Infectious disease prediction method considering trend learning and cross-regional transmission

Abstract

The invention discloses an infectious disease prediction method considering trend learning and cross-regional transmission, and the method comprises the steps: firstly dividing a region range and an infectious disease group type, obtaining infectious disease infection data, defining a use interval of the data, then defining an effect evaluation index, and setting an infectious disease related parameter; dividing an infectious disease stage into an infectious disease stable scene and an infectious disease mutation scene according to different acquired infectious disease infection data; an M-SEIYAQURD-H model is established on the basis of an improved multi-band learning algorithm; and obtaining a static prediction result of the M-SEIYAQURD-H model under the fitting sub-cycle framework, obtaining a dynamic prediction result of the M-SEIYAQURD-H model under the rolling prediction framework, and respectively comparing the dynamic and static prediction effects of each scene. According to the method, the trend change of epidemic situation development is brought into a model consideration range, so that the influence of dynamics and uncertainty of epidemic situation evolution on model prediction is solved to a certain extent, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of infectious disease prediction, and in particular to an infectious disease prediction method considering trend learning and cross-regional transmission. Background Art

[0002] Major public health emergencies, such as infectious disease outbreaks, continue to threaten human health and the country's social and economic development. Predicting complex infectious disease outbreaks and understanding their development trends has become a crucial research topic.

[0003] In existing infectious disease prediction research, traditional forecasting methods are primarily based on simple mathematical models, such as the Susceptible-Infectious-Recovered (SIR) model and the Susceptible-Exposed-Infected-Recovered (SEIR) model. While these models can describe the spread of infectious diseases to a certain extent, they have significant limitations. They typically assume a relatively stable epidemic environment and ignore the significant impact of key factors such as virus mutations and policy changes on epidemic trends. Furthermore, most of these models treat the study area as an isolated entity, failing to fully account for cross-infection caused by the movement of people and goods between different regions. When faced with complex epidemics, the predictions of traditional models often deviate significantly from the actual situation, making them unable to provide a reliable basis for public health decision-making.

[0004] With the rapid development of information technology, emerging technologies such as big data and artificial intelligence are increasingly being applied to infectious disease prediction. Some studies have attempted to utilize machine learning algorithms, such as neural networks and support vector machines, to analyze and predict infectious disease data. While these methods offer advantages in processing large-scale, high-dimensional data, they also face challenges such as uneven data quality and poor model interpretability. Furthermore, existing prediction methods based on emerging technologies have also failed to adequately address the challenges of sudden changes in epidemic trends and cross-regional spread, leaving significant room for improvement in practical applications.

[0005] In order to effectively respond to the challenges brought by infectious diseases and improve the accuracy and reliability of predictions, there is an urgent need for a new prediction method that fully considers the factors of changing epidemic trends and cross-regional transmission and overcomes the shortcomings of traditional methods and existing technologies. Summary of the Invention

[0006] The purpose of this invention is to provide an infectious disease prediction method that takes into account trend learning and cross-regional transmission, incorporating the trend changes of epidemic development into the model consideration range, to a certain extent solving the impact of the dynamics and uncertainty of epidemic evolution on model prediction, and improving the accuracy of prediction.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A method for predicting infectious diseases considering trend learning and cross-regional transmission includes the following steps:

[0009] S1. First, divide the regional scope and infectious disease group type, obtain infectious disease infection data, define the data usage range, then define the effect evaluation indicators and set infectious disease related parameters;

[0010] S2. Based on the acquired infectious disease infection data, the infectious disease stage is divided into an infectious disease stable scenario and an infectious disease mutation scenario;

[0011] S3, establish the M-SEIYAQURD-H model based on the improved multi-band learning algorithm;

[0012] S4. Obtain the static prediction results of the M-SEIYAQURD-H model in the fitting sub-period framework, and obtain the dynamic prediction results of the M-SEIYAQURD-H model in the rolling prediction framework, and compare the dynamic and static prediction results of each region respectively;

[0013] S5. Perform static prediction of the M-SEIYAQURD-H model and dynamic prediction of the M-SEIYAQURD-H model in the infectious disease stable scenario and infectious disease mutation scenario respectively, and calculate the effect evaluation index to compare the effects.

