infected person I. The initial value of the infected person I is the number of existing confirmed cases on 2022 / 11 / 01, and the initial value of the recovered person R is 0. During training, the initial values of the parameters beta and gamma are both 0.1, and the optimization method is the L-BFGS-B (Limited-memory Broyden-Fletcher-Goldfarb-Shanno) algorithm. The L-BFGS-B algorithm uses limited memory to store and update the approximate inverse Hessian matrix, which is suitable for large-scale optimization problems and allows upper and lower boundary constraints to be imposed on variables. In each iteration, the algorithm uses gradient information to gradually update the parameter values to minimize the objective function. The boundaries of beta and gamma are (0.0, 10) and (0.07, 0.2), respectively. The loss function is defined as the mean square error between the predicted number of confirmed cases per day and the actual number of confirmed cases on that day. After obtaining R0 (basic reproduction number) through training, the number of confirmed cases per day for the next 200 days is predicted starting from 2022 / 11 / 01. For the SEIR model, the initial values of beta, gamma, and alpha are 0.73, 0.16, and 0.14, respectively, and the boundaries are (0.7, 0.77), (0.16, 0.3), and (0.05, 0.165), respectively. The initial value of the exposed person E is 0. The training data for the second version of the case prediction model is the daily confirmed case data from 2022 / 11 / 01 to 2022 / 12 / 12 (42 days in total). The initial value definition, parameter initial value, optimization parameter, and loss function definition are the same as those of the first version of the case prediction model. After obtaining R0 through training, the number of current confirmed cases per day for the next 200 days will be predicted starting from 2022 / 11 / 01.
[0066] (3) Construct SIR model and SEIR model of keywords.
[0067] In this embodiment, the keyword prediction model refers to a model for predicting the keyword search share. For the SIR model and the SEIR model, this embodiment constructs two groups of keyword prediction models. The first group of keyword prediction models includes a first keyword prediction SIR model and a first keyword prediction SEIR model, and the second group of keyword prediction models includes a second keyword prediction SIR model and a second keyword prediction SEIR model. Among them, the first group of keyword prediction models is used to train the regression model of keywords and case numbers, and the second group of keyword prediction models is used as the input of each keyword feature in the regression model.
[0068] In the first group of keyword prediction models, the training data of the first keyword prediction SIR model and the first keyword prediction SEIR model use data from 2022 / 11 / 01 to 2022 / 11 / 17 (a total of 17 days). The specific data used is the daily keyword search volume share minus the keyword search volume share on the corresponding date in 2019, and the calculation results less than zero are deleted. For missing values, linear interpolation is used. For the case where the first value is empty, take the nearest value divided by 2 to assign it a value. For the case where the last value is empty, take the nearest value multiplied by 1.2 to assign it a value. Then perform linear interpolation. The total search volume share N' is defined as 1, the initial value of the susceptible person S is defined as 1 minus the initial value of the infected person I, the initial value of the infected person I is defined as the keyword search volume share on 2022 / 11 / 01, and the initial value of the recovered person R is 0. During training, the initial values of the parameters beta and gamma are the parameter values trained by the corresponding case prediction model, and the optimization method is Nelder-Mead (simplex algorithm). Nelder-Mead is an iterative method for optimizing derivative-free multivariable functions. It adjusts the position of the simple shape vertices in each iteration through reflection, expansion, and contraction steps, and gradually approaches the minimum value of the function. The boundaries of beta and gamma are (0.01, 1000) and (0.07, 0.2), respectively. The loss function is defined as the mean square error between the predicted daily search volume share of this keyword and the actual daily search volume share of this keyword. After obtaining R0 through training, the daily keyword search volume share for the next 200 days is predicted from 2022 / 11 / 01. The same parameter settings are used for the SEIR model. For the initial value of the alpha parameter, it is set to 0.03 and the boundary is set to (1 / 30, 1). In the second group of keyword prediction models, the training data of each keyword prediction model uses the data from 2022 / 11 / 01 to 2022 / 12 / 12 (a total of 42 days). The rest of the processing is the same as the first group of keyword prediction models. After obtaining R0 through training, the daily keyword search volume share for the next 200 days is predicted from 2022 / 11 / 01.
[0069] (4) Regression model of keywords and number of cases.
