Dengue fever prediction method and device considering extreme weather driving and medium

Abstract

The invention discloses a dengue prediction method and device considering extreme weather driving and a medium, and the method comprises the steps: obtaining case data and multi-source data of a target region, the multi-source data comprising meteorological data, environmental data and social economic data; according to the meteorological data and a preset extreme weather threshold value, constructing extreme weather characteristics; obtaining epidemic situation time-space lag data according to the spatial autocorrelation and the time autocorrelation of the case data; adding an epidemic situation time-space lag item into the zero-expansion negative binomial regression model to obtain an initial epidemic situation prediction model; the epidemic situation prediction model comprises quadratic terms of extreme weather characteristics; and performing parameter fitting on the initial epidemic situation prediction model based on case data, extreme weather characteristics, epidemic situation time-space lag data and multi-source data to obtain a fitted epidemic situation prediction model so as to predict future epidemic situations. According to the method, the credibility of regional dengue epidemic situation prediction under the climate change background can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of infectious disease prediction, and in particular to a dengue fever prediction method, device and medium taking extreme weather drive into consideration. Background Art

[0002] Dengue fever is a mosquito-borne infectious disease caused by the dengue virus. With global climate change, extreme weather events such as heat waves, extreme precipitation and meteorological drought have increased, and these changes have had a significant impact on the living environment of mosquitoes. Climate warming not only expands the distribution range and survival period of mosquitoes, but also causes the dengue fever outbreak area to expand from low altitudes to high altitudes, and from tropical regions to temperate regions, thereby increasing the risk of dengue fever transmission. Therefore, accurate prediction of dengue fever is of great significance for the prevention and control of infectious diseases.

[0003] Current dengue fever prediction technology focuses on meteorological, demographic and socioeconomic factors observed daily, while neglecting in-depth analysis of extreme weather and environmental factors. Existing prediction methods use technologies such as time series statistical analysis, machine learning, and deep learning, but they generally have problems such as weak explanatory power of model variables, insufficient consideration of the detailed characteristics of extreme weather and influencing factors within the city, making dengue fever prediction results less credible. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a dengue fever prediction method, device and medium taking into account extreme weather driving, which can effectively improve the credibility of regional dengue fever epidemic prediction under the background of climate change.

[0005] The embodiment of the present invention provides a dengue fever prediction method considering extreme weather driving, including:

[0006] Acquire case data and multi-source data of the target area, wherein the multi-source data includes meteorological data, environmental data and socio-economic data;

[0007] constructing extreme weather features based on the meteorological data and a preset extreme weather threshold;

[0008] According to the spatial autocorrelation and temporal autocorrelation of the case data, the spatiotemporal lag data of the epidemic are obtained;

[0009] Adding the spatiotemporal lag term of the epidemic situation to the zero-inflated negative binomial regression model to obtain an initial epidemic situation prediction model; the epidemic situation prediction model includes a quadratic term of the extreme weather characteristics;

[0010] Based on the case data, the extreme weather characteristics, the spatiotemporal lag data of the epidemic, and the multi-source data, parameter fitting is performed on the initial epidemic prediction model to obtain a fitted epidemic prediction model;

[0011] The fitted epidemic prediction model is used to predict the future epidemic situation in the target area.

[0012] As an improvement of the above solution, the step of obtaining case data of the target area includes:

[0013] Obtain unstructured address text from dengue case reports in the target area;

[0014] Mapping the unstructured address text into a standard address library to obtain the geographic coordinates corresponding to the case;

[0015] The target area is segmented into grids, and the number of cases in each grid unit is calculated based on the geographic coordinates to obtain gridded case data.

[0016] As an improvement of the above solution, the meteorological data includes but is not limited to daily maximum temperature, daily minimum temperature, daily precipitation and drought condition data;

[0017] The environmental data include but are not limited to air quality data, vegetation coverage and impervious surface ratio;

[0018] The socioeconomic data include but are not limited to mosquito density, population density, road network density, distribution of medical facilities and urban construction data.

[0019] As an improvement of the above solution, constructing extreme weather features according to the meteorological data and a preset extreme weather threshold includes:

[0020] Based on the historical meteorological data of the historical base period, the quantiles (such as the 90th or 95th quantiles) of the historical meteorological data are calculated using a sliding window method to determine the extreme weather threshold;

[0021] According to the extreme weather threshold, extreme weather characteristics are extracted from the meteorological data, wherein the extreme weather characteristics include the frequency, intensity and duration of each extreme weather event.

[0022] As an improvement of the above scheme, the spatiotemporal lag data of the epidemic situation is obtained according to the spatial autocorrelation relationship and the temporal autocorrelation relationship of the case data, including:

[0023] Based on the case data, a spatial weight matrix is ​​constructed by calculating the spatial correlation index value of each grid unit and the adjacent area, and an inverse distance weighted matrix is ​​obtained by combining the preset inverse distance weighted radius;

[0024] Calculating the spatial lag data of each grid cell according to the inverse distance weight matrix and the case data;

[0025] Performing least squares regression on the case data to obtain residuals, and calculating the DW statistic based on the residuals;

[0026] It is determined whether the case data has a first-order positive correlation based on the DW statistic. If so, the spatial lag data lagged by one period is used as the spatiotemporal lag data of the epidemic.

