Method and system for dynamic integrated prediction of infectious diseases based on multi-source space-time

Abstract

This invention provides a multi-source spatiotemporal dynamic integrated prediction method and system for infectious diseases, belonging to the field of infectious disease prediction technology. It includes classifying and processing multi-source data, establishing a SPE IQDR model, performing monotonicity analysis on time series to identify multiple waves of transmission, marking the start date of a new wave of transmission, dynamically adjusting parameters according to the characteristics of each wave of transmission, and superimposing the prediction results to form a final prediction. This invention can adapt to the characteristics of multi-wave transmission of infectious diseases, dynamically adjust model parameters, improve prediction accuracy, and has important guiding significance for public health decision-making.

Description

Technical Field

[0001] This invention relates to infectious disease prediction technology, and more particularly to a method and system for dynamic integrated prediction of infectious diseases based on multi-source spatiotemporal data. Background Technology

[0002] Infectious disease prediction models are crucial tools for public health decision-making and the formulation of prevention and control strategies, and their accuracy directly impacts the effectiveness of epidemic prevention and control measures. Traditional infectious disease prediction methods are mainly based on the classic SIR (Susceptibility-Infection-Removal) model and its variants, which describe the transition process between different disease states in a population by constructing a system of differential equations. With the development of big data and artificial intelligence technologies, infectious disease prediction methods are gradually evolving towards multi-source data fusion and intelligent analysis, such as combining search engine query data, social media sentiment analysis, population flow data, and other information sources for comprehensive prediction.

[0003] Current infectious disease prediction technologies suffer from the following deficiencies and shortcomings: First, existing prediction models generally lack accurate estimation and inclusion of asymptomatic carriers, leading to significant discrepancies between predicted results and actual situations in the early stages of infectious diseases, particularly for diseases with long incubation periods and high asymptomatic infection rates, such as COVID-19-induced pneumonia and measles. Second, existing models typically employ static parameter settings, failing to effectively capture the dynamic characteristics of multiple outbreaks of infectious diseases. Their predictive performance is poor when multiple peaks occur, and they struggle to adapt to dynamic changes in transmission parameters caused by changes in prevention and control policies, seasonal factors, or viral mutations. Third, existing methods do not adequately integrate and utilize multi-source data, failing to fully combine information from different levels, such as confirmed cases, asymptomatic infections, and those who have progressed to confirmed cases, for dynamic parameter adjustments. This results in insufficient flexibility and adaptability in model predictions, making it difficult to accurately reflect the complex development of epidemics. Summary of the Invention

[0004] The embodiments of the present invention provide a method and system for dynamic integrated prediction of infectious diseases based on multi-source spatiotemporal data, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a method for dynamic integrated prediction of infectious diseases based on multi-source spatiotemporal data, comprising:

[0006] Multi-source data related to infectious diseases are classified into different levels, including the number of confirmed cases, the number of asymptomatic infections, and the number of cases that have progressed to confirmed cases. The time series of the cumulative number of confirmed cases is calculated using a data analysis library. The time series, the start time of the infectious disease, the data cutoff time, the region, and the population are saved to a configuration information file.

[0007] A SPEIQDR model is established based on the time series. Data series with cumulative confirmed cases greater than a preset confirmed case threshold are selected. The Runge-Kutta method is used to solve the SPEIQDR model to obtain the number series of each group. The basic parameters are obtained by least squares calculation of the number series of each group and the actual data.

[0008] Monotonicity analysis is performed on the time series. When the time series shows multiple monotonically increasing or monotonically decreasing trends, it is determined to be a multi-wave development and dynamic prediction is initiated. Minimum points are obtained through the minimum value judgment method. Minimum points with a date interval greater than a preset time threshold corresponding to adjacent minimum points are marked as the start date of a new wave of propagation.

[0009] Based on the start date of the new wave of propagation, the number of latent individuals, the response start time, and the propagation attenuation rate are quantitatively adjusted in combination with the characteristics of each wave of propagation. The model calculation results are then fitted with the actual data to dynamically correct the parameters, resulting in the corrected and optimized parameters.

[0010] The number of confirmed cases is predicted based on the modified tuning parameters. When dynamic prediction is enabled, the prediction results of each prediction are superimposed to obtain the final prediction result and saved.

[0011] A SPEIQDR model is established based on the time series. Data sequences with cumulative confirmed cases exceeding a preset confirmed case threshold are selected. The Runge-Kutta method is used to solve the SPEIQDR model to obtain the population sequences of each group. The basic parameters are obtained by least squares calculation of the population sequences and the actual data, including:

[0012] Based on the time series, a SPEIQDR model is established, dividing the total population into susceptible population, non-susceptible population, latent population, infected population, isolated population, deceased population, and recovered population. A set of dynamic equations is constructed based on the SPEIQD model.

[0013] Formula for establishing the SPEIQD model:

[0014]

[0015]

[0016] N = S + P + E + I + Q + R + D;

[0017] Wherein, coefficients α, β, γ, δ, λ and κ represent the protection rate, infection rate, reciprocal of the average incubation time, rate at which infected individuals enter isolation, time-dependent recovery rate and time-dependent mortality rate, respectively.

[0018] Model parameters are extracted from the set of dynamic equations and the initial range of parameter values ​​is determined based on epidemiological characteristics. The initial range of parameter values ​​is set according to a preset interval based on the epidemiological characteristics.

[0019] Based on the preset interval, the Runge-Kutta method is used to numerically solve the dynamic equations. The time step is set to a preset time unit. The evolutionary sequence of different populations over time is obtained by integral calculation. The evolutionary sequence is fitted with the time series to construct the mean square error objective function.

[0020] The mean squared error objective function is used as the fitness function to iteratively optimize the model parameters. The iteration stops when the fitness function value is less than a preset error threshold or when a preset number of iterations is reached, and the optimized model parameters are obtained.

[0021] The optimized model parameters are substituted into the dynamic equations and solved using the fourth-order Runge-Kutta method to obtain the infectious disease prediction results. The infectious disease prediction results are then converted into basic parameters through de-standardization.

[0022] Based on the preset interval, the Runge-Kutta method is used to numerically solve the dynamic equations. The time step is set to a preset time unit. The evolutionary sequence of different populations over time is obtained through integral calculation. The evolutionary sequence is then fitted to the time series to construct the mean squared error objective function, which includes:

[0023] Construct a state vector for the dynamic equation system, which includes state components for susceptible populations, non-susceptible populations, latent populations, infected populations, isolated populations, deceased populations, and recovered populations.

[0024] Based on the state vector, a fourth-order Runge-Kutta iterative scheme is set up. The time step is determined according to the characteristic time scale of the dynamic equation system. The time step is substituted into the fourth-order Runge-Kutta iterative scheme. The state vector is iteratively calculated to obtain the state vector value at each time point. The state vector values ​​are stored in chronological order to form a time evolution sequence.

[0025] The state vector values ​​in the time evolution sequence are compared with the actual observation sequence. The squared error between the predicted state vector value and the actual observation value at each time point is calculated, and a weighting coefficient is assigned to each time point.

[0026] The mean square error objective function is constructed based on the sum of squared errors and the weighting coefficients. The weighted mean square error objective function is the sum of squared weighted errors at each time point divided by the total number of time points.

[0027] Monotonicity analysis is performed on the time series. When the time series exhibits multiple monotonically increasing or decreasing trends, it is determined to be a multi-wave development, and dynamic prediction is initiated. Minimum points are obtained through a minimum value judgment method. Minimum points with a date interval greater than a preset time threshold corresponding to adjacent minimum points are marked as the start date of a new wave of propagation, including:

[0028] Obtain confirmed case data in a continuous time series, and perform a difference operation on the confirmed case data at adjacent time points to obtain a difference sequence;

[0029] A monotonicity indicator function is constructed based on the difference sequence. When the difference value is greater than a preset difference threshold, the monotonicity indicator function takes a value of positive one, and when the difference value is less than the preset difference threshold, the monotonicity indicator function takes a value of negative one.