[0014] Preferably, in S1, the infectious disease group types include susceptible people, latent people, infected people, symptomatic infected people, asymptomatic infected people, isolated and treated people, people with symptoms but not isolated and treated, recovered people and deceased people.

[0015] Preferably, in S1, the infectious disease related parameters include the geographical mixing coefficient h k,j :

[0016]

[0017] Among them, h0 represents the proportion of transmission contact within the region, and (1-h0) represents the proportion of transmission contact to other external regions; d k,j Represents the geographic coordinate Euclidean distance between region k and region j, 1 / d k,j To represent the possibility of individuals moving from region k to region j.

[0018] Preferably, in S3, obtaining the static prediction results of the M-SEIYAQURD-H model under the fitting sub-period framework specifically includes:

[0019] Input the training sample data of the full time period and the population size status of the initial prediction period; initialize the fitting starting point and sub-period values; divide the sub-period and use the MH algorithm to fit and train to obtain the M-SEIYAQURD-H model of the sub-period; update the fitting starting point and initial population size status of the next sub-period, repeat the above process until all data are fitted, and output the final M-SEIYAQURD model to achieve static prediction.

[0020] Preferably, the sub-period is divided and the MH algorithm is used for fitting and training to obtain the M-SEIYAQURD-H model of the sub-period, specifically including:

[0021] Input the training sample data of the time period to be fitted and the population status of the initial period of prediction; extract the relevant parameters of the infectious disease from the random distribution, generate the initial parameter set and calculate the loss function; extract the new parameter set through the maximum likelihood ratio and calculate the likelihood ratio; update the parameter set according to the likelihood ratio, repeat the iteration until the upper limit of the number of iterations is reached, and output the final fitting parameter set;

[0022] Preferably, in S3, obtaining the dynamic prediction results of the M-SEIYAQURD-H model under the rolling prediction framework specifically includes:

[0023] Input the length of the prediction time period, the trained M-SEIYAQURD-H model, and the fitting sub-period; initialize the prediction starting point and the value of the rolling prediction period; divide the sub-period and use the trained M-SEIYAQURD-H model to iteratively calculate the predicted value within this rolling period; update the training set and fit the next sub-period, and repeat the above process until the prediction of all period data is completed.

[0024] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it implements an infectious disease prediction method that considers trend learning and cross-regional transmission as described above.

[0025] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0026] (1) This paper studies the problem of infectious disease prediction considering trend learning and cross-regional transmission, establishes an infectious disease prediction model, proposes a band learning algorithm, and verifies the effectiveness of the model and algorithm through numerical experiments. An improved multi-band learning algorithm is used to establish the M-SEIYAQURD-H model. Taking trend learning into account, the trend changes in the development of the epidemic are taken into account in the model. This solves the impact of the dynamics and uncertainty of the epidemic evolution on the model prediction to a certain extent, and improves the accuracy of the prediction.

[0027] (2) Due to the mutual influence of epidemic spread among multiple regions, the M-SEIYAQURD-H model of the present invention establishes the population conversion relationship between various groups in different regions, taking into account regional heterogeneous mixing, making the model more in line with the actual situation of epidemic spread and the prediction results more in line with actual expectations. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0029] Figure 1 A flow chart of an infectious disease prediction method considering trend learning and cross-regional transmission provided by the present invention;

[0030] Figure 2 Schematic diagram of input and output of the M-SEIYAQURD-H model of the present invention;

[0031] Figure 3 A flow chart for constructing the M-SEIYAQURD-H model of the present invention;

[0032] Figure 4 This is a flow chart of the improved multi-band learning algorithm of the present invention;

[0033] Figure 5 Flowchart of the Metropolis-Hastings algorithm used in the present invention;

[0034] Figure 6 This is a schematic diagram of the static prediction cycle structure of the present invention;

[0035] Figure 7 This is a schematic diagram of the dynamic prediction cycle structure of the present invention. DETAILED DESCRIPTION