[0027] As an improvement of the above solution, the initial epidemic prediction model is parameter-fitted based on the case data, the extreme weather characteristics, the epidemic spatiotemporal lag data, and the multi-source data to obtain a fitted epidemic prediction model, including:

[0028] Taking the extreme weather characteristics, the spatiotemporal lag data of the epidemic, and the multi-source data as a first variable set, and constructing a basic regression model for each type of extreme weather according to the first variable set and the case data;

[0029] Based on the basic regression model, respectively calculating the variance inflation factor of each of the extreme weather characteristics, and selecting variables in the first variable set according to the variance inflation factor to obtain a second variable set;

[0030] Based on the second variable set, a penalty term is added to the basic regression model of each type of extreme weather according to the Lasso regression algorithm, and the variables whose contribution rate to the model is less than a preset threshold are eliminated by the stepwise regression method to obtain a third variable set;

[0031] According to the third variable set and the case data, the initial epidemic prediction model is fitted with parameters to obtain a fitted epidemic prediction model.

[0032] As an improvement of the above scheme, the epidemic prediction model includes a counting part and a zero-inflation part;

[0033] The expression of the counting part is:

[0034]

[0035] Where i represents the grid number, t represents the time, n represents the number of grids, log(λ it ) is the natural logarithm expectation of the number of dengue cases in the grid cell with base e, X it represents the observed value of extreme weather characteristics, α it The quadratic term of each extreme weather characteristic has an impact on the prevalence of dengue fever, β it represents the influence coefficient of extreme weather characteristics on the prevalence of dengue fever, w i y i(t-1) is the spatiotemporal lag data of the epidemic, ρ is the spatial effect coefficient of dengue fever, ε itRepresents the unobservable part and individual heterogeneity, the multi-source data is used as a control variable, CV iT represents the observed value of the control variable, γ it represents the influence coefficient of each control variable on the prevalence of dengue fever;

[0036] The observed values ​​of extreme weather characteristics include the observed values ​​of a single extreme weather characteristic and / or the cross-product terms of different extreme weather characteristics.

[0037] As an improvement of the above solution, the expression of the zero expansion part is:

[0038]

[0039] Among them, p it is the probability of zero expansion, τ i is the influence coefficient of the zero expansion factor, G i is the observed value of the zero inflation factor, and τ0 is a constant term; the zero inflation factor includes the night light intensity in the multi-source data and the imported cases from abroad in the case data.

[0040] The embodiment of the present invention further provides a dengue fever prediction device considering extreme weather driving, comprising:

[0041] A data acquisition module, used to acquire case data and multi-source data in the target area, wherein the multi-source data includes meteorological data, environmental data and socio-economic data;

[0042] An extreme weather feature module, used to construct extreme weather features based on the meteorological data and a preset extreme weather threshold;

[0043] A spatiotemporal lag data module, used to obtain the spatiotemporal lag data of the epidemic according to the spatial autocorrelation relationship and the temporal autocorrelation relationship of the case data;

[0044] A model building module, used to add the spatiotemporal lag term of the epidemic situation to the zero-inflated negative binomial regression model to obtain an initial epidemic prediction model; the epidemic prediction model includes a quadratic term of the extreme weather characteristics;

[0045] A parameter fitting module, used to perform parameter fitting on the initial epidemic prediction model based on the case data, the extreme weather characteristics, the epidemic spatiotemporal lag data, and the multi-source data to obtain a fitted epidemic prediction model;

[0046] The epidemic prediction module is used to predict the future epidemic situation in the target area using the fitted epidemic prediction model.

[0047] An embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any of the above-mentioned dengue fever prediction methods considering extreme weather drive.

[0048] Compared with the prior art, the beneficial effects of a dengue fever prediction method, device and medium considering extreme weather driving provided by the embodiment of the present invention are: by constructing a zero-inflated negative binomial spatiotemporal lag model, and combining case data, extreme weather characteristics, epidemic spatiotemporal lag data and multi-source data (including meteorological data, environmental data and socio-economic data), a reliable prediction of the dengue fever epidemic is achieved; through multiple screening of model input variables, the model input data has a more reasonable driving mechanism, which improves the prediction performance of the model; by considering the spatiotemporal autocorrelation of dengue fever cases in neighboring areas, as well as the time lag and nonlinear relationship of climate factors, the dynamic relationship between extreme weather and dengue fever transmission can be more accurately captured, and the accuracy of model prediction is improved. Compared with prediction methods such as deep learning models and large models, the model of the embodiment of the present invention has low complexity, low cost, strong interpretability, and high credibility, which helps public health departments to accurately grasp the epidemic trend of dengue fever and provide a scientific basis for formulating effective prevention and control strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flow chart of a dengue fever prediction method considering extreme weather drive provided by an embodiment of the present invention;

[0050] Figure 2 It is a structural schematic diagram of a dengue fever prediction device considering extreme weather drive provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0052] See also Figure 1 , Figure 1 1 is a flow chart of a dengue fever prediction method considering extreme weather driving provided by an embodiment of the present invention. The dengue fever prediction method considering extreme weather driving comprises:

[0053] S1: Acquire case data and multi-source data of the target area, wherein the multi-source data includes meteorological data, environmental data and socio-economic data;

[0054] Specifically, first determine the target area for research based on historical dengue epidemic data. Areas with high epidemic incidence or representativeness can be selected as target areas, and ensure that the target area has accessible high-precision meteorological data to facilitate the subsequent construction of a prediction model. When determining the size of the target area, it is necessary to consider the spatial scale of the data to avoid a large sample size of high-precision data due to an area that is too large, which makes the calculation cost too high, and also to avoid the inability to accurately evaluate the driving effect of each variable due to an area that is too small.