[0030] When the time interval in which the monotonicity indicator function takes a continuous positive value, it constitutes an increasing interval; when the time interval in which the monotonicity indicator function takes a continuous negative value, it constitutes a decreasing interval.

[0031] Local minimum points are identified in the decreasing interval, where the number of confirmed cases corresponding to the local minimum point is less than the number of confirmed cases at two adjacent time points. The identified local minimum points form an initial set of minimum points.

[0032] Calculate the time interval between adjacent minimum points in the initial minimum point set. When the time interval is less than a preset time threshold, delete the next minimum point from the initial minimum point set to obtain the effective minimum point set.

[0033] The time points in the set of effective minimum points are marked as the wave start time points, and the time difference between two adjacent wave start time points is marked as the start date of the new wave propagation.

[0034] Based on the start date of the new wave of propagation, and combined with the characteristics of each wave of propagation, the number of latent individuals, the response start time, and the propagation attenuation rate are quantitatively adjusted. The model calculation results are then fitted with actual data to dynamically correct the parameters, resulting in the following optimized parameters:

[0035] Acquire historical epidemic transmission data and extract the actual number of infections, predicted number of infections, number of latent individuals, response time, and transmission attenuation rate of each wave of the epidemic as initial optimization parameters.

[0036] Based on the actual infection number sequence, identify the peak infection number and peak time of each wave of the epidemic, obtain the infection number and end time of each wave of the epidemic, and calculate the natural logarithm of the ratio of the peak infection number to the end infection number divided by the time difference between the end time and the peak time to obtain the actual decay rate.

[0037] The actual number of infected persons sequence is compared with the predicted number of infected persons sequence to obtain the prediction error at each time point. Based on the prediction error, the difference ratio between the actual number of latent persons and the predicted number of latent persons is calculated to obtain the latent person change rate. The time difference between the actual response time and the predicted response time is calculated to obtain the response time delay value.

[0038] The initial parameters are jointly corrected based on the rate of change of the number of lurkers, the response time delay value, and the actual attenuation rate. The product of the number of lurkers and the rate of change of the number of lurkers plus one is used as the corrected number of lurkers. The sum of the response time and the response time delay value is used as the corrected response time. The sum of the differences between the propagation attenuation rate and the actual attenuation rate is used as the corrected tuning parameters.

[0039] The corrected optimization parameters are obtained by using the corrected number of latent agents, the corrected response time, and the corrected decay rate.

[0040] Based on the revised optimization parameters, the number of confirmed cases is predicted. When dynamic prediction is enabled, the results of each prediction are summed to obtain the final prediction result, which is then saved.

[0041] Construct a model parameter vector, combine the optimized model parameters into a parameter set, divide the prediction time series, and combine the prediction time points into a time window series;

[0042] Based on the parameter set, the first prediction calculation is performed within the time window sequence to obtain the first predicted sequence of confirmed cases.

[0043] Calculate the mean square error between the first predicted sequence of confirmed cases and the actual sequence of confirmed cases, compare the product of the mean square error and the step size adjustment coefficient with the preset maximum step size, and select the smaller value as the dynamic prediction step size.

[0044] The parameter set is adjusted according to the dynamic prediction step size to obtain a new parameter set. The new parameter set is then used for the next prediction calculation to obtain a new sequence of predicted confirmed cases.

[0045] Calculate the new mean square error between the new predicted confirmed case number sequence and the actual confirmed case number sequence, calculate the prediction weight based on the new mean square error, and use the ratio of the exponential function of the prediction weight to the sum of the exponential functions of the weights corresponding to all prediction times as the normalized prediction weight.

[0046] The sequence of predicted confirmed cases is multiplied by the normalized prediction weights and summed to obtain the fused sequence of predicted confirmed cases.

[0047] The historical fusion prediction sequence of confirmed cases is merged with the fusion prediction sequence of confirmed cases to obtain the final prediction result.

[0048] A second aspect of the present invention provides a multi-source spatiotemporal dynamic integrated prediction system for infectious diseases, comprising:

[0049] The first unit is used to classify multi-source data related to infectious diseases according to different levels. The data at different levels include the number of confirmed cases, the number of asymptomatic infections, and the number of confirmed cases that have progressed to confirmed cases. The time series of the cumulative number of confirmed cases is calculated using the data analysis library. The time series, the start time of the infectious disease, the data cutoff time, the region, and the population are saved to the configuration information file.

[0050] The second unit is used to establish a SPEIQDR model based on the time series, select data series with cumulative confirmed cases greater than a preset confirmed case threshold, use the Runge-Kutta method to solve the SPEIQDR model to obtain the number series of each group, and perform least squares calculation on the number series of each group and the actual data to obtain the basic parameters.

[0051] The third unit is used to perform monotonicity analysis on the time series. When the time series shows multiple monotonically increasing or monotonically decreasing trends, it is determined to be a multi-wave development and dynamic prediction is initiated. The minimum value point is obtained through the minimum value judgment method. The minimum value point with a date interval greater than a preset time threshold corresponding to adjacent minimum value points is marked as the start date of the new wave of propagation.

[0052] The fourth unit is used to quantitatively adjust the number of latent individuals, response start time, and propagation attenuation rate based on the start date of the new wave of propagation and the characteristics of each wave of propagation. It also dynamically corrects the parameters by fitting the model calculation results with the actual data to obtain the corrected and optimized parameters.

[0053] The fifth unit is used to predict the number of confirmed cases based on the modified tuning parameters. When dynamic prediction is enabled, the prediction results of each prediction are superimposed to obtain the final prediction result and saved.

[0054] A third aspect of the present invention provides an electronic device, comprising:

[0055] processor;

[0056] Memory used to store processor-executable instructions;

[0057] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0058] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0059] The beneficial effects of this application are as follows:

[0060] The present invention provides a multi-source spatiotemporal dynamic integrated prediction method for infectious diseases. By introducing a multi-source data classification and processing mechanism, it achieves a comprehensive analysis of the number of confirmed cases, the number of asymptomatic infections, and the number of cases that have progressed to confirmed cases, thereby improving the data foundation integrity and reliability of the prediction model.

[0061] This invention employs the SPEIQDR model combined with the Runge-Kutta method to accurately model the transmission process of infectious diseases. By analyzing the monotonicity of time series data, it identifies the characteristics of multiple waves of transmission, automatically identifies different stages of the epidemic's development, flexibly adjusts prediction strategies, and effectively responds to the complex and ever-changing situation of infectious disease transmission.

[0062] This invention implements a dynamic parameter correction mechanism, which quantitatively adjusts key parameters based on the characteristics of each wave of propagation, and optimizes the model calculation results by fitting them with actual data. This significantly improves the accuracy and adaptability of predictions, providing more scientific decision support for epidemic prevention and control, and has important practical value. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating the multi-source spatiotemporal dynamic integrated prediction method for infectious diseases according to an embodiment of the present invention.

[0064] Figure 2 This is an overall flowchart of an embodiment of the present invention;

[0065] Figure 3 This is a schematic diagram of the program flow for automatically determining whether an epidemic is developing in multiple waves, according to an embodiment of the present invention.

[0066] Figures 4-5 This is a schematic diagram of the time-dimensional dynamic integrated prediction method according to an embodiment of the present invention;

[0067] Figure 6 This is a schematic diagram illustrating the prediction results of the MDEP method in six Chinese cities according to an embodiment of the present invention.