[0070] From the prediction results of the first case prediction SIR model and the first keyword prediction SIR model, the data from 2022 / 11 / 18 to 2022 / 12 / 12 (a total of 25 days) are taken to form the training set of the regression model. Among them, the actual number of confirmed cases per day is the label, and the features are composed of "case prediction SIR model prediction results", "'influenza (flu)' keyword prediction model prediction results", "'pneumonia' keyword prediction model prediction results", "'body temperature' keyword prediction model prediction results", "'fever' keyword prediction model prediction results", "'antipyretic' keyword prediction model prediction results", "'pharyngitis' keyword prediction model prediction results", "'sum' keyword prediction model prediction results" and so on. In the keyword feature, the "sum" keyword represents the proportion of the sum of all keyword search volumes to the total search volume of the day. The model training set is finally divided into a training set containing 20 samples and a validation set containing 5 samples.
[0071] In this embodiment, in the processing part of the input data, the case prediction model features are not preprocessed, but all keyword features are log-scaled using a scaling formula, and the scaling formula is as follows:
[0072] N t = -np.log2(O t +1e-10);
[0073] Among them, N t represents the new eigenvalue, O t Represents the original eigenvalue, and -np is the scaling factor.
[0074] In the present embodiment, the regression model is a computational model for determining the relationship between the independent variable and the dependent variable, and the case prediction SIR model, the keyword prediction SIR model, the keyword search volume, the actual number of cases, etc. are combined as independent variables to predict the number of confirmed cases per day in the future. The regression model of the present embodiment adopts elastic network regression (Elastic Net Regression), which is a linear regression model that combines the advantages of ridge regression (Ridge Regression) and lasso regression (LassoRegression). By introducing two regularization terms of L1 and L2, elastic network regression can handle the multicollinearity problem in high-dimensional data, and variable selection can be performed at the same time. This model is suitable for the case where the number of variables is greater than the number of observed values, and can effectively improve the stability and prediction performance of the model. In the present embodiment, the l1_ratio parameter is set to 0.5, which is used to control the proportion of the L1 regularization term in the whole model. The eps parameter is set to 0.001, which is used to define the threshold value of model convergence. The n_alphas parameter is set to 100, which represents the number of alpha values on the regularization path, which is used to control the intensity of regularization. The alphas parameter is None, which means that a list of candidate values is automatically generated based on n_alphas. The fit_intercept parameter is set to True, which means that the intercept is considered when fitting the model. The precompute parameter is set to auto, which means that the precomputed Gram matrix is automatically selected according to the situation. The max_iter parameter is set to 1000, which indicates the maximum number of iterations of the model. The tol parameter is set to 0.0001, which indicates the convergence tolerance of the model. The cv parameter is set to 5, which indicates that 5-fold cross-validation is used for parameter optimization. The copy_X parameter is set to True, which means that the input data is copied to prevent it from being modified. The verbose parameter is set to 0, which means that no fitting information is output. The n_jobs parameter is None, which means that a single job is used for calculation. The positive parameter is set to False, which means that the coefficient is not forced to be non-negative. The random_state parameter is None, which means that the seed of the random number generator is not fixed. The selection parameter is set to cyclic, which means that the cyclic path update rule is used in the elastic network. The training set fitting of the first case prediction SIR model is shown in the figure. Figure 4 As shown. In the prediction results of the second case prediction SIR model and the second keyword prediction SIR model, the data from 2022 / 12 / 13 to 2022 / 12 / 20 (a total of 8 days) are taken to form the final test set, which is input into the trained regression model to compare the prediction results of the traditional SIR model and the method of this embodiment. For the first case prediction SEIR model, the first keyword prediction SEIR model, the second case prediction SEIR model and the second keyword prediction SEIR model, the above steps are also followed.
[0075] (5) Experimental results.
[0076] 1) Results of the SIR model.
[0077] The experimental results of the first case prediction SIR model and the first keyword prediction SIR model are shown in Table 1, as follows:
[0078] Table 1 Results of the first case prediction SIR model and the first keyword prediction SIR model
[0079]
[0080]
[0081] As can be seen from Table 1, the RMSE of the first case prediction SIR model on the training set is 125.60, R 2 is 0.91, which is a good fit. The R0 obtained from training is 2.06, and the recovery rate is 0.10, which means that the infected person can recover in about 10 days on average. Figure 5 It is known that the peak of influenza will be in mid-February 2023, with the highest daily confirmed cases exceeding 3.5 million. Figures 6 to 12 It can be seen that in the first keyword prediction SIR model, the keyword "body temperature" is not well fitted, the prediction peaks of the keywords "fever", "pneumonia" and "pharyngitis" are in March to April, and the prediction peaks of the remaining keywords are in December to February. In their prediction results, the data from 2022 / 11 / 18 to 2022 / 12 / 12 constitute the training set of the regression model.