[0055] As one of the optional embodiments, the step of obtaining case data of the target area includes:

[0056] Obtain unstructured address text from dengue case reports in the target area;

[0057] Mapping the unstructured address text into a standard address library to obtain the geographic coordinates corresponding to the case;

[0058] The target area is segmented into grids, and the number of cases in each grid unit is calculated based on the geographic coordinates to obtain gridded case data.

[0059] Specifically, the dengue case reports of the target area are obtained, and the unstructured address text of each case is obtained from the attribute data of the report; then the place name address matching engine is used to map the unstructured address text to the standard address library to realize the geographical coding of the address. Specifically, the data is collected by combining the Internet and open source geographic information, and after automatic address data cleaning and quality inspection, a high-quality standard address library is constructed. By combining quantitative matching strategies and bidirectional indexes, the unstructured address text to be matched is quickly matched with the standard address of hierarchical segmentation to obtain the precise geographical coordinates of each case; the geographical coordinates of the case are spatially distributed and visualized. Specifically, a fishing net of the same size as the research unit is constructed, and the number of dengue cases in each grid is calculated to form gridded dengue case spatial distribution data on a time scale such as daily or monthly, i.e., case data. Among them, the number of dengue cases can also be subdivided into the number of dengue cases with different population characteristics (such as gender, age, etc.).

[0060] The embodiment of the present invention is based on a standard address database and adopts an efficient and accurate place name address matching technology to ensure that dengue fever cases can be accurately located in space and reduce the accuracy loss caused by model input errors.

[0061] As one of the optional embodiments, the meteorological data includes but is not limited to daily maximum temperature, daily minimum temperature, daily precipitation and drought condition data;

[0062] The environmental data include but are not limited to air quality data, vegetation coverage and impervious surface ratio;

[0063] The socioeconomic data include but are not limited to mosquito density, population density, road network density, distribution of medical facilities and urban construction data.

[0064] Specifically, high-precision meteorological data (such as an accuracy of 1km or 0.05 degrees of longitude and latitude) of the target area is obtained, including daily maximum temperature, daily minimum temperature, daily precipitation and drought conditions data.

[0065] Obtain air quality data that have an impact on the spread of dengue fever (such as PM2.5, PM10, O3, SO2 concentrations, etc.), as well as environmental data such as vegetation coverage and proportion of impervious surfaces.

[0066] Obtain socioeconomic data closely related to the spread of dengue fever, such as mosquito density, population density, road network density, distribution of medical facilities, and urban construction conditions (such as sewage concentration, density of public toilets, and other facilities where there may be stagnant water, providing a breeding environment for mosquitoes).

[0067] Furthermore, the acquired case data and multi-source data were cleaned and quality checked, including 99% tailing outliers, deletion of missing values, approximate value replacement of missing values, etc. At the same time, standardization was performed to ensure the consistency of data coordinates, numerical format, and the rationality of units of dimensions.

[0068] The embodiment of the present invention comprehensively considers various driving factors closely related to the incidence of dengue fever, such as historical case data, meteorological environment, and socioeconomic conditions, so that the input data of the prediction model has a more reasonable driving mechanism, thereby enhancing the interpretability of the model and facilitating the disclosure of the influencing mechanism and law of the dengue fever epidemic.

[0069] S2: constructing extreme weather characteristics according to the meteorological data and a preset extreme weather threshold;

[0070] As one of the optional embodiments, constructing extreme weather features according to the meteorological data and a preset extreme weather threshold includes:

[0071] Based on the historical meteorological data of the historical base period, the quantiles of the historical meteorological data are calculated using a sliding window method to determine the extreme weather threshold;

[0072] According to the extreme weather threshold, extreme weather characteristics are extracted from the meteorological data, wherein the extreme weather characteristics include the frequency, intensity and duration of each extreme weather event.

[0073] Specifically, the threshold for each extreme weather is first determined: based on the historical meteorological data of the historical base period (such as 1961-1990), the sliding window method (such as taking 5 days as a window period) is used to calculate the quantiles of the same historical period as the threshold for extreme weather events. For example, the daytime heat wave threshold is obtained according to the 90th percentile of the daily maximum temperature; the nighttime heat wave threshold is obtained according to the 90th percentile of the daily minimum temperature; the nighttime heat wave threshold is obtained according to the 95th percentile of the daily precipitation. For extreme climate events such as drought, the level boundary values ​​representing no drought and mild drought in meteorological drought indices such as SPEI can be directly used as the corresponding extreme weather thresholds.

[0074] Then, extreme weather characteristic variables are constructed: based on the meteorological data of the target area obtained, the monthly frequency, monthly average intensity and monthly average duration of extreme weather events such as daytime heat wave events (DHW), nighttime heat wave events (NHW) and extreme precipitation events (EP) that exceed the extreme weather threshold for three consecutive days are calculated; for drought events, the severity of the drought events is determined based on the drought index and the corresponding threshold.