[0068] Figure 7 This is a schematic diagram illustrating the evaluation of prediction results using the MDEP method according to an embodiment of the present invention;

[0069] Figure 8 This is a schematic diagram comparing the prediction results of the TEP and MDEP methods for Shanghai and Beijing in an embodiment of the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0072] Figure 1 This is a flowchart illustrating the multi-source spatiotemporal dynamic integrated prediction method for infectious diseases according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes:

[0073] Multi-source data related to infectious diseases are classified into different levels, including the number of confirmed cases, the number of asymptomatic infections, and the number of cases that have progressed to confirmed cases. The time series of the cumulative number of confirmed cases is calculated using a data analysis library. The time series, the start time of the infectious disease, the data cutoff time, the region, and the population are saved to a configuration information file.

[0074] A SPEIQDR model is established based on the time series. Data series with cumulative confirmed cases greater than a preset confirmed case threshold are selected. The Runge-Kutta method is used to solve the SPEIQDR model to obtain the number series of each group. The basic parameters are obtained by least squares calculation of the number series of each group and the actual data.

[0075] Monotonicity analysis is performed on the time series. When the time series shows multiple monotonically increasing or monotonically decreasing trends, it is determined to be a multi-wave development and dynamic prediction is initiated. Minimum points are obtained through the minimum value judgment method. Minimum points with a date interval greater than a preset time threshold corresponding to adjacent minimum points are marked as the start date of a new wave of propagation.

[0076] Based on the start date of the new wave of propagation, the number of latent individuals, the response start time, and the propagation attenuation rate are quantitatively adjusted in combination with the characteristics of each wave of propagation. The model calculation results are then fitted with the actual data to dynamically correct the parameters, resulting in the corrected and optimized parameters.

[0077] The number of confirmed cases is predicted based on the modified tuning parameters. When dynamic prediction is enabled, the prediction results of each prediction are superimposed to obtain the final prediction result and saved.

[0078] In one optional implementation, a SPEIQDR model is established based on the time series. Data sequences with cumulative confirmed cases exceeding a preset confirmed case threshold are selected. The Runge-Kutta method is used to solve the SPEIQDR model to obtain the population sequence for each group. The basic parameters are obtained by performing least squares calculations on the population sequence and the actual data, including:

[0079] Based on the time series, a SPEIQDR model is established, dividing the total population into susceptible population, non-susceptible population, incubation period population, infected population, quarantined population, deceased population, and recovered population. A set of dynamic equations is constructed based on the SPEIQDR model.

[0080] Formula for establishing the SPEIQDR model:

[0081]

[0082] N = S + P + E + I + Q + R + D;

[0083] Wherein, coefficients α, β, γ, δ, λ and κ represent the protection rate, infection rate, reciprocal of the average incubation time, rate at which infected individuals enter isolation, time-dependent recovery rate and time-dependent mortality rate, respectively.

[0084] Model parameters are extracted from the set of dynamic equations and the initial range of parameter values ​​is determined based on epidemiological characteristics. The initial range of parameter values ​​is set according to a preset interval based on the epidemiological characteristics.

[0085] Based on the preset interval, the Runge-Kutta method is used to numerically solve the dynamic equations. The time step is set to a preset time unit. The evolutionary sequence of different populations over time is obtained by integral calculation. The evolutionary sequence is fitted with the time series to construct the mean square error objective function.

[0086] The mean squared error objective function is used as the fitness function to iteratively optimize the model parameters. The iteration stops when the fitness function value is less than a preset error threshold or when a preset number of iterations is reached, and the optimized model parameters are obtained.

[0087] The optimized model parameters are substituted into the dynamic equations and solved using the fourth-order Runge-Kutta method to obtain the infectious disease prediction results. The infectious disease prediction results are then converted into basic parameters through de-standardization.

[0088] A time-series model, SPEIQDR, is established, which divides the total population into six categories: susceptible population (S), unsustainable population (P), latent population (E), infected population (I), isolated population (Q), deceased population (D), and recovered population (R). Based on the characteristics of infectious disease transmission, a set of dynamic equations is constructed to describe the changes in the number of each category of population over time.

[0089] In the SPEIQDR model, parameter α represents the protection rate, which is the rate at which a susceptible population becomes a non-susceptible population; β is the infection rate, which represents the probability that a susceptible population will become infected after contact with an infected person; γ is the reciprocal of the average incubation time, which represents the rate at which a population in the incubation period becomes an infected population; δ is the rate at which an infected person enters isolation; λ is the time-dependent recovery rate; and κ is the time-dependent mortality rate.

[0090] Collect infectious disease outbreak data, including time-series data such as daily new confirmed cases, cumulative confirmed cases, cured cases, and deaths. Select data points where the cumulative number of confirmed cases exceeds a preset threshold (e.g., 100 cases) to begin modeling analysis, in order to reduce the impact of early data fluctuations on model fitting.

[0091] Taking the COVID-19 pneumonia epidemic as an example, data from the outbreak of the epidemic in a certain region was collected for 90 days. The total population was 1 million. Initially, the susceptible population S(0) = 999900, the unsustainable population P(0) = 0, the incubation period population E(0) = 50, the infected population I(0) = 30, the quarantined population Q(0) = 20, the deceased population D(0) = 0, and the recovered population R(0) = 0.

[0092] The initial range of model parameters was determined based on epidemiological characteristics: the protection rate α ranged from [0.01, 0.1], indicating that 1%-10% of susceptible individuals took protective measures each day; the infection rate β ranged from [0.2, 0.7], calculated based on the basic reproduction number R0 of infectious diseases; the incubation period parameter γ ranged from [1 / 14, 1 / 3], corresponding to an incubation period of 3-14 days; the isolation rate δ ranged from [0.1, 0.5], indicating the speed at which infected individuals were discovered and isolated; the recovery rate λ ranged from [0.01, 0.2]; and the mortality rate κ ranged from [0.001, 0.05].

[0093] The fourth-order Runge-Kutta method was used to solve the dynamic equations. This method is a classic approach in numerical solutions of ordinary differential equations and has high accuracy. The time step was set to one day, and the number of people in each population group at each time point was calculated iteratively.

[0094] Taking the COVID-19 pandemic as an example, according to the Runge-Kutta algorithm, for each time point t, the number of people in each group at time t+1 is obtained through four intermediate calculations. For example, when calculating the number of susceptible people S, the rate of change k1 in the current state is first obtained, then k2 is obtained based on k1, and so on to obtain k3 and k4. Finally, the value of S at the next time point is determined by a weighted average. Other group categories are calculated using the same method.

[0095] The calculated population size sequences for each group are compared with actual observation data to construct a mean squared error objective function. This function measures the difference between the model's predicted values ​​and the actual observed values, calculating the sum of the squared differences between the actual number of isolated people and the model's predicted number of isolated people, which serves as the optimization objective.

[0096] The particle swarm optimization algorithm is used to iteratively optimize the model parameters. First, the positions of the parameter particles are randomly initialized, and the position vector contains all the parameters to be optimized. The number of particles is set to 50, and the maximum number of iterations is set to 1000. During the optimization process, the position and velocity of each particle are continuously updated until the mean squared error is less than a preset threshold of 0.001 or the maximum number of iterations is reached.

[0097] Taking the COVID-19 pandemic as an example, the optimal parameter values ​​were obtained through parameter optimization: protection rate α = 0.056, infection rate β = 0.42, incubation period parameter γ = 0.14 (corresponding to an average incubation period of approximately 7 days), isolation rate δ = 0.31, recovery rate λ = 0.068, and mortality rate κ = 0.012. This means that approximately 5.6% of susceptible individuals become non-susceptible each day, infected individuals have a 31% probability of being isolated each day, and among those isolated, 6.8% recover and 1.2% die each day.