[0082] 2) Results of SEIR model.
[0083] In order to conduct an ablation study, this embodiment replaces the SIR model in the above method with the SEIR model and conducts the same experiment. The experimental results are as follows: Figures 13 to 21 As shown, according to Figures 13 to 21 It can be seen that the experimental results of the first case prediction SEIR model and the first keyword prediction SEIR model are extremely similar to the experimental results of the first case prediction SIR model and the first keyword prediction SIR model, which shows that the method of this embodiment is also applicable to the SEIR model. Table 2 shows the results of the first case prediction SEIR model and the first keyword prediction SEIR model.
[0084] Table 2 Results of SEIR model for first case prediction and SEIR model for first keyword prediction
[0085] Keywords <![CDATA[R 2 ]]> RMSE R0 Infection rate Recovery rate Latent period influenza 0.77 2.69e-4 2.27 0.16 0.07 1.00 pneumonia 0.62 5.65e-5 1.96 0.14 0.07 1.00 body temperature -0.36 2.44e-5 1.49 0.10 0.07 1.00 fever 0.25 1.75e-5 15.72 1.10 0.07 29.99 Antipyretics 0.77 2.66e-6 14.77 1.03 0.07 5.52 pharyngitis 0.75 1.84e-6 2.52 0.18 0.07 1.62 sum 0.70 4.45e-4 3.53 0.25 0.07 1.00 Number of infected cases 0.81 182.27 4.81 0.77 0.16 6.06
[0086] As shown in Table 2, the RMSE of the SEIR model for the first case prediction on the training set is 182.27, R 2 is 0.81, compared with the first case prediction SIR model, R 2 The R0 has decreased and the RMSE has increased, indicating that the fitting effect of the SEIR model for first case prediction is slightly worse than that of the SIR model for first case prediction. This may be because the SEIR model for first case prediction takes into account the incubation period, making the model more complex. The R0 obtained after training is 4.81, which means that each infected person will infect 4.81 people on average. The recovery rate is 0.16, and the incubation period is 6.06 days, which means that it takes an average of about 6 days for an infected person to develop symptoms. Compared with the SIR model for first case prediction, R0 has increased, the recovery rate has decreased, and the incubation period has also increased. This may be because the SEIR model for first case prediction better captures the disease transmission dynamics. Fig. 22 As shown in the figure, for the regression equation construction results, the SIR model version is first used. The RMSE of the SIR model on the validation set is 869.79, R 2 The optimal alpha value obtained by training is 2267130.47, and the optimal l1_ratio value is 0.5. Then the SEIR model is used. The RMSE of the SEIR model on the validation set is 1010.52, R 2 The optimal alpha value obtained through training is 3902904.18, and the optimal l1_ratio value is 0.5.
[0087] Based on the fitting results of the second case prediction SIR model, the second case prediction SEIR model, the second keyword prediction SIR model, and the second keyword prediction SEIR model, the second case prediction SIR model and the second keyword prediction SEIR model are used. As shown in Table 3, the RMSE of the second case prediction SIR model on the training set is 902.40, R 2 The R0 obtained from training is 1.48, and the recovery rate is 0.20, which means that infected people can recover in an average of 5 days.