[0075] S3: Obtaining the spatiotemporal lag data of the epidemic according to the spatial autocorrelation and temporal autocorrelation of the case data;

[0076] As one of the optional embodiments, obtaining the spatiotemporal lag data of the epidemic situation according to the spatial autocorrelation relationship and the temporal autocorrelation relationship of the case data includes:

[0077] Based on the case data, a spatial weight matrix is ​​constructed by calculating the spatial correlation index value of each grid unit and the adjacent area, and an inverse distance weighted matrix is ​​obtained by combining the preset inverse distance weighted radius;

[0078] Calculating the spatial lag data of each grid cell according to the inverse distance weight matrix and the case data;

[0079] Performing least squares regression on the case data to obtain residuals, and calculating the DW statistic based on the residuals;

[0080] It is determined whether the case data has a first-order positive correlation based on the DW statistic. If so, the spatial lag data lagged by one period is used as the spatiotemporal lag data of the epidemic.

[0081] Specifically, the spatial correlation index includes Moran's I and LISA (Local Local Indicators of Spatial Association) index. The global Moran's I index and the local LISA index are used to analyze the spatial distribution of dengue fever case data in each grid unit of the target area to determine whether there is a spatial autocorrelation relationship in the dengue fever distribution. Among them, when Moran's I>0, it means that there is a spatial positive correlation. The larger the index value, the greater the spatial autocorrelation. The calculation formula of Moran's I index is:

[0082]

[0083] Where n represents the number of grid cells in the target area, x i or x j is the number of cases in cell i or j, is the average number of cases in n units, w ij is a weight matrix that defines the relationships between geographic units ij.

[0084] The LISA index represents the agglomeration relationship between each grid unit and its adjacent areas, and its calculation formula is:

[0085]

[0086] Where n represents the number of grid cells in the target area, x i or x j is the number of cases in cell i or j, is the average number of cases in n units, S 2 is the variance, w ij is a weight matrix that defines the relationships between geographic units ij.

[0087] Furthermore, the spatial weight matrix is ​​constructed by calculating the spatial correlation index. Taking into account the spatial scale of the study, the spatial distribution range of dengue fever and the range of mosquito vector activities, an appropriate inverse distance weighted radius (such as 2 km) is set to reflect the influence relationship between various geographical locations based on the inverse of the distance square. According to the spatial weight matrix and the inverse distance weighted radius, the inverse distance weight matrix of the target area is obtained.

[0088] Furthermore, the spatial lag data of each grid cell were calculated using the inverse distance weight matrix and dengue case data. The spatial lag data indicated the extent to which the grid cell was affected by surrounding cases, reflecting the spatial correlation of the dengue epidemic situation affected by neighboring areas.

[0089] Furthermore, based on the time series data of the case data, the data was regressed using the least squares OLS method, and the DW statistic was calculated based on the model residuals:

[0090]

[0091] When the DW statistic is in the interval (0, dL), it indicates that the case data presents a first-order positive autocorrelation in time, where dL is the lower critical value of the DW test. Taking into account the possible time lag effect between adjacent regions, the dengue fever epidemic data of the past time unit (such as the last week or the last month) is regarded as a key reference for predicting the current epidemic. This past time point is called lag 1. In other words, the dengue fever epidemic data with a lag of 1 period (i.e., epidemic spatiotemporal lag data) that takes into account the dual lag effects of space and time are incorporated into the epidemic prediction model as the input variable of the epidemic prediction model.

[0092] S4: adding the spatiotemporal lag term of the epidemic situation to the zero-inflated negative binomial regression model to obtain an initial epidemic prediction model; the epidemic prediction model includes the quadratic term of the extreme weather characteristics;

[0093] Specifically, case data generally do not occur in every grid cell. Therefore, in the results obtained from the zero-inflation and over-dispersion analysis, the case data distribution is more likely to have zero-inflation and over-dispersion distributions. Therefore, the zero-inflation negative binomial regression model is selected as the basic model of the epidemic prediction model.

[0094] The judgment of zero expansion phenomenon is: draw a frequency distribution diagram based on case data, calculate the zero value ratio of grid cells, if the zero value ratio is too high, it means that zero expansion phenomenon exists. The judgment of overdispersion is: calculate the O statistic based on the distribution characteristics of case data, and the calculation formula is:

[0095]

[0096] In the formula, n represents the number of samples, represents the mean of all sample dengue cases, and s represents the standard deviation of all dengue cases. If the O statistic is ≥1.96, it means that the data is overdispersed and the negative binomial distribution model should be selected.

[0097] Furthermore, if the data shows zero inflation, the zero part of the value needs to be modeled separately. Usually, the occurrence of dengue fever is related to imported cases from abroad and human activities, and the night light intensity is a proxy variable that comprehensively reflects human activities. Therefore, the number of imported cases from abroad and the night light intensity are selected as the zero inflation factors of the zero-inflation negative binomial regression model.

[0098] Furthermore, considering that the impact of extreme weather on dengue fever generally presents a nonlinear relationship of first promoting and then inhibiting, a quadratic term is added to the model to express this nonlinearity. The spatiotemporal lag term of dengue fever cases (corresponding to the spatiotemporal lag data of the epidemic) is integrated into the zero-inflated negative binomial regression model to form a zero-inflated negative binomial spatiotemporal lag regression model (ZINB-STLRM), i.e., an epidemic prediction model, in order to improve the fitting effect of the model.