[0098] The optimized parameters are substituted into the dynamic equations, and the Runge-Kutta method is used to solve for the number of various population groups at future time points. For example, predicting the epidemic trend over the next 30 days yields key indicators such as daily new confirmed cases, cumulative confirmed cases, active cases, and deaths.

[0099] The predicted results are converted into actual values ​​through destandardization. For example, the predicted results show that there will be 120 new confirmed cases on day 100, bringing the cumulative total to 3,500, with 850 active cases and 42 cumulative deaths.

[0100] In one optional implementation, the Runge-Kutta method is used to numerically solve the dynamic equations based on the preset interval. The time step is set to a preset time unit, and the evolutionary sequence of different populations over time is obtained through integral calculation. The evolutionary sequence is then fitted to the time series to construct a mean squared error objective function, including:

[0101] Construct a state vector for the dynamic equation system, which includes state components for susceptible populations, non-susceptible populations, latent populations, infected populations, isolated populations, deceased populations, and recovered populations.

[0102] Based on the state vector, a fourth-order Runge-Kutta iterative scheme is set up. The time step is determined according to the characteristic time scale of the dynamic equation system. The time step is substituted into the fourth-order Runge-Kutta iterative scheme. The state vector is iteratively calculated to obtain the state vector value at each time point. The state vector values ​​are stored in chronological order to form a time evolution sequence.

[0103] The state vector values ​​in the time evolution sequence are compared with the actual observation sequence. The squared error between the predicted state vector value and the actual observation value at each time point is calculated, and a weighting coefficient is assigned to each time point.

[0104] The mean square error objective function is constructed based on the sum of squared errors and the weighting coefficients. The weighted mean square error objective function is the sum of squared weighted errors at each time point divided by the total number of time points.

[0105] A state vector needs to be constructed from the equations governing the dynamics of infectious diseases. This state vector contains seven components: S (susceptible population), P (unsusceptible population), E (incubation period population), I (infected population), Q (isolated population), D (deceased population), and R (recovered population). These seven components together constitute a complete state vector, used to describe the dynamics of infectious disease transmission within the population.

[0106] Let the state vector be X, which contains the seven components mentioned above, each corresponding to the proportion of each group in the total population. For example, in a city with a total population of 1 million, the initial state vector might be: S = 0.95 (950,000 people are susceptible), U = 0.005 (5,000 people are not susceptible), E = 0.02 (20,000 people are in the incubation period), I = 0.01 (10,000 people are infected), Q = 0.005 (5,000 people are quarantined), D = 0.005 (5,000 people have died), and R = 0.005 (5,000 people have recovered).

[0107] A fourth-order Runge-Kutta iterative scheme is set up based on the constructed state vector. The Runge-Kutta method is a numerical method for solving ordinary differential equations, and this embodiment adopts the classic fourth-order Runge-Kutta method. An appropriate time step h is determined according to the time scale of the infectious disease dynamics. In this embodiment, assuming the average incubation period of the infectious disease is 7 days, a time step h = 0.5 days can be chosen as a suitable unit for numerical calculation.

[0108] The specific implementation steps of the fourth-order Runge-Kutta iteration are as follows: For the state vector X(t) at the current time t, first calculate four increments k1, k2, k3, and k4. Here, k1 represents the rate of change of the state vector in the current state; k2 represents the rate of change at state X(t) + 0.5h·k1; k3 represents the rate of change at state X(t) + 0.5h·k2; and k4 represents the rate of change at state X(t) + h·k3. The calculation of these four increments is based on the derivatives of the dynamic equations with respect to the state vector.

[0109] After calculating these four increments, the final change in the state vector is obtained by weighted averaging, i.e., X(t+h) = X(t) + (k1+2k2+2k3+k4) / 6·h. In this way, starting from the initial time t0, the state vector values ​​at times t0+h, t0+2h, ..., t0+Nh are calculated sequentially, and these state vectors are stored in chronological order to form a complete time evolution sequence.

[0110] For example, assuming the initial state is the example value above, the infection rate β = 0.3, the average rate of conversion from incubation period to infection α = 0.14 (corresponding to an average incubation period of 7 days), the isolation rate γ = 0.2, the recovery rate δ = 0.1, and the calculation is performed using a step size h = 0.5 days.

[0111] After obtaining the time evolution sequence, it is necessary to compare it with the actual observation sequence, calculate the error, and construct the objective function.

[0112] To construct the mean squared error objective function, it is necessary to calculate the squared error between the predicted value and the actual observed value at each observation time point. In this embodiment, the focus is on the total number of infected people, so the error is calculated as: (predicted I+Q value - actual observed I+Q value)².

[0113] Considering that the importance of data may vary at different time points, a weight coefficient is assigned to each time point. In this embodiment, the weight can be set according to the reliability or importance of the data. For example, the weight of the most recent time point is set higher: the weight of day 5 is w1 = 0.1, the weight of day 10 is w2 = 0.2, the weight of day 15 is w3 = 0.3, and the weight of day 20 is w4 = 0.4.

[0114] The final weighted mean squared error objective function is calculated as the sum of the weighted squared errors at each time point divided by the total number of time points:

[0115] (0.1×0.0000025+0.2×0.0001+0.3×0.00000025+0.4×0.00000025)÷4=0.000005075;

[0116] By optimizing the model parameters to minimize this objective function, the error between the model's predictions and actual observations can be minimized, thereby improving the model's fitting accuracy and predictive ability. This numerical solution scheme based on the Runge-Kutta method can effectively handle the complex nonlinear characteristics of the infectious disease dynamics equations, providing a scientific basis for predicting infectious disease epidemic trends.

[0117] In one optional implementation, monotonicity analysis is performed on the time series. When the time series exhibits multiple monotonically increasing or decreasing trends, it is determined to be a multi-wave development, and dynamic prediction is initiated. Minimum points are obtained using a minimum value determination method. Minimum points with a date interval greater than a preset time threshold corresponding to adjacent minimum points are marked as the start date of a new wave of propagation, including:

[0118] Obtain confirmed case data in a continuous time series, and perform a difference operation on the confirmed case data at adjacent time points to obtain a difference sequence;

[0119] A monotonicity indicator function is constructed based on the difference sequence. When the difference value is greater than a preset difference threshold, the monotonicity indicator function takes a value of positive one, and when the difference value is less than the preset difference threshold, the monotonicity indicator function takes a value of negative one.

[0120] When the time interval in which the monotonicity indicator function takes a continuous positive value, it constitutes an increasing interval; when the time interval in which the monotonicity indicator function takes a continuous negative value, it constitutes a decreasing interval.

[0121] Local minimum points are identified in the decreasing interval, where the number of confirmed cases corresponding to the local minimum point is less than the number of confirmed cases at two adjacent time points. The identified local minimum points form an initial set of minimum points.

[0122] Calculate the time interval between adjacent minimum points in the initial minimum point set. When the time interval is less than a preset time threshold, delete the next minimum point from the initial minimum point set to obtain the effective minimum point set.

[0123] The time points in the set of effective minimum points are marked as the wave start time points, and the time difference between two adjacent wave start time points is marked as the start date of the new wave propagation.

[0124] Obtain continuous time series data of confirmed COVID-19 cases. Taking the daily new confirmed COVID-19 cases in a certain region from January 2022 to June 2022 as an example, collect 180 days of confirmed case data to form a time series D = {d1, d2, ..., d...} 180}, where d_i represents the number of newly confirmed cases on day i. For example, the data for the previous 10 days might be {120,145,168,190,210,225,236,242,233,218}.