[0088] Table 3 Results of the second case prediction SIR model and the second keyword prediction SIR model
[0089] Keywords <![CDATA[R 2 ]]> RMSE R0 Infection rate Recovery rate influenza 0.79 3.46e-3 1.41 0.28 0.20 pneumonia 0.84 2.84e-4 1.29 0.26 0.20 body temperature 0.67 3.54e-4 1.50 0.30 0.20 fever 0.71 1.98e-4 1.51 0.30 0.20 Antipyretics 0.95 1.26e-4 2.07 0.41 0.20 pharyngitis 0.68 1.72e-05 1.32 0.26 0.20 sum 0.84 6.87e-3 1.62 0.32 0.20 Number of infected cases 0.94 902.40 1.48 0.30 0.20
[0090] Using the SEIR model for the second case prediction, the SIR model in the above method was replaced with the SEIR model and the same experiment steps were performed. The following are the corresponding results:
[0091] In the scenarios of the second case prediction SEIR model and the second keyword prediction SEIR model, the training set RMSE of the second case prediction SEIR model is 1174.67, R 2 Compared with the second case prediction SIR model (RMSE is 902.40, R 2 The R of the SEIR model for the second case prediction is 0.94. 2 The R0 has decreased and the RMSE has increased, indicating that the fitting effect is slightly worse. This may be because the SEIR model for second case prediction takes into account the incubation period, making the model more complicated. The R0 obtained by the SEIR model for second case prediction is 4.67, which means that each infected person will infect 4.67 people on average. The recovery rate is 0.16, and the incubation period is 6.63 days, which means that it takes an average of about 6.63 days for the infected person to develop symptoms. Compared with the SIR model for second case prediction (R0 is 1.48 and the recovery rate is 0.20), R0 has increased and the recovery rate has decreased. This may be because the SEIR model for second case prediction better captures the transmission dynamics of the disease.
[0092] Table 4 shows the SEIR model for second case prediction for different keywords in Beijing. It can be seen that the parameters such as R0, infection rate, recovery rate and incubation period of different keywords are different, which reflects the difference in the impact of different keywords on the spread of influenza. For example, the R0 of "antipyretics" is the highest (35.74), indicating that the demand for antipyretics is highly correlated with the spread of influenza; while the R0 of "pharyngitis" is relatively low (11.99), indicating that the correlation between pharyngitis symptoms and the spread of influenza is relatively low.
[0093] Table 4 Results of SEIR model for second case prediction and SEIR model for second keyword prediction
[0094] Keywords <![CDATA[R 2 ]]> RMSE R0 Infection rate Recovery rate Latent period influenza 0.88 2.64e-3 8.82 0.62 0.07 16.38 pneumonia 0.91 2.09e-4 11.18 0.78 0.07 29.99 body temperature 0.74 3.14e-4 17.61 1.23 0.07 30.00 fever 0.77 1.75e-4 18.03 1.26 0.07 29.99 Antipyretics 0.96 1.06e-4 35.74 2.50 0.07 24.25 pharyngitis 0.76 1.51e-05 11.99 0.84 0.07 29.99 sum 0.91 5.15e-3 7.19 0.50 0.07 7.27 Number of infected cases 0.89 1174.67 4.67 0.75 0.16 6.63
[0095] For the final test results, the version of the SIR model is adopted, which is Fig.23 It can be seen that during the period from December 13, 2022 to December 20, 2022, the predicted results of the method proposed in this embodiment are significantly better than the classic SIR results, and the RMSE drops from 11330.99 to 3284.47. Fig.24It can be seen that the peak value of the daily current confirmed cases predicted by the classic SIR method is greater than the peak value predicted by the method proposed in this embodiment. The peak value of influenza predicted by the classic SIR method is around mid-February 2023, with the highest daily current confirmed cases reaching more than 1.2 million. The peak value of influenza predicted by the method proposed in this embodiment is similar to that of the classic SIR method, with the highest daily current confirmed cases reaching more than 600,000. In the version using the SEIR model, this embodiment replaces the SIR model in the above method with the SEIR model and conducts the same experiment steps. Fig.25 and Fig.26 It can be seen that the method proposed in this embodiment can also achieve good results on the SEIR model. It can be seen that by adding keyword search results, the prediction effect of the infectious disease model (including the SIR model and the SEIR model) can be significantly improved.
[0096] This embodiment uses keyword search data and infection case data to construct a SIR model, and associates the two through a regression model to predict the development trend and epidemic situation of influenza. The final results show that the established keyword SIR model (including the first keyword prediction SIR model and the second keyword prediction SIR model) is very effective in predicting the development trend of influenza, and the regression model can also successfully associate the keyword search data with the number of infection cases. First, the keyword SIR model shows a certain effectiveness in predicting the development trend of influenza. By constructing a keyword SIR model, this embodiment can capture the correlation between the search volume of different keywords and the spread of influenza. For example, in the second keyword prediction SIR model, the predicted peak value of the keyword "pharyngitis" is between March and April, indicating that the search trend of this keyword is similar to that of influenza, while the search peak values of "influenza", "body temperature" and "antipyretic drugs" are between December and January, indicating that these symptoms appear the earliest and disappear the earliest. These results are consistent with the actual spread trend of influenza, indicating that the keyword SIR model can reflect the development of influenza to a certain extent. Secondly, the regression model performs well in associating keyword search data with the number of cases. By training the regression model, this embodiment can use the prediction results of the keyword SIR model as features to predict the number of confirmed cases per day. On the validation set, the regression model shows a lower RMSE and a higher R 2 The value indicates that the model can fit the actual data well. This confirms that there is a certain correlation between keyword search data and the number of cases, which can be predicted through the regression model.
