MaxEnt model-combined plague risk assessment method and MaxEnt model-combined plague risk assessment system

Abstract

The invention relates to the crossing field of public health and geographic information technology, and discloses a plague risk assessment method and system combined with a MaxEnt model, and the method comprises the steps: obtaining a multi-source environment variable, and carrying out the standardization; reducing a colinear variable through variance threshold preliminary screening, a Pearson's correlation coefficient and plague point significance test; performing variable importance sorting by using recursive feature elimination and a support vector machine, and dynamically determining an optimal variable subset; and inputting a MaxEnt model to train ecological niche probability distribution, and dividing risk levels based on historical occurrence point probability quantiles. The system comprises corresponding function modules. According to the system, through a three-stage automatic variable screening mechanism, the generalization ability, interpretability and prediction precision of the model are remarkably improved, and reliable support is provided for accurate prevention and control of plague.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of public health and geographic information technology, specifically involving a plague risk assessment method and system that combines the MaxEnt model. Background Technology

[0002] Plague, a highly virulent zoonotic disease caused by Yersinia pestis, is influenced by a complex interplay of ecological and environmental factors, including climate conditions, vegetation cover, host animal population dynamics, and geographic spatial characteristics. In recent years, risk assessment methods based on Species Distribution Models (SDMs) have been widely applied in infectious disease prediction, with the Maximum Entropy Model (MaxEnt) becoming a mainstream tool due to its high predictive accuracy even with small sample sizes. This model integrates known outbreak distributions with environmental variables to infer potential suitable habitats for pathogens, providing spatial decision support for public health interventions.

[0003] Plague risk assessment methods based on the MaxEnt model rely on the systematic input of multi-source environmental variables to characterize the ecological suitability of plague foci. The basic principle is to construct a statistical correlation model between environmental variables and the probability of plague occurrence, using the information entropy maximization criterion, while satisfying the constraints of observational data. However, the effectiveness of this method is highly dependent on the quality and relevance of the input variables; the scientific nature of variable selection directly determines the model's generalization ability and explanatory power.

[0004] Environmental variables are typically manually selected by researchers based on experience or literature reviews, lacking objective quantitative standards. This leads to subjective preferences significantly influencing the composition of the variable set. The inclusion of numerous redundant or weakly correlated variables (such as highly collinear temperature indicators or remote sensing indices not directly related to plague transmission mechanisms) in the model not only exacerbates computational complexity but also causes overfitting, resulting in poor stability and high false alarm rates in new regions or time periods.

[0005] Traditional screening methods struggle to capture the complex nonlinear and non-monotonic relationship between variables and plague outbreaks, and cannot dynamically reflect the time-varying characteristics of variable contributions under different seasons or ecological backgrounds, resulting in weakened model interpretability and difficulty in supporting the formulation of precise prevention and control strategies. Summary of the Invention

[0006] This invention provides a method and system for plague risk assessment combined with the MaxEnt model, aiming to solve the technical problems of overfitting, poor interpretability, and high false alarm rate in existing MaxEnt models caused by the high subjectivity of manual selection of environmental variables and the high proportion of redundant variables. This invention constructs an automated variable processing mechanism based on multi-stage feature selection and dynamic weight optimization to achieve efficient dimensionality reduction and information preservation of the original environmental variable set, thereby improving the generalization ability, interpretability, and prediction accuracy of the MaxEnt model in plague risk prediction.

[0007] This invention provides a plague risk assessment method combining the MaxEnt model, which includes:

[0008] Obtain the original environmental variable dataset covering the target area. The original environmental variable dataset includes temperature, precipitation, vegetation index, host animal population density, soil type, altitude, land use type, and human activity intensity index. Perform spatial resolution unification processing and geographic coordinate alignment on the original environmental variable dataset to form a standardized environmental variable matrix.

[0009] The first stage of variable screening is performed on the standardized environmental variable matrix, and environmental variables with variances lower than a preset threshold are removed using the variance threshold method.

[0010] The second stage of variable correlation reduction was performed on the environmental variables after the initial screening. The Pearson correlation coefficient between any two environmental variables was calculated. If the absolute value was greater than or equal to 0.7, the variable with a higher correlation to the spatial distribution of historical plague occurrence points was retained, and the other variable was removed.

[0011] The reduced set of environmental variables is subjected to a third stage of importance ranking based on recursive feature elimination using support vector machine. Historical plague occurrence points are used as positive samples and randomly generated non-occurrence points are used as negative samples. The support vector machine classifier is trained and sorted in descending order according to the absolute value of the weight of each environmental variable in the classification decision function. The top N environmental variables are selected from the ranking results to form the optimal variable subset, where N is the minimum number of variables that makes the cross-validation accuracy reach the plateau period for the first time.

[0012] The optimal subset of variables is input into the MaxEnt model, and combined with the spatial coordinates of historical plague occurrence points, a plague niche probability distribution model is trained.