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] like Figure 1As shown, the present invention provides an infectious disease prediction method considering trend learning and cross-regional transmission, comprising the following steps:

[0039] S1. First, divide the regional scope and infectious disease group type, obtain infectious disease infection data, define the data usage range, then define the effect evaluation indicators and set infectious disease related parameters;

[0040] S2. Based on the acquired infectious disease infection data, the infectious disease stage is divided into an infectious disease stable scenario and an infectious disease mutation scenario;

[0041] S3, establish the M-SEIYAQURD-H model based on the improved multi-band learning algorithm;

[0042] S4. Obtain the static prediction results of the M-SEIYAQURD-H model in the fitting sub-period framework, and obtain the dynamic prediction results of the M-SEIYAQURD-H model in the rolling prediction framework, and compare the dynamic and static prediction results of each region respectively;

[0043] S5. Perform static prediction of the M-SEIYAQURD-H model and dynamic prediction of the M-SEIYAQURD-H model in the infectious disease stable scenario and infectious disease mutation scenario respectively, and calculate the effect evaluation index to compare the effects.

[0044] Specifically, this paper considers the interplay between different regions and proposes an infectious disease prediction method that takes into account regional heterogeneity. This is explained below using a specific region as an example. The problem is defined as follows:

[0045] (1) Group division: According to the actual characteristics of a region, the population can be divided into nine different groups during the outbreak process, namely, susceptible (S), latent (E), infected (I), symptomatic (Y), asymptomatic (A), quarantined (Q), symptomatic but not quarantined (U), recovered (R), and deceased (D).

[0046] (2) Transformation relationship: As time goes by and the epidemic spreads, there is a relationship of transmission and state transfer between different groups.

[0047] Define N k is the population contained in region k; S k,t 、E k,t , I k,t 、Y k,t 、Ak,t , Q k,t 、U k,t 、R k,t 、D k,t They represent the number of susceptible people, latent people, infected people, symptomatic infected people, asymptomatic infected people, symptomatic people with isolation treatment, symptomatic people without isolation treatment, recovered people and deaths in region k at period t; Represent the latent population E k,t , Asymptomatic infection group A k,t , Group U with symptoms but not isolated for treatment k,t The disease transmission rate, that is, the susceptible S k,t The probability of transmission per contact; α k For the lurker group E k,t Transformed into infected group I k,t The potential conversion rate; They are the asymptomatic infected group A without considering the natural mortality rate. k,t , group Q of people with symptoms and isolated treatment k,t , Group U with symptoms but not isolated for treatment k,t mortality rate due to disease; Represents asymptomatic infected group A k,t , group Q of people with symptoms and isolated treatment k,t , Group U with symptoms but not isolated for treatment k,t The recovery rate due to self-healing or treatment; p is the infected population I k,t Among them, asymptomatic infection group A k,t The proportion of symptomatic infection group Y k,t Middle, isolate the healer group Q k,t The ratio of d k,j Represents the geographic coordinate Euclidean distance between region k and region j.

[0048] (3) Effective contact transmission rate: by constructing the geographical mixing coefficient h k,j , to describe the degree of mutual transmission contact between people in different regions, as shown in the following formula:

[0049]

[0050] Among them, h0 represents the proportion of transmission contact within the region, and (1-h0) represents the proportion of transmission contact with other external regions. Therefore, the geographical mixing coefficient within this region is h k,j =h0; and for the interaction between regions, d k,j It represents the geographical coordinate Euclidean distance between region k and region j, and the proportion of its reciprocal represents the possibility of individuals moving from region k to region j. In summary, the geographical mixing coefficient hk,j It is to characterize the cross-transmission rate between different regions through the inverse of geographical distance and contact ratio.

[0051] (4) Define the evaluation index: Use the mean absolute percentage error (MAPE) as the evaluation index for fitting or prediction effect. The calculation formula is as follows. t ′ represents the actual observed data of infected people in period t. The lower the MAPE value, the more accurate the model fitting or prediction.