[0097] It should be noted that this embodiment still has some limitations and room for improvement. First, the time period of the training data of the SIR model is in the early stage of influenza. Since there is not enough data on the outbreak of influenza, the calculation of R0 may not be accurate enough. The R0 of some keyword prediction models may not be accurate enough. 2 It is not very high. In addition, the training set data of the regression model is relatively small, only 25, which may have a certain impact on the generalization of the regression model and cause overfitting. The main reason for these problems is that the case data used stopped counting on 2022 / 12 / 20, so the data of the influenza stage in this embodiment only has a small amount of data from 2022 / 11 / 01 to 2022 / 12 / 20. In addition, the SIR model itself has certain limitations. For example, it may not be able to accurately predict the spread of complex influenza, and it has high requirements for the time period and quality of the data. In addition, the SIR model assumes that the behavior and infection rate of the population remain unchanged, and cannot consider the impact of changes in people's behavior and the implementation of prevention and control measures on the spread of influenza. In the future, it can be considered to collect more data through different platforms or methods to build a model with better performance. This embodiment also uses the SEIR model to predict the trend of the number of infected cases and the proportion of keywords, and the method of this embodiment is also applied to the SEIR model. The results of the experiment are basically consistent with the SIR model. Therefore, the influenza trend prediction method of this embodiment based on the traditional SIR model / SEIR model and the search volume of influenza-related keywords can effectively improve the performance of the traditional SIR model and the SEIR model, obtain more accurate prediction results, and provide new ideas and methods for early warning research on respiratory infectious diseases.
[0098] At present, big data technology is widely used in the prediction of respiratory infectious diseases. The big data technology in this embodiment is specifically embodied as follows:
[0099] 1) By monitoring online search engine queries submitted by millions of users around the world every day, this embodiment can analyze the behavior of users in consulting health issues, especially the relative frequency of certain queries is highly correlated with the proportion of patients with flu-like symptoms, which enables this embodiment to accurately estimate the current level of weekly influenza activity in various regions, with a reporting lag of only one day. This method utilizes a large amount of network search data and provides a new way to predict the prevalence of influenza, which helps to detect disease activity early and take countermeasures quickly, thereby reducing the impact of seasonal influenza and pandemic influenza.
[0100] 2) Using the time series of online search query frequencies, the prevalence of influenza in various regions can be inferred. Unsupervised modeling techniques and transfer learning techniques can reduce the expected bias caused by public interest and improve prediction accuracy. Incorporating web search data into the prediction model improves the short-term prediction accuracy of the influenza autoregressive model. Therefore, online search data can be used to develop complementary public health surveillance methods to provide information for response measures for various respiratory infectious diseases such as influenza.
[0101] 3) By analyzing online search data, the correlation between the public's concern about infectious diseases and the actual number of infected cases was found. This embodiment can use Google TrendsTM data to compare the relative search volume index with the number of influenza cases and find that the public's concern is highest on average 11.5 days before the peak of new infected cases.
[0102] 4) The role of big data in public health was highlighted by showing that Twitter data can improve flu forecasts. Prediction models using Twitter data can reduce prediction errors, predict earlier, and predict flu prevalence better than models using Google Flu Trends data to some extent.
[0103] 5) Using real-time, geographically concentrated and easily accessible data, an influenza forecasting model can be developed to provide early warning to medical centers, so that there is enough time to implement intervention measures. The forecasting model is mainly based on the Box-Jenkins forecasting procedure, the generalized linear model (GLM) and the generalized linear autoregressive moving average (GARMA) method. As the only external source of information, Google Flu Trends data can significantly improve the accuracy of the forecast when added to the forecasting model.