[0099] As one of the optional embodiments, the epidemic prediction model includes a counting part and a zero-inflation part;

[0100] The expression of the counting part is:

[0101]

[0102] The expression of the zero expansion part is:

[0103]

[0104] Where i represents the grid number, t represents the time, n represents the number of grids, log(λ it ) is the natural logarithm expectation of the number of dengue cases in the grid cell with base e, X it represents the observed value of extreme weather characteristics, α iy The quadratic term of each extreme weather characteristic has an impact on the prevalence of dengue fever, β it represents the influence coefficient of extreme weather characteristics on the prevalence of dengue fever, w i y i(t-1) is the spatiotemporal lag data of the epidemic (i.e., the observed value y of the i-th grid cell at time t-1 i(t-1) According to the spatial weight w i The weighted result represents the dengue cases in the adjacent areas lagged by one period), ρ is the dengue spatial effect coefficient, ε it Represents the unobservable part and individual heterogeneity, the multi-source data is used as a control variable, CV it represents the observed value of the control variable, γ it represents the influence coefficient of each control variable on the prevalence of dengue fever; p it is the probability of zero expansion, τ i is the influence coefficient of the zero expansion factor, G i is the observed value of zero inflation factor, τ0 is a constant term;

[0105] The zero inflation factor includes the nighttime light intensity in the multi-source data and the number of imported cases from abroad in the case data to reflect the characteristics of population density and socioeconomic activities.

[0106] The zero-inflated negative binomial spatiotemporal lag regression model constructed by the embodiment of the present invention comprehensively considers the zero-inflation phenomenon, the overdispersion phenomenon and the spatiotemporal autocorrelation of dengue cases in neighboring areas, thereby improving the prediction accuracy. In the context of global warming, extreme weather is increasingly driving the spread of dengue fever, and the occurrence of dengue fever is the result of the combined effects of multiple meteorological factors. Small extreme weather may promote the spread of dengue virus, while extreme weather exceeding a certain intensity may inhibit the occurrence of dengue fever, and this driving effect may have a time lag effect. Therefore, the embodiment of the present invention can more accurately capture the dynamic relationship between extreme weather and dengue fever transmission by considering the time lag effect and nonlinear relationship of climate factors, thereby improving the credibility of the prediction results.

[0107] S5: Based on the case data, the extreme weather characteristics, the spatiotemporal lag data of the epidemic, and the multi-source data, parameter fitting is performed on the initial epidemic prediction model to obtain a fitted epidemic prediction model;

[0108] As one of the optional embodiments, the initial epidemic prediction model is parameter-fitted based on the case data, the extreme weather characteristics, the epidemic spatiotemporal lag data, and the multi-source data to obtain a fitted epidemic prediction model, including:

[0109] Taking the extreme weather characteristics, the spatiotemporal lag data of the epidemic, and the multi-source data as a first variable set, and constructing a basic regression model for each type of extreme weather according to the first variable set and the case data;

[0110] Based on the basic regression model, respectively calculating the variance inflation factor of each of the extreme weather characteristics, and selecting variables in the first variable set according to the variance inflation factor to obtain a second variable set;

[0111] Based on the second variable set, a penalty term is added to the basic regression model of each type of extreme weather according to the Lasso regression algorithm, and the variables whose contribution rate to the model is less than a preset threshold are eliminated by the stepwise regression method to obtain a third variable set;

[0112] According to the third variable set and the case data, the initial epidemic prediction model is fitted with parameters to obtain a fitted epidemic prediction model.

[0113] Specifically, in order to improve the prediction performance of the model, the input variables of the model are first screened. First, a basic regression model for each type of extreme weather is constructed based on extreme weather characteristics, spatiotemporal lag data of the epidemic, multi-source data, and case data. The variance inflation factor (VIF) is used to detect multicollinearity. Multicollinearity refers to the existence of a high linear correlation between variables in the regression model, which will lead to inaccurate estimation of the regression coefficient. Therefore, highly correlated variables are screened out with VIF>10 as the standard. Then, the extreme weather variables are locked, and the control variables are selected using Lasso regression. A penalty term is added to the linear regression model of each type of extreme weather, that is, the sum of the absolute values ​​of the characteristic coefficients is multiplied by the parameter λ to control the complexity of the model, so that the model selects fewer features, thereby reducing the risk of overfitting; then the stepwise regression method is used to further screen the control variables and determine the control variables that are significantly associated with the dengue fever epidemic. Specifically, according to the variables on the model R 2 The contribution rate of R 2 The variables whose contribution rate is less than the threshold (such as 0.01) obtain the final model input variables (i.e., the third variable set).

[0114] Furthermore, based on the case data and the screened variables (including multi-source data, extreme weather characteristics, and epidemic spatiotemporal lag data), the epidemic prediction model is fitted with parameters, the fitting effect of the model is judged by the AIC criterion and / or the BIC criterion, and the parameters of the optimal model are selected as the model parameters after fitting to obtain the fitted epidemic prediction model.

[0115] During the model fitting process, the relative risk ratio IRR, i.e. exp(β), is selected for reporting, so that the various regression coefficients of the model can be explained by scientific theory, and the IRR coefficient can be used to explain the promotion or inhibition of the risk of dengue fever transmission by various extreme weather characteristics and their related driving factors. At the same time, the performance of the model in terms of spatiotemporal lag and nonlinear effects is checked to verify the rationality of the model selection. For example, whether the spatiotemporal lag term is positive indicates that the number of cases in the adjacent region in the previous period will lead to an increase in the number of dengue fever cases in the current region.