[0052]

[0053] (5) Set infectious disease related parameters: disease transmission rate β of each group E ,β A ,β U The priori value range of is set to U(0,0.08) uniformly distributed; the priori value range of the latent conversion rate α of the latent person is set to U(0.2,0.3) uniformly distributed; the mortality rate δ of the three groups is set to U(0,0.08) uniformly distributed; A ,δ Q ,δ U The priori value range of is set to U(0,0.0005) uniformly distributed; the recovery rate of the three groups γ A ,γ Q ,γ U The prior value range of is set to U(0,0.05) uniform distribution; the prior value range of the proportion of asymptomatic infections p is set to U(0,1) uniform distribution; the prior value range of the proportion of isolated treatment persons q is set to U(0,1) uniform distribution.

[0054] like Figure 2 As shown, based on the known data on the number of 9 types of groups in each region and related parameters, and on the basis of setting information such as the conversion relationship between groups and the geographical mixing coefficient, the present invention can use the model to predict the number of people in each group in the future, calculate the effect evaluation index, and compare the difference between direct prediction and rolling prediction, verifying the significant role of rolling prediction in improving prediction accuracy.

[0055] like Figure 3 As shown in Figure 2, this method is mainly divided into four modules, namely the improved multi-band learning algorithm, the Metropolis-Hastings algorithm, the fitting sub-period framework, and the prediction framework based on the rolling perspective. The overall process is as follows:

[0056] Improved multi-band learning algorithm: Due to the influence of factors such as policy changes, holiday gatherings and virus mutations, the development trend of the epidemic is sudden and random. In order to take this feature into account, the present invention refers to the multi-band learning algorithm (Multipeak Learning Algorithm, MLA) to learn and capture the occurrence of new bands of the epidemic, and based on the structure and characteristics of the above model, improves the structure of MLA and related calculation formulas, and further expands the model into a multi-band SEIYAQURD-H (Multipeak SEIYAQURD-H, M-SEIYAQURD-H) model. The algorithm mainly has the following 6 steps: Figure 4 As shown:

[0057] (1) Input: training sample data for the entire time period and the population size status at the initial prediction period;

[0058] (2) Initialization: Initialize the first band of the current epidemic, as well as the upper and lower limits of the martingale value in subsequent detections;

[0059] (3) Model training: The algorithm trains the model of the current band o, and the starting point of the training set is the initial period T of the current band. o , the end point is the last period T, and the SEIYAQURD-H model under band o is obtained through training and fitting;

[0060] (4) Calculation: Take the initial period of the band T o As the starting point of the new band detection, the weighted error value Z corresponding to each period t is calculated in turn t , error anomaly W t and the martingale value M t ;

[0061] (5) Detecting new bands: If the martingale value of the current period exceeds the specified upper threshold It means that a new band has been detected, the detection result is recorded and the detection is stopped; if the martingale value of the current period is less than the specified lower limit threshold This means that the fitting effect is gradually improving over time, so this is used as a new starting point for band detection and the above detection steps are repeated;

[0062] (6) Determine the termination condition: After the detection is completed, if there is a new band, the SEIYAQURD-H model is used to iteratively calculate the initial state of the population in the next band for the next round of model training, and the above steps are repeated to continue searching for the next new band; if no new band is detected, the algorithm stops and outputs the multi-band SEIYAQURD-H (Multipeak SEIYAQURD-H, M-SEIYAQURD-H) model.

[0063] Metropolis-Hastings (MH) algorithm: This paper uses the Metropolis-Hastings (MH) algorithm in the Markov Chain Monte Carlo (MCMC) method to fit the model. This algorithm is often used in the fitting process of classic infectious disease cabin models such as SIR and SEIR and has good results. The algorithm mainly has the following 5 steps: Figure 5 As shown:

[0064] (1) Input: Input the training sample data of the time period to be fitted, and the population size status of the initial period of prediction;

[0065] (2) Extract parameters and calculate SSE: Randomly extract infectious disease-related parameters from a given random distribution to generate an initial parameter set ω, substitute it into the SEIYAQURD model and calculate the corresponding loss function SSE;

[0066] (3) Extract a new parameter set and calculate LR: In each iteration, a new parameter set ω is extracted by maximum likelihood. l , and calculate the likelihood ratio LR between the parameter set ω of the previous round;

[0067] (4) Update iteration: Draw a random number ν from the uniform distribution U(0,1). If LR>ν, update the parameter set ω=ω l ;

[0068] (5) Determine the termination condition: Repeat the above process until the upper limit of the number of iterations L is reached, the algorithm stops, and the final fitting parameter set ω is output.