[0104] This embodiment successfully improves the accuracy of the traditional dynamic model in predicting the development trend of influenza by constructing a traditional SIR model and combining it with big data technology and introducing keyword search data. The results show that there is a certain correlation between the search volume of keywords related to influenza and the spread trend of influenza, which can be predicted by the keyword SIR model, and the keyword search volume can be linked to the number of cases by the regression model. The research results of this embodiment have certain guiding significance for the monitoring and prediction of similar infectious diseases such as influenza. By utilizing the association between Internet search data and traditional influenza data, the trend and possible changes of influenza can be understood more timely, which is of great significance for the formulation of public health strategies, resource allocation and risk management.
[0105] In an exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the above-mentioned method for predicting the epidemic status of respiratory infectious diseases are implemented.
[0106] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for predicting the epidemic status of respiratory infectious diseases.
[0107] In an exemplary embodiment, a computer program product is provided, including a computer program, which implements the steps of the above-mentioned method for predicting the epidemic status of respiratory infectious diseases when executed by a processor.
[0108] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0109] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A method for predicting the prevalence of respiratory infectious diseases, characterized in that: The method for predicting the epidemic situation of respiratory infectious diseases comprises: Obtain case data and keyword search data of respiratory infectious diseases in the area to be tested; the case data includes the number of existing confirmed cases, the number of existing recovered people and the total number of people, the total number of people is the total permanent population of the area to be tested, the keyword search data includes the existing keyword search ratio, the existing non-searched keyword ratio and the total search volume ratio, the existing keyword search ratio refers to the ratio of the search volume of keywords related to respiratory infectious diseases to the total search volume in the same period, the existing non-searched keyword ratio refers to the ratio of the search volume of non-respiratory infectious disease keywords in the same period, the total search volume ratio refers to the ratio of the total search volume in the same period to the total search volume in previous years in the same period, keywords refer to words related to respiratory infectious diseases, and the total search volume refers to the sum of the search volumes of all words; The case data is input into a case prediction model to obtain a case number prediction result; the case prediction model is a trained case SIR model or a case SEIR model, and the case number prediction result is the number of confirmed cases per day in a future time period predicted based on the case data; The keyword search data is input into a keyword prediction model to obtain a keyword search volume prediction result; the keyword prediction model is a trained keyword SIR model or a keyword SEIR model, and the keyword search volume prediction result is a daily keyword search percentage predicted based on the keyword search data and in the same time period as the case number prediction result; the daily keyword search percentage refers to the proportion of daily searches for keywords related to respiratory infectious diseases to the search volume of related terms in previous years during the same time period; The case number prediction result and the keyword search volume prediction result are input into the regression model to obtain the final case number prediction result; the regression model is a trained calculation model for determining the correlation between independent variables and dependent variables, the independent variables include the case prediction model, the keyword prediction model, the keyword search volume and the actual number of confirmed cases, the dependent variables include the number of confirmed cases per day in a future time period, and the final case number prediction result is the number of confirmed cases per day in a future time period predicted based on the case data and the keyword search data.
2. The method for predicting the prevalence of respiratory infectious diseases according to claim 1, characterized in that: Obtain case data and keyword search data of respiratory infectious diseases in the area to be tested, including: Using a web crawler to obtain the case data; Using a search engine, the keyword search data in units of days is obtained.
3. The method for predicting the prevalence of respiratory infectious diseases according to claim 1, characterized in that: The optimization algorithm used in training the case prediction model is the L-BFGS-B algorithm, and the loss function is the mean square error between the predicted number of confirmed cases per day and the actual number of confirmed cases on that day.
4. The method for predicting the prevalence of respiratory infectious diseases according to claim 1, characterized in that: The optimization algorithm used in the keyword prediction model training is the Nelder-Mead algorithm, and the loss function is the mean square error between the predicted daily keyword search proportion and the actual keyword search proportion on that day.
5. The method for predicting the prevalence of respiratory infectious diseases according to claim 1, characterized in that: The keywords include the name, symptoms and symptomatic drugs of the respiratory infectious disease.
6. The method for predicting the prevalence of respiratory infectious diseases according to claim 5, characterized in that: The respiratory infectious disease is influenza.
7. The method for predicting the prevalence of respiratory infectious diseases according to claim 6, characterized in that: The keywords include influenza, flu, fever, pharyngitis, pneumonia, body temperature and antipyretics.
8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for predicting the epidemic status of respiratory infectious diseases according to any one of claims 1 to 7.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the epidemic status of respiratory infectious diseases described in any one of claims 1 to 7 is implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting the epidemic status of respiratory infectious diseases described in any one of claims 1 to 7 is implemented.