[0116] Furthermore, the observed values ​​of the extreme weather characteristics of the epidemic prediction model include the observed values ​​of a single extreme weather characteristic and / or the cross-product terms of different extreme weather characteristics.

[0117] Specifically, if the observed value of a single extreme weather feature is taken as X it Substituting into the model, the effect of a single extreme weather feature on the model prediction can be analyzed through the model; if the cross-product terms of different extreme weather features are added to the model, the interaction between different extreme weather can be studied through the cross-product terms, such as taking the daytime heat wave intensity and the nighttime heat wave intensity and their cross-product terms as X itSubstituting into the model, if a negative coefficient appears in the cross-product term after model fitting, that is, the relative risk ratio IRR is less than 1, it means that the two extreme weather characteristics show antagonistic effects and can reduce the risk of dengue fever transmission.

[0118] Furthermore, a test set was obtained to verify the fitted epidemic prediction model, and the prediction performance of the model was evaluated using indicators such as MAE and RMSE to ensure the reliability of the epidemic prediction model.

[0119] S6: Use the fitted epidemic prediction model to predict future epidemics in the target area.

[0120] Specifically, the future climate prediction data set and related control variable prediction data set of the target area are obtained, and the dengue fever epidemic in the target area is predicted using the fitted epidemic prediction model to obtain the prediction results. Then, according to the mechanism behind the model coefficients and the regional dengue fever epidemic development forecast, early warning and prevention and control recommendations are provided to the public health department to optimize the dengue fever prevention and control strategy.

[0121] The embodiment of the present invention realizes the reliable prediction of dengue fever epidemic situation by constructing a zero-inflated negative binomial spatiotemporal lag model, and combining case data, extreme weather characteristics, epidemic spatiotemporal lag data and multi-source data (including meteorological data, environmental data and socio-economic data); through multiple screening of model input variables, the model input data has a more reasonable driving mechanism, which improves the prediction performance of the model; by considering the spatiotemporal autocorrelation of dengue fever cases in neighboring areas, as well as the time lag and nonlinear relationship of climate factors, it can more accurately capture the dynamic relationship between extreme weather and dengue fever transmission, and improve the accuracy of model prediction. Compared with prediction methods such as deep learning models and large models, the embodiment of the present invention has low complexity, low cost, strong interpretability, and high credibility, which helps public health departments to accurately grasp the epidemic trend of dengue fever epidemic situation and provide a scientific basis for formulating effective prevention and control strategies.

[0122] Accordingly, the present invention also provides a dengue fever prediction device considering extreme weather-driven factors, which can implement all the processes of the dengue fever prediction method considering extreme weather-driven factors in the above-mentioned embodiment.

[0123] See also Figure 2 , Figure 2 1 is a schematic diagram of a dengue fever prediction device considering extreme weather driving provided by an embodiment of the present invention. The dengue fever prediction device considering extreme weather driving comprises:

[0124] A data acquisition module 201 is used to acquire case data and multi-source data of a target area, wherein the multi-source data includes meteorological data, environmental data, and socio-economic data;

[0125] The extreme weather feature module 202 is used to construct extreme weather features according to the meteorological data and a preset extreme weather threshold;

[0126] The spatiotemporal lag data module 203 is used to obtain the spatiotemporal lag data of the epidemic situation according to the spatial autocorrelation relationship and the temporal autocorrelation relationship of the case data;

[0127] A model building module 204 is used to add the spatiotemporal lag term of the epidemic situation to the zero-inflated negative binomial regression model to obtain an initial epidemic situation prediction model; the epidemic situation prediction model includes the quadratic term of the extreme weather characteristics;

[0128] A parameter fitting module 205 is used to perform parameter fitting on the initial epidemic prediction model based on the case data, the extreme weather characteristics, the epidemic spatiotemporal lag data, and the multi-source data to obtain a fitted epidemic prediction model;

[0129] The epidemic prediction module 206 is used to predict the future epidemic situation in the target area by using the fitted epidemic prediction model.

[0130] Preferably, obtaining case data of the target area includes:

[0131] Obtain unstructured address text from dengue case reports in the target area;

[0132] Mapping the unstructured address text into a standard address library to obtain the geographic coordinates corresponding to the case;

[0133] The target area is segmented into grids, and the number of cases in each grid unit is calculated based on the geographic coordinates to obtain gridded case data.

[0134] Preferably, the meteorological data include but are not limited to daily maximum temperature, daily minimum temperature, daily precipitation and drought condition data;

[0135] The environmental data include but are not limited to air quality data, vegetation coverage and impervious surface ratio;

[0136] The socioeconomic data include but are not limited to mosquito density, population density, road network density, distribution of medical facilities and urban construction data.

[0137] Preferably, the extreme weather feature module 202 is specifically used for:

[0138] Based on the historical meteorological data of the historical base period, the quantiles of the historical meteorological data are calculated using a sliding window method to determine the extreme weather threshold;

[0139] According to the extreme weather threshold, extreme weather characteristics are extracted from the meteorological data, wherein the extreme weather characteristics include the frequency, intensity and duration of each extreme weather event.