[0069] Fitting sub-period framework: To improve fitting accuracy and reduce the number of iterations required for fitting, this section proposes a fitting sub-period framework for model fitting. The specific process is as follows:

[0070] (1) Input: Input the training sample data of the entire time period and the population size status of the initial period of prediction;

[0071] (2) Initialization: Initialize the fitting starting point t s and the value of the sub-period π;

[0072] (3) Divide into sub-periods: Calculate the end point t of the sub-period fitting in each round e If the remaining time period is less than the sub-period π, it is directly regarded as the last sub-period. Figure 6 As shown;

[0073] (4) Fitting training: The M-SEIYAQURD-H model of the sub-period is obtained by fitting training using the MH algorithm;

[0074] (5) Update and iteration: Use the model iterative calculation to update the fitting starting point and the initial state of the population size for the next sub-cycle;

[0075] (6) Fitting completed: Repeat the above process until all data are fitted, the algorithm stops, and the M-SEIYAQURD-H model with t∈{1,2,...,T} is output.

[0076] Prediction framework based on rolling perspective: To improve prediction accuracy, this section proposes a prediction framework based on rolling perspective for model prediction. The specific process is as follows:

[0077] (1) Input: Input the prediction time period length T p , the trained M-SEIYAQURD-H model and the fitting sub-period π;

[0078] (2) Initialization: Initialize the prediction starting point t s ′ and the value of the rolling forecast period π′;

[0079] (3) Divide into sub-periods: Calculate the end point t of the rolling forecast in each round e ′, if the remaining time period is less than the rolling forecast period π′, it is directly regarded as the last forecast period. Figure 7 As shown;

[0080] (4) Update the training set: Using the trained model, iteratively calculate the predicted value within the current rolling cycle, and incorporate the real observation data within the corresponding cycle into the training sample set;

[0081] (5) Fitting training: Update the fitting starting point to the πth period preceding the current prediction end point, and similarly fit the sub-period framework, train the data of this π-length time period and obtain the model;

[0082] (6) Update iteration: Update the starting point t of the next round of prediction s ′=t e ';

[0083] (7) Prediction completion: Repeat the above process until all period data are predicted, the algorithm stops, and outputs t∈{T,T+1,...,T+T p}’s prediction results.

[0084] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an infectious disease prediction method that considers trend learning and cross-regional transmission as described above.

[0085] In a specific embodiment, taking the United States as an example, the epidemic in most parts of the United States was in a stage of stable development in the time period around 2020-09-01. Therefore, the present invention uses 2020-04-12 to 2020-09-01 as the training set time period, and the prediction time period length is 30 days, that is, the number of infected people from 2020-09-02 to 2020-10-01 is predicted to explore the fitting effect of the M-SEIYAQURD-H model in the stable epidemic scenario, and the comparison of the effects of using static prediction and dynamic prediction respectively. The comparison of the effect data of the stable epidemic scenario is shown in Table 1 below.

[0086] Table 1

[0087]

[0088] The epidemic situation in most parts of the United States experienced a significant change in trend around January 15, 2021. Therefore, this patent uses the training period from April 12, 2020 to January 15, 2021, and a 30-day prediction period. Specifically, it predicts the number of infections from January 16, 2021 to February 15, 2021. This is to explore the M-SEIYAQURD-H model's performance in the context of sudden epidemic events, and to compare the performance of static and dynamic predictions. Table 2 shows a comparison of the performance data for sudden epidemic events.