[0125] Perform difference operations on adjacent time points in the acquired confirmed case data. For each time point i (i>1), calculate the difference value Δd. i =d i -d i-1 The difference sequence ΔD = {Δd2, Δd3, ..., Δd} is obtained. 180 Taking the aforementioned 10 days of data as an example, the first 9 terms of the difference sequence are {25,23,22,20,15,11,6,-9,-15}.

[0126] A monotonicity indicator function I is constructed based on the difference sequence. A preset difference threshold δ is set, for example, δ = 5. When Δd i When Δd > δ, I(i) = 1, indicating that the epidemic shows a significant growth trend on day i; when Δd i When |Δd| < -δ, I(i) = -1, indicating that the epidemic shows a significant downward trend on day i; when |Δd| < -δ, ... i When |≤δ, I(i)=0, indicating that the epidemic is relatively stable. For the above difference sequence, the corresponding monotonicity indicator function value is {1,1,1,1,1,1,1,1,-1,-1,-1}.

[0127] Based on the monotonicity indicator function, the increasing and decreasing intervals of the time series are identified. An increasing interval consists of consecutive I(i) = 1, and a decreasing interval consists of consecutive I(i) = -1. In the example above, the interval from day 2 to day 8 is an increasing interval, and days 9 and 10 form a decreasing interval.

[0128] Identify local minima within the decreasing interval. For time point j, if d j <d j-1 And d j <d j+1 If j is a local minimum, then j is a local minimum point. Traverse all decreasing intervals, identify all local minimum points that satisfy the conditions, and form an initial set of minimum points M_init.

[0129] Suppose that through the above steps, five local minima are identified in 180 days of data, with corresponding time points {30, 45, 85, 105, 160}, and corresponding confirmed cases {50, 65, 30, 45, 25}.

[0130] Calculate the time intervals between adjacent minimum points in the initial set of minimum points. Let the preset time threshold be T days, for example, T = 30. For adjacent minimum points j and k, if |k - j| < T, it is considered that the waves corresponding to these two minimum points may not be independent waves and need to be screened. According to the screening rule, if the time interval between two adjacent minimum points is less than the preset time threshold, the latter minimum point is deleted from the initial set of minimum points.

[0131] The time intervals corresponding to the minimum points are 15 days (45 - 30), 40 days (85 - 45), 20 days (105 - 85), and 55 days (160 - 105) respectively. According to the screening rule, among the adjacent minimum points with a time interval of 15 days, the latter minimum point, that is, the 45th day, is deleted; similarly, among the adjacent minimum points with a time interval of 20 days, the minimum point on the 105th day is deleted. After screening, the effective set of minimum points M_valid = {30, 85, 160} is obtained.

[0132] Mark the time points in the effective set of minimum points as the starting time points of the waves. In this example, the 30th day, the 85th day, and the 160th day are respectively marked as the starting times of the three waves of the epidemic. The time differences between adjacent starting time points of the waves are 55 days (85 - 30) and 75 days (160 - 85) respectively, and these time differences are marked as the durations of new waves of transmission.

[0133] Based on the identified multi - wave characteristics, a dynamic prediction model can be further activated. Separate prediction models are established for each wave of the epidemic, such as the SEIR model or the time - series model, to predict the future development trend of the epidemic. For example, for the third wave of the epidemic (starting from the 160th day), based on the development laws of the first two waves of the epidemic, it can be predicted that its peak may appear around the 190th day, and the peak number of confirmed cases is about 350 cases.

[0134] When new epidemic data is monitored and shows new minimum points, the system automatically identifies it as the start of a new wave of the epidemic and activates the corresponding prediction mechanism to achieve dynamic identification and prediction of multi - wave epidemics.

[0135] In an optional implementation manner, according to the start date of the new wave of transmission, combined with the characteristics of each wave of transmission, the number of latent individuals, the start time of response, and the transmission decay rate are quantitatively adjusted, and the model calculation results are fitted with the actual data for dynamic parameter correction. The corrected optimized parameters include:

[0136] Obtain the historical epidemic transmission data of each wave, and extract the actual number of infected people sequence, the predicted number of infected people sequence, the number of latent individuals, the response time, and the transmission decay rate of each wave of epidemic transmission as the initial optimized parameters;

[0137] Based on the actual infection number sequence, identify the peak infection number and peak time of each wave of the epidemic, obtain the infection number and end time of each wave of the epidemic, and calculate the natural logarithm of the ratio of the peak infection number to the end infection number divided by the time difference between the end time and the peak time to obtain the actual decay rate.

[0138] The actual number of infected persons sequence is compared with the predicted number of infected persons sequence to obtain the prediction error at each time point. Based on the prediction error, the difference ratio between the actual number of latent persons and the predicted number of latent persons is calculated to obtain the latent person change rate. The time difference between the actual response time and the predicted response time is calculated to obtain the response time delay value.

[0139] The initial parameters are jointly corrected based on the rate of change of the number of lurkers, the response time delay value, and the actual attenuation rate. The product of the number of lurkers and the rate of change of the number of lurkers plus one is used as the corrected number of lurkers. The sum of the response time and the response time delay value is used as the corrected response time. The sum of the differences between the propagation attenuation rate and the actual attenuation rate is used as the corrected tuning parameters.

[0140] The corrected optimization parameters are obtained by using the corrected number of latent agents, the corrected response time, and the corrected decay rate.

[0141] Data on the spread of multiple waves of the COVID-19 pandemic were obtained from historical databases. Taking three waves of the COVID-19 pandemic in a certain city as an example, the actual number of infections, the predicted number of infections, and the initial optimization parameters for each wave were obtained. The initial parameters for the first wave were: 120 latent cases, response start time on day 5, and a transmission attenuation rate of 0.15; the initial parameters for the second wave were: 85 latent cases, response start time on day 3, and a transmission attenuation rate of 0.18; the initial parameters for the third wave were: 160 latent cases, response start time on day 4, and a transmission attenuation rate of 0.12.

[0142] The model identifies key features of each wave of the epidemic based on the actual number of infections. Taking the first wave as an example, actual data shows a peak of 892 infections on day 18; by the end of the epidemic (day 45), the number of infections had dropped to 37. To calculate the actual decay rate, the natural logarithm of the ratio of peak to end infection is first calculated: ln(892 / 37) = 3.18, then divided by the time difference (45-18) = 27 days, yielding an actual decay rate of 0.118. The initial model used a decay rate of 0.15, a difference of 0.032. Similarly, the calculated actual decay rate for the second wave was 0.163 (initially 0.18, difference 0.017), and for the third wave it was 0.135 (initially 0.12, difference -0.015).

[0143] The prediction error was calculated and the latent population change rate was analyzed. For the first wave of the epidemic, the actual number of infections each day was compared with the predicted number of infections. It was found that the predicted values ​​were generally lower before the peak, indicating that the actual number of latent individuals may be higher than the model's prediction. In the specific calculation, the average prediction error during the prediction period was taken as 32%, indicating that the difference between the actual number of latent individuals and the predicted value was 0.32, that is, the latent population change rate was 0.32. The analysis of the second wave of the epidemic showed that the latent population change rate was -0.15 (prediction was too high), and for the third wave it was 0.25 (prediction was too low).

[0144] Analyze the response time delay. By analyzing the difference between the actual response time and the predicted time, the actual response time for the first wave of the epidemic was 2 days later than predicted (predicted day 5, actual day 7), i.e., the response time delay was 2 days. The response time delay for the second wave of the epidemic was -1 day (predicted lag), and for the third wave it was 1 day (predicted advance).

[0145] Based on the above analysis results, the system performs joint corrections to the initial parameters. Taking the first wave of the epidemic as an example, the corrected number of latent cases is the product of the initial value and the rate of change: 120 × (1 + 0.32) = 158.4 people, rounded down to 158 people. The corrected response time is the initial response time plus a delay value: 5 + 2 = 7 days. The corrected decay rate is the adjustment of the difference between the initial decay rate and the actual decay rate: 0.15 - (0.15 - 0.118) = 0.118.

[0146] For the second wave of the epidemic, the revised number of latent cases is 85×(1-0.15)=72 people, the revised response time is 3-1=2 days, and the revised decay rate is 0.18-(0.18-0.163)=0.163.

[0147] For the third wave of the epidemic, the revised number of latent individuals is 160×(1+0.25)=200, the revised response time is 4+1=5 days, and the revised decay rate is 0.12-(0.12-0.135)=0.135.

[0148] The corrected parameters were re-input into the model for validation testing. Taking the first wave of the epidemic as an example, the corrected parameters (158 latent cases, 7-day response time, and a decay rate of 0.118) were used to re-predict the number of infections. The average error between the model and the actual number of infections decreased from 32% to 8.7%, and the peak prediction accuracy increased from 86% to 97.2%, demonstrating the effectiveness of the parameter correction.

[0149] The second wave of the epidemic was re-predicted using revised parameters (72 latent cases, 2-day response time, and a decay rate of 0.163), with the average error decreasing from 27% to 9.3%. The third wave of the epidemic was re-predicted using revised parameters (200 latent cases, 5-day response time, and a decay rate of 0.135), with the average error decreasing from 29% to 11.5%.

[0150] The corrected parameters are used as initial parameters for predicting the spread of the next wave of the epidemic, enabling dynamic model optimization and improving prediction accuracy. Furthermore, the system establishes a parameter correction rule base based on factors such as epidemic type and transmission environment, providing guidance for parameter adjustments in different scenarios and further enhancing the model's adaptability and predictive performance.

[0151] In one optional implementation, the number of confirmed cases is predicted based on the modified tuning parameters. When dynamic prediction is enabled, the final prediction result is obtained by summing up each prediction result and saving it, including:

[0152] Construct a model parameter vector, combine the optimized model parameters into a parameter set, divide the prediction time series, and combine the prediction time points into a time window series;

[0153] Based on the parameter set, the first prediction calculation is performed within the time window sequence to obtain the first predicted sequence of confirmed cases.

[0154] Calculate the mean square error between the first predicted sequence of confirmed cases and the actual sequence of confirmed cases, compare the product of the mean square error and the step size adjustment coefficient with the preset maximum step size, and select the smaller value as the dynamic prediction step size.

[0155] The parameter set is adjusted according to the dynamic prediction step size to obtain a new parameter set. The new parameter set is then used for the next prediction calculation to obtain a new sequence of predicted confirmed cases.

[0156] Calculate the new mean square error between the new predicted confirmed case number sequence and the actual confirmed case number sequence, calculate the prediction weight based on the new mean square error, and use the ratio of the exponential function of the prediction weight to the sum of the exponential functions of the weights corresponding to all prediction times as the normalized prediction weight.

[0157] The sequence of predicted confirmed cases is multiplied by the normalized prediction weights and summed to obtain the fused sequence of predicted confirmed cases.

[0158] The historical fusion prediction sequence of confirmed cases is merged with the fusion prediction sequence of confirmed cases to obtain the final prediction result.

[0159] The model parameter vector is constructed and the prediction time series is divided. In this embodiment, it is assumed that the corrected parameters include the infection rate β = 0.32, the recovery rate γ = 0.045, and the isolation rate α = 0.13, which form the parameter set P = {0.32, 0.045, 0.13}. The prediction time window is set to 14 days, i.e., the time window sequence T = {t1, t2, ..., t...}. 14}, where t1 represents the prediction start date, such as January 1, 2022.

[0160] The first prediction calculation is performed based on the parameter set P within the time window sequence T. Specifically, by substituting the parameter set P into the SEIR infectious disease model, the sequence of confirmed cases predicted for the first time is calculated as C1 = {120, 135, 152, 170, 188, 207, 224, 240, 255, 268, 280, 290, 298, 305}.

[0161] Calculate the mean squared error (MSE) between the first predicted confirmed case number sequence C1 and the actual confirmed case number sequence A = {118, 132, 149, 165, 184, 200, 218, 232, 245, 260, 273, 285, 295, 302}. The calculated MSE1 for the first prediction is 25.36. Assume a step size adjustment coefficient k = 0.05 and a preset maximum step size S. ax =0.02, then the dynamic prediction step size S1 = min(25.36×0.05,0.02) = min(1.268,0.02) = 0.02.

[0162] The parameter set P is adjusted according to the dynamic prediction step size 1, resulting in a new parameter set P' = {0.3136, 0.0459, 0.1326}. This is achieved by adding a small random perturbation to each parameter and multiplying it by the step size coefficient. For example, for the infection rate β, the new value is calculated as 0.32 - (0.32 × 0.02) = 0.3136.

[0163] Using the new parameter set P', a second prediction calculation is performed to obtain a new predicted sequence of confirmed cases C2={118,132,150,168,185,203,220,236,250,263,275,286,295,302}.

[0164] The new mean squared error (MSE2) between the new predicted confirmed case number sequence C2 and the actual confirmed case number sequence A is calculated to be 9.21. Based on the new MSE2, the prediction weights are calculated using an exponential function: w2 = exp(-MSE2) = exp(-9.21) = 0.0001.

[0165] Assuming the initial prediction weights are w1 = exp(-MSE1) = exp(-25.36) = 0.000000001, then the normalized prediction weights are:

[0166] W1=w1 / (w1+w2)=0.000000001 / 0.0001000001=0.00001;

[0167] W2=w2 / (w1+w2)=0.0001 / 0.0001000001=0.99999.

[0168] The fusion predicted confirmed case count sequence F is obtained by multiplying each predicted confirmed case count sequence by the normalized prediction weights and summing the results. For example, the fusion predicted confirmed case count for the first day is: F1 = C 11 ×W1+C 21 ×W2=120×0.00001+118×0.99999=118.00002≈118. Following this pattern, the complete fusion prediction sequence of confirmed cases is obtained: F={118,132,150,168,185,203,220,236,250,263,275,286,295,302}.

[0169] If the historical prediction sequence is empty, the final prediction result is F; if there is a historical fusion prediction sequence of confirmed cases H = {105, 112}, then H and F are merged to obtain the final prediction result R = {105, 112, 118, 132, 150, 168, 185, 203, 220, 236, 250, 263, 275, 286, 295, 302}.

[0170] In practical applications, a third and fourth prediction can be made, with each prediction based on the results of the previous one, adjusting the parameters accordingly. For example, the third prediction uses the parameter set P" = {0.3073, 0.0468, 0.1353}, resulting in the prediction sequence C3 = {117, 131, 148, 166, 183, 201, 218, 234, 248, 261, 273, 284, 293, 300}, with a mean squared error MSE3 = 4.57 and a prediction weight w3 = exp(-4.57) = 0.0104.

[0171] Recalculate the normalized weights:

[0172] W1=w1 / (w1+w2+w3)=0.000000001 / (0.000000001+0.0001+0.0104)=0.0000000001;

[0173] W2=w2 / (w1+w2+w3)=0.0001 / (0.000000001+0.0001+0.0104)=0.0095;

[0174] W3=w3 / (w1+w2+w3)=0.0104 / (0.000000001+0.0001+0.0104)=0.9905.

[0175] The three prediction results are fused to obtain a new fused prediction sequence F' = {117,131,148,166,183,201,218,234,248,261,273,284,293,300}. The historical sequences are then merged to obtain the final prediction result R' = {105,112,117,131,148,166,183,201,218,234,248,261,273,284,293,300}.

[0176] The key advantage of this method lies in its ability to dynamically adjust model parameters based on real-time data and reduce the impact of single-prediction bias by fusing multiple prediction results through weighted fusion. Simultaneously, the adaptive step size adjustment mechanism makes parameter updates more flexible, allowing for rapid adjustments with larger step sizes during periods of rapid change in the epidemic, and fine-tuning with smaller step sizes during periods of stability, thus improving the model's adaptability to different stages.

[0177] like Figures 2-8 As shown, the method described in this embodiment further includes:

[0178] (1) Data collection and preprocessing

[0179] For the 12 cities in China where outbreaks occurred, epidemic data for each city was collected and saved to separate text files. Simultaneously, historical epidemic data for similar situations in each city were collected as inversion parameters.

[0180] (2) Historical data inversion parameters

[0181] For cities experiencing outbreaks within China, historical epidemic data similar to those of each city are searched. Given an initial parameter range, the SPEIQDR model introduced earlier is used, and the Runge-Kutta method is employed to solve the equation. By integrating, a sequence of the number of people in each population group can be obtained. The number of infected individuals in the integrated sequence is then compared with the actual number of infected individuals using least squares calculation. This process is repeated until the sum of squared residuals between the two sequences is minimized, making the predicted result most similar to the actual result, thus obtaining a set of basic parameters.

[0182] (3) Determine whether the epidemic is showing a multi-wave development trend.

[0183] according to Figure 3The flowchart shown utilizes the monotonicity of functions to determine whether the epidemic development in cities with outbreaks in China exhibits a multi-wave pattern. The epidemic trend within a given time period can be expressed by an appropriate function, and the monotonicity of the function is used to determine whether the epidemic development within that time period is a monotonically increasing or monotonically decreasing function. A sliding window period of 5 days is set, and the monotonicity of the epidemic data within each time period is checked. If it is a monotonically increasing function, it indicates that the epidemic is still in its upward phase, and the basic model can be used for prediction updates. When the epidemic development shows a monotonically decreasing characteristic, the time of the minimum point is further determined. To prevent data anomalies caused by detection or recording problems, additional judgment conditions are added: That is, the interval between two adjacent minimum values ​​needs to be greater than 5 days. Only when such a condition is met can it be judged that the epidemic has entered a new outbreak period.

[0184] (4) Multi-source spatiotemporal dynamic prediction method

[0185] When the development of the epidemic conforms to the characteristics of a multi-spread epidemic, and the accuracy of the epidemic data can be precise down to the district level, according to Figure 4 and Figure 5 The display allows for the overlay of prediction results along both the time and time-space dimensions to obtain the final dynamic ensemble result. This result is then compared with actual data to verify the model's accuracy.

[0186] A second aspect of the present invention provides a multi-source spatiotemporal dynamic integrated prediction system for infectious diseases, comprising:

[0187] The first unit is used to classify multi-source data related to infectious diseases according to different levels. The data at different levels include the number of confirmed cases, the number of asymptomatic infections, and the number of confirmed cases that have progressed to confirmed cases. The time series of the cumulative number of confirmed cases is calculated using the data analysis library. The time series, the start time of the infectious disease, the data cutoff time, the region, and the population are saved to the configuration information file.

[0188] The second unit is used to establish a SPEIQDR model based on the time series, select data series with cumulative confirmed cases greater than a preset confirmed case threshold, use the Runge-Kutta method to solve the SPEIQDR model to obtain the number series of each group, and perform least squares calculation on the number series of each group and the actual data to obtain the basic parameters.

[0189] The third unit is used to perform monotonicity analysis on the time series. When the time series shows multiple monotonically increasing or monotonically decreasing trends, it is determined to be a multi-wave development and dynamic prediction is initiated. The minimum value point is obtained through the minimum value judgment method. The minimum value point with a date interval greater than a preset time threshold corresponding to adjacent minimum value points is marked as the start date of the new wave of propagation.

[0190] The fourth unit is used to quantitatively adjust the number of latent individuals, response start time, and propagation attenuation rate based on the start date of the new wave of propagation and the characteristics of each wave of propagation. It also dynamically corrects the parameters by fitting the model calculation results with the actual data to obtain the corrected and optimized parameters.

[0191] The fifth unit is used to predict the number of confirmed cases based on the modified tuning parameters. When dynamic prediction is enabled, the prediction results of each prediction are superimposed to obtain the final prediction result and saved.

[0192] A third aspect of the present invention provides an electronic device, comprising:

[0193] processor;

[0194] Memory used to store processor-executable instructions;

[0195] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0196] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0197] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0198] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-source spatiotemporal dynamic integrated prediction method for infectious diseases, characterized in that, include: Multi-source data related to infectious diseases are classified into different levels, including the number of confirmed cases, the number of asymptomatic infections, and the number of cases that have progressed to confirmed cases. The time series of the cumulative number of confirmed cases is calculated using a data analysis library. The time series, the start time of the infectious disease, the data cutoff time, the region, and the population are saved to a configuration information file. A SPEIQDR model is established based on the time series. Data sequences with cumulative confirmed cases exceeding a preset confirmed case threshold are selected. The Runge-Kutta method is used to solve the SPEIQDR model to obtain the population sequences of each group. The population sequences of each group are then compared with the actual data using least squares calculations to obtain the basic parameters, including: Based on the time series, a SPEIQDR model is established, dividing the total population into susceptible population, non-susceptible population, latent population, infected population, isolated population, deceased population, and recovered population. A set of dynamic equations is constructed based on the SPEIQDR model. SEIQRP model establishment formula: ; ； ； ； ； ； ； ； Wherein, coefficients α, β, γ, δ, λ and κ represent the protection rate, infection rate, reciprocal of the average incubation time, rate at which infected persons enter isolation, time-dependent recovery rate and time-dependent mortality rate, respectively; susceptible population (S), non-susceptible population (P), incubation period population (E), infected population (I), isolated population (Q), deceased population (D) and recovered population (R). Model parameters are extracted from the set of dynamic equations and the initial range of parameter values ​​is determined based on epidemiological characteristics. The initial range of parameter values ​​is set according to a preset interval based on the epidemiological characteristics. Based on the preset interval, the Runge-Kutta method is used to numerically solve the dynamic equations. The time step is set to a preset time unit. The evolutionary sequence of different populations over time is obtained by integral calculation. The evolutionary sequence is fitted with the time series to construct the mean square error objective function. The mean squared error objective function is used as the fitness function to iteratively optimize the model parameters. The iteration stops when the fitness function value is less than a preset error threshold or when a preset number of iterations is reached, and the optimized model parameters are obtained. Substitute the optimized model parameters into the dynamic equations and solve them using the fourth-order Runge-Kutta method to obtain the infectious disease prediction results. Then, convert the infectious disease prediction results into basic parameters through de-standardization. Monotonicity analysis is performed on the time series. When the time series shows multiple monotonically increasing or monotonically decreasing trends, it is determined to be a multi-wave development and dynamic prediction is initiated. Minimum points are obtained through the minimum value judgment method. Minimum points with a date interval greater than a preset time threshold corresponding to adjacent minimum points are marked as the start date of a new wave of propagation. Based on the start date of the new wave of propagation, the number of latent individuals, the response start time, and the propagation attenuation rate are quantitatively adjusted in combination with the characteristics of each wave of propagation. The model calculation results are then fitted with the actual data to dynamically correct the parameters, resulting in the corrected and optimized parameters. The number of confirmed cases is predicted based on the modified tuning parameters. When dynamic prediction is enabled, the prediction results are superimposed to obtain the final prediction result and saved.

2. The method according to claim 1, characterized in that, Based on the preset interval, the Runge-Kutta method is used to numerically solve the dynamic equations. The time step is set to a preset time unit. The evolutionary sequence of different populations over time is obtained through integral calculation. The evolutionary sequence is then fitted to the time series to construct the mean squared error objective function, which includes: Construct a state vector for the dynamic equation system, which includes state components for susceptible populations, non-susceptible populations, latent populations, infected populations, isolated populations, deceased populations, and recovered populations. Based on the state vector, a fourth-order Runge-Kutta iterative scheme is set up. The time step is determined according to the characteristic time scale of the dynamic equation system. The time step is substituted into the fourth-order Runge-Kutta iterative scheme. The state vector is iteratively calculated to obtain the state vector value at each time point. The state vector values ​​are stored in chronological order to form a time evolution sequence. The state vector values ​​in the time evolution sequence are compared with the actual observation sequence. The squared error between the predicted state vector value and the actual observation value at each time point is calculated, and a weighting coefficient is assigned to each time point. The mean square error objective function is constructed based on the sum of squared errors and the weighting coefficients. The mean square error objective function is the weighted sum of squared errors at each time point divided by the total number of time points.

3. The method according to claim 1, characterized in that, Monotonicity analysis is performed on the time series. When the time series exhibits multiple monotonically increasing or decreasing trends, it is determined to be a multi-wave development, and dynamic prediction is initiated. Minimum points are obtained through a minimum value judgment method. Minimum points with a date interval greater than a preset time threshold corresponding to adjacent minimum points are marked as the start date of a new wave of propagation, including: Obtain confirmed case data in a continuous time series, and perform a difference operation on the confirmed case data at adjacent time points to obtain a difference sequence; A monotonicity indicator function is constructed based on the difference sequence. When the difference value is greater than a preset difference threshold, the monotonicity indicator function takes a value of positive one, and when the difference value is less than the preset difference threshold, the monotonicity indicator function takes a value of negative one. The time interval in which the monotonicity indicator function continuously takes a positive value constitutes an increasing interval, and the time interval in which the monotonicity indicator function continuously takes a negative value constitutes a decreasing interval. Local minimum points are identified in the decreasing interval, where the number of confirmed cases corresponding to the local minimum point is less than the number of confirmed cases at two adjacent time points. The identified local minimum points form an initial set of minimum points. Calculate the time interval between adjacent minimum points in the initial minimum point set. When the time interval is less than a preset time threshold, delete the next minimum point from the initial minimum point set to obtain the effective minimum point set. The time points in the set of effective minimum points are marked as the wave start time points, and the time difference between two adjacent wave start time points is marked as the start date of the new wave propagation.

4. The method according to claim 1, characterized in that, Based on the start date of the new wave of propagation, and combined with the characteristics of each wave of propagation, the number of latent individuals, the response start time, and the propagation attenuation rate are quantitatively adjusted. The model calculation results are then fitted with actual data to dynamically correct the parameters, resulting in the following optimized parameters: Acquire historical epidemic transmission data and extract the actual number of infections, predicted number of infections, number of latent individuals, response time, and transmission attenuation rate of each wave of the epidemic as initial optimization parameters. Based on the actual infection number sequence, identify the peak infection number and peak time of each wave of the epidemic, obtain the infection number and end time of each wave of the epidemic, and calculate the natural logarithm of the ratio of the peak infection number to the end infection number divided by the time difference between the end time and the peak time to obtain the actual decay rate. The actual number of infected persons sequence is compared with the predicted number of infected persons sequence to obtain the prediction error at each time point. Based on the prediction error, the difference ratio between the actual number of latent persons and the predicted number of latent persons is calculated to obtain the latent person change rate. The time difference between the actual response time and the predicted response time is calculated to obtain the response time delay value. The initial parameters are jointly corrected based on the rate of change of the number of lurkers, the response time delay value, and the actual attenuation rate. The product of the number of lurkers and the rate of change of the number of lurkers plus one is used as the corrected number of lurkers. The sum of the response time and the response time delay value is used as the corrected response time. The sum of the differences between the propagation attenuation rate and the actual attenuation rate is used as the corrected tuning parameters. The corrected optimization parameters are obtained by using the corrected number of latent agents, the corrected response time, and the corrected decay rate.

5. The method according to claim 1, characterized in that, Based on the revised optimization parameters, the number of confirmed cases is predicted. When dynamic prediction is enabled, the results of each prediction are summed to obtain the final prediction result, which is then saved. Construct a model parameter vector, combine the optimized model parameters into a parameter set, divide the prediction time series, and combine the prediction time points into a time window series; Based on the parameter set, the first prediction calculation is performed within the time window sequence to obtain the first predicted sequence of confirmed cases. Calculate the mean square error between the first predicted sequence of confirmed cases and the actual sequence of confirmed cases, compare the product of the mean square error and the step size adjustment coefficient with the preset maximum step size, and select the smaller value as the dynamic prediction step size. The parameter set is adjusted according to the dynamic prediction step size to obtain a new parameter set. The new parameter set is then used for the next prediction calculation to obtain a new sequence of predicted confirmed cases. Calculate the new mean square error between the new predicted confirmed case number sequence and the actual confirmed case number sequence, calculate the prediction weight based on the new mean square error, and use the ratio of the exponential function of the prediction weight to the sum of the exponential functions of the weights corresponding to all prediction times as the normalized prediction weight. The sequence of predicted confirmed cases is multiplied by the normalized prediction weights and summed to obtain the fused sequence of predicted confirmed cases. The historical fusion prediction sequence of confirmed cases is merged with the fusion prediction sequence of confirmed cases to obtain the final prediction result.

6. A multi-source spatiotemporal dynamic integrated prediction system for infectious diseases, used to implement the method of any one of claims 1-5, characterized in that, include: The first unit is used to classify multi-source data related to infectious diseases according to different levels. The data at different levels include the number of confirmed cases, the number of asymptomatic infections, and the number of confirmed cases that have progressed to confirmed cases. The time series of the cumulative number of confirmed cases is calculated using the data analysis library. The time series, the start time of the infectious disease, the data cutoff time, the region, and the population are saved to the configuration information file. The second unit is used to establish a SPEIQDR model based on the time series, select data series with cumulative confirmed cases greater than a preset confirmed case threshold, use the Runge-Kutta method to solve the SPEIQDR model to obtain the number series of each group, and perform least squares calculation on the number series of each group and the actual data to obtain the basic parameters. The third unit is used to perform monotonicity analysis on the time series. When the time series shows multiple monotonically increasing or monotonically decreasing trends, it is determined to be a multi-wave development and dynamic prediction is initiated. The minimum value point is obtained through the minimum value judgment method. The minimum value point with a date interval greater than a preset time threshold corresponding to adjacent minimum value points is marked as the start date of the new wave of propagation. The fourth unit is used to quantitatively adjust the number of latent individuals, response start time, and propagation attenuation rate based on the start date of the new wave of propagation and the characteristics of each wave of propagation. It also dynamically corrects the parameters by fitting the model calculation results with the actual data to obtain the corrected and optimized parameters. The fifth unit is used to predict the number of confirmed cases based on the modified tuning parameters. When dynamic prediction is enabled, the prediction results of each prediction are superimposed to obtain the final prediction result and saved.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images