[0013] Based on the plague niche probability distribution model, the probability score of plague occurrence is calculated for all geographical units in the target area, and low-risk, medium-risk and high-risk areas are divided according to the preset risk level threshold; a plague risk assessment map containing the risk level division results is output.

[0014] Furthermore, in the initial screening of variables in the first stage, the variance threshold is set to the tenths of the variance of all variables in the original environmental variable dataset to ensure that environmental factors with significant spatial variability are retained.

[0015] Furthermore, in the second stage of variable correlation reduction, the correlation with the spatial distribution of historical plague occurrence points is quantified by calculating the significance level of the difference between the mean value of each environmental variable at the location of the plague occurrence point and the global mean. The p-value of the two-sample t-test is used as the discrimination criterion, and variables with a p-value less than 0.05 are judged to have significant correlation.

[0016] Furthermore, in the third stage of support vector machine importance ranking based on recursive feature elimination, the support vector machine classifier adopts a radial basis kernel function, with its penalty parameter C set to 10 and the kernel function parameter gamma set to the reciprocal of the dimension of the input variable; the recursive feature elimination process is executed iteratively, and each iteration removes the last 10% of the variables in the importance ranking until the number of remaining variables is 1.

[0017] Furthermore, the determination process of N specifically includes: training the MaxEnt model with the set of variables remaining in each step of the recursive feature elimination process, and calculating the average area under the receiver operating characteristic curve using five-fold cross-validation; performing first-order difference on the curve of the area under the receiver operating characteristic curve changing with the number of variables, and determining that the curve has entered a plateau period when the absolute value of the difference values ​​for three consecutive times is less than 0.015, and taking the number of variables corresponding to the starting point of the plateau period as N.

[0018] Furthermore, during the training process of the MaxEnt model, the number of background points is set to 10 times the number of historical plague occurrence points, the regularization multiplier is set to 1.5, the maximum number of iterations is set to 5000, and the convergence threshold is set to 0.00001.

[0019] Furthermore, the preset risk level threshold is determined based on the probability score distribution of historical plague occurrence points, wherein the high-risk area threshold is set to the 90th percentile of the probability score of historical occurrence points, the medium-risk area threshold is set to the 50th percentile, and areas below the 50th percentile are classified as low-risk areas.

[0020] According to another aspect of the present invention, a plague risk assessment system incorporating the MaxEnt model is provided, comprising:

[0021] The original environmental variable acquisition module is used to acquire the original environmental variable dataset covering the target area. The original environmental variable dataset includes temperature, precipitation, vegetation index, host animal population density, soil type, altitude, land use type, and human activity intensity index.

[0022] The data standardization module is used to perform spatial resolution unification and geographic coordinate alignment on the original environmental variable dataset to form a standardized environmental variable matrix.

[0023] The variable initial screening module is used to perform the first stage of variable initial screening on the standardized environmental variable matrix, and to remove environmental variables with variances lower than a preset threshold using the variance threshold method; the correlation reduction module is used to perform the second stage of variable correlation reduction on the environmental variables after initial screening, and to calculate the Pearson correlation coefficient between any two environmental variables. If its absolute value is greater than or equal to 0.7, the variable with a higher correlation to the spatial distribution of historical plague occurrence points is retained, and the other variable is removed.

[0024] The importance ranking module is used to perform the third stage of support vector machine importance ranking based on recursive feature elimination on the reduced set of environmental variables. It uses historical plague occurrence points as positive samples and randomly generated non-occurrence points as negative samples to train the support vector machine classifier and sorts them in descending order according to the absolute value of the weight of each environmental variable in the classification decision function.

[0025] The optimal variable subset determination module is used to select the top N environmental variables from the ranking results to form the optimal variable subset, where N is the minimum number of variables that makes the cross-validation accuracy reach the plateau period for the first time.

[0026] The MaxEnt model training module is used to input the optimal subset of variables into the MaxEnt model and, in combination with the spatial coordinates of historical plague occurrence points, train a plague niche probability distribution model.

[0027] The risk level classification module is used to calculate the probability score of plague occurrence for all geographical units in the target area based on the plague niche probability distribution model, and to classify low-risk areas, medium-risk areas and high-risk areas according to the preset risk level threshold.

[0028] The risk map output module is used to output a plague risk assessment map that includes risk level classification results.

[0029] Furthermore, in the variable initial screening module, the variance threshold is set to the tenths of the variance of all variables in the original environmental variable dataset.

[0030] Furthermore, in the correlation reduction module, the correlation with the spatial distribution of historical plague occurrence points is quantified by calculating the significance level of the difference between the mean value of each environmental variable at the location of the plague occurrence point and the global mean. The p-value of a two-sample t-test is used as the discrimination criterion, and variables with a p-value less than 0.05 are judged to have significant correlation.

[