[0140] Preferably, the spatiotemporal lag data module 203 is specifically used for:

[0141] Based on the case data, a spatial weight matrix is ​​constructed by calculating the spatial correlation index value of each grid unit and the adjacent area, and an inverse distance weighted matrix is ​​obtained by combining the preset inverse distance weighted radius;

[0142] Calculating the spatial lag data of each grid cell according to the inverse distance weight matrix and the case data;

[0143] Performing least squares regression on the case data to obtain residuals, and calculating the DW statistic based on the residuals;

[0144] It is determined whether the case data has a first-order positive correlation based on the DW statistic. If so, the spatial lag data lagged by one period is used as the spatiotemporal lag data of the epidemic.

[0145] Preferably, the parameter fitting module 205 is specifically used for:

[0146] Taking the extreme weather characteristics, the spatiotemporal lag data of the epidemic, and the multi-source data as a first variable set, and constructing a basic regression model for each type of extreme weather according to the first variable set and the case data;

[0147] Based on the basic regression model, respectively calculating the variance inflation factor of each of the extreme weather characteristics, and selecting variables in the first variable set according to the variance inflation factor to obtain a second variable set;

[0148] Based on the second variable set, a penalty term is added to the basic regression model of each type of extreme weather according to the Lasso regression algorithm, and the variables whose contribution rate to the model is less than a preset threshold are eliminated by the stepwise regression method to obtain a third variable set;

[0149] According to the third variable set and the case data, the initial epidemic prediction model is fitted with parameters to obtain a fitted epidemic prediction model.

[0150] Preferably, the epidemic prediction model includes a counting part and a zero-inflation part;

[0151] The expression of the counting part is:

[0152]

[0153] Where i represents the grid number, t represents the time, n represents the number of grids, log(λit ) is the natural logarithm expectation of the number of dengue cases in the grid cell with base e, X it represents the observed value of extreme weather characteristics, α it The quadratic term of each extreme weather characteristic has an impact on the prevalence of dengue fever, β it represents the influence coefficient of extreme weather characteristics on the prevalence of dengue fever, w i y i(t-1) is the spatiotemporal lag data of the epidemic, ρ is the spatial effect coefficient of dengue fever, ε it Represents the unobservable part and individual heterogeneity, the multi-source data is used as a control variable, CV it represents the observed value of the control variable, γ it represents the influence coefficient of each control variable on the prevalence of dengue fever;

[0154] The observed values ​​of extreme weather characteristics include the observed values ​​of a single extreme weather characteristic and / or the cross-product terms of different extreme weather characteristics.

[0155] Preferably, the expression of the zero expansion part is:

[0156]

[0157] Among them, p it is the probability of zero expansion, v i is the influence coefficient of the zero expansion factor, G i is the observed value of the zero inflation factor, and τ0 is a constant term; the zero inflation factor includes the night light intensity in the multi-source data and the imported cases from abroad in the case data.

[0158] In the specific implementation, the working principle, control process and technical effect achieved by the dengue fever prediction device considering extreme weather-driven provided in the embodiment of the present invention are the same as those of the dengue fever prediction method considering extreme weather-driven in the above-mentioned embodiment, and will not be repeated here.

[0159] An embodiment of the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the dengue fever prediction method considering extreme weather drive as described in any of the above embodiments.

[0160] The embodiment of the present invention provides a dengue fever prediction method, device and medium considering extreme weather driving, and its beneficial effects are: by constructing a zero-inflated negative binomial spatiotemporal lag model, and combining case data, extreme weather characteristics, epidemic spatiotemporal lag data and multi-source data (including meteorological data, environmental data and socio-economic data), a reliable prediction of the dengue fever epidemic is achieved; by multiple screening of model input variables, the model input data has a more reasonable driving mechanism, which improves the prediction performance of the model; by considering the spatiotemporal autocorrelation of dengue fever cases in neighboring areas, as well as the time lag and nonlinear relationship of climate factors, the dynamic relationship between extreme weather and dengue fever transmission can be more accurately captured, and the accuracy of model prediction is improved. Compared with prediction methods such as deep learning models and large models, the embodiment of the present invention has low complexity, low cost, strong interpretability, and high credibility, which helps public health departments to accurately grasp the epidemic trend of dengue fever and provide a scientific basis for formulating effective prevention and control strategies.

[0161] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A dengue fever prediction method considering extreme weather driving, characterized in that: include: Acquire case data and multi-source data of the target area, wherein the multi-source data includes meteorological data, environmental data and socio-economic data; constructing extreme weather features based on the meteorological data and a preset extreme weather threshold; According to the spatial autocorrelation and temporal autocorrelation of the case data, the spatiotemporal lag data of the epidemic are obtained; Adding the spatiotemporal lag term of the epidemic situation to the zero-inflated negative binomial regression model to obtain an initial epidemic situation prediction model; the epidemic situation prediction model includes a quadratic term of the extreme weather characteristics; Based on the case data, the extreme weather characteristics, the spatiotemporal lag data of the epidemic, and the multi-source data, parameter fitting is performed on the initial epidemic prediction model to obtain a fitted epidemic prediction model; The fitted epidemic prediction model is used to predict the future epidemic situation in the target area.

2. The dengue fever prediction method considering extreme weather drive as claimed in claim 1, characterized in that: The step of obtaining case data of the target area includes: Obtain unstructured address text from dengue case reports in the target area; Mapping the unstructured address text into a standard address library to obtain the geographic coordinates corresponding to the case; The target area is segmented into grids, and the number of cases in each grid unit is calculated based on the geographic coordinates to obtain gridded case data.