[0089] Table 2

[0090]

[0091]

[0092] The above results demonstrate that the M-SEIYAQURD-H model performs well in fitting epidemic data from multiple regions. Comparing these results reveals that direct predictions are significantly affected by the starting point and future trends. Furthermore, the rolling forecast method effectively improves prediction accuracy, both in stable and sudden epidemic scenarios, and mitigates the impact of different scenario selection on prediction results.

[0093] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0094] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for predicting infectious diseases considering trend learning and cross-regional transmission, characterized in that: The following steps are involved: S1. Divide the region and infectious disease group types, obtain infectious disease infection data, define the data usage range, then define effect evaluation indicators and set infectious disease related parameters; S2. Dividing the infectious disease stage into an infectious disease stable scenario and an infectious disease mutation scenario based on the acquired infectious disease infection data; S3, establish the M-SEIYAQURD-H model based on the improved multi-band learning algorithm; S4. Obtain the static prediction results of the M-SEIYAQURD-H model in the fitting sub-period framework, and obtain the dynamic prediction results of the M-SEIYAQURD-H model in the rolling prediction framework, and compare the dynamic and static prediction results of each region respectively; S5. Perform static prediction of the M-SEIYAQURD-H model and dynamic prediction of the M-SEIYAQURD-H model in the infectious disease stable scenario and infectious disease mutation scenario respectively, and calculate the effect evaluation index to compare the effects.

2. The infectious disease prediction method considering trend learning and cross-regional transmission according to claim 1, characterized in that: In S1, the infectious disease group types include susceptible people, latent people, infected people, symptomatic infected people, asymptomatic infected people, people receiving isolation treatment, people with symptoms but not receiving isolation treatment, recovered people and deceased people.

3. The infectious disease prediction method considering trend learning and cross-regional transmission according to claim 1, characterized in that: In S1, the infectious disease related parameters include the geographical mixing coefficient h k,j : Among them, h0 represents the proportion of transmission contact within the region, and (1-h0) represents the proportion of transmission contact to other external regions; d k,j Represents the geographic coordinate Euclidean distance between region k and region j, 1 / d k,j To represent the possibility of individuals moving from region k to region j.

4. The infectious disease prediction method considering trend learning and cross-regional transmission according to claim 1, characterized in that: In S3, the static prediction results of the M-SEIYAQURD-H model are obtained under the fitting sub-period framework, specifically including: Input the training sample data of the full time period and the population size status of the initial prediction period; initialize the fitting starting point and sub-period values; divide the sub-period and use the MH algorithm to fit and train to obtain the M-SEIYAQURD-H model of the sub-period; update the fitting starting point and initial population size status of the next sub-period, repeat the above process until all data are fitted, and output the final M-SEIYAQURD model to achieve static prediction.

5. The infectious disease prediction method considering trend learning and cross-regional transmission according to claim 4, characterized in that: The sub-period division and the use of the MH algorithm for fitting and training to obtain the M-SEIYAQURD-H model of the sub-period specifically include: Input the training sample data of the time period to be fitted and the population status of the initial period of prediction; extract infectious disease related parameters from the random distribution, generate the initial parameter set and calculate the loss function; extract a new parameter set through the maximum likelihood ratio and calculate the likelihood ratio; update the parameter set according to the likelihood ratio, repeat the iteration until the upper limit of the number of iterations is reached, and output the final fitting parameter set.

6. The infectious disease prediction method considering trend learning and cross-regional transmission according to claim 4, characterized in that: In S3, the dynamic prediction results of the M-SEIYAQURD-H model obtained under the rolling prediction framework specifically include: Input the length of the prediction time period, the trained M-SEIYAQURD-H model, and the fitting sub-period; initialize the prediction starting point and the value of the rolling prediction period; divide the sub-period and use the trained M-SEIYAQURD-H model to iteratively calculate the predicted value within this rolling period; update the training set and fit the next sub-period, and repeat the above process until the prediction of all period data is completed.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements an infectious disease prediction method considering trend learning and cross-regional transmission as described in any one of claims 1 to 6.

Images