3. The dengue fever prediction method considering extreme weather drive as claimed in claim 1, characterized in that: The meteorological data include but are not limited to daily maximum temperature, daily minimum temperature, daily precipitation and drought conditions data; The environmental data include but are not limited to air quality data, vegetation coverage and impervious surface ratio; The socioeconomic data include but are not limited to mosquito density, population density, road network density, distribution of medical facilities and urban construction data.

4. The dengue fever prediction method considering extreme weather drive as claimed in claim 1, characterized in that: The step of constructing extreme weather characteristics according to the meteorological data and a preset extreme weather threshold comprises: Based on the historical meteorological data of the historical base period, the quantiles of the historical meteorological data are calculated using a sliding window method to determine the extreme weather threshold; According to the extreme weather threshold, extreme weather characteristics are extracted from the meteorological data, wherein the extreme weather characteristics include the frequency, intensity and duration of each extreme weather event.

5. The dengue fever prediction method considering extreme weather drive as claimed in claim 1, characterized in that: The spatial and temporal autocorrelation relationship of the case data is used to obtain the spatiotemporal lag data of the epidemic, including: Based on the case data, a spatial weight matrix is ​​constructed by calculating the spatial correlation index value of each grid unit and the adjacent area, and an inverse distance weighted matrix is ​​obtained by combining the preset inverse distance weighted radius; Calculating the spatial lag data of each grid cell according to the inverse distance weight matrix and the case data; Performing least squares regression on the case data to obtain residuals, and calculating the DW statistic based on the residuals; It is determined whether the case data has a first-order positive correlation based on the DW statistic. If so, the spatial lag data lagged by one period is used as the spatiotemporal lag data of the epidemic.

6. The dengue fever prediction method considering extreme weather drive as claimed in claim 1, characterized in that: The performing parameter fitting on the initial epidemic prediction model based on the case data, the extreme weather characteristics, the epidemic spatiotemporal lag data, and the multi-source data to obtain a fitted epidemic prediction model includes: Taking the extreme weather characteristics, the spatiotemporal lag data of the epidemic, and the multi-source data as a first variable set, and constructing a basic regression model for each type of extreme weather according to the first variable set and the case data; Based on the basic regression model, respectively calculating the variance inflation factor of each of the extreme weather characteristics, and selecting variables in the first variable set according to the variance inflation factor to obtain a second variable set; Based on the second variable set, a penalty term is added to the basic regression model of each type of extreme weather according to the Lasso regression algorithm, and the variables whose contribution rate to the model is less than a preset threshold are eliminated by the stepwise regression method to obtain a third variable set; According to the third variable set and the case data, the initial epidemic prediction model is fitted with parameters to obtain a fitted epidemic prediction model.

7. The dengue fever prediction method considering extreme weather drive as claimed in claim 6, characterized in that: The epidemic prediction model includes a counting part and a zero-inflation part; The expression of the counting part is: Where i represents the grid number, t represents the time, n represents the number of grids, log(λ it ) is the natural logarithm expectation of the number of dengue cases in the grid cell with base e, X it represents the observed value of extreme weather characteristics, α it The quadratic term representing the influence coefficient of each extreme weather characteristic on the prevalence of dengue fever, β it represents the influence coefficient of extreme weather characteristics on the prevalence of dengue fever, w i y i(t-1) is the spatiotemporal lag data of the epidemic, ρ is the spatial effect coefficient of dengue fever, ε it Represents the unobservable part and individual heterogeneity, the multi-source data is used as a control variable, CV ut represents the observed value of the control variable, γ it represents the influence coefficient of each control variable on the prevalence of dengue fever; The observed values ​​of extreme weather characteristics include the observed values ​​of a single extreme weather characteristic and / or the cross-product terms of different extreme weather characteristics.

8. The dengue fever prediction method considering extreme weather drive as claimed in claim 7, characterized in that: The expression of the zero expansion part is: Among them, p it is the probability of zero expansion, v i is the influence coefficient of the zero expansion factor, G i is the observed value of the zero inflation factor, and τ0 is a constant term; the zero inflation factor includes the night light intensity in the multi-source data and the number of imported cases from abroad in the case data.

9. A dengue fever prediction device considering extreme weather driving, characterized in that: include: A data acquisition module, used to acquire case data and multi-source data in the target area, wherein the multi-source data includes meteorological data, environmental data and socio-economic data; An extreme weather feature module, used to construct extreme weather features based on the meteorological data and a preset extreme weather threshold; A spatiotemporal lag data module, used to obtain the spatiotemporal lag data of the epidemic according to the spatial autocorrelation relationship and the temporal autocorrelation relationship of the case data; A model building module, used to add the spatiotemporal lag term of the epidemic situation to the zero-inflated negative binomial regression model to obtain an initial epidemic prediction model; the epidemic prediction model includes a quadratic term of the extreme weather characteristics; A parameter fitting module, used to perform parameter fitting on the initial epidemic prediction model based on the case data, the extreme weather characteristics, the epidemic spatiotemporal lag data, and the multi-source data to obtain a fitted epidemic prediction model; The epidemic prediction module is used to predict the future epidemic situation in the target area using the fitted epidemic prediction model.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the dengue fever prediction method considering extreme weather drive as described in any one of claims 1 to 8 is implemented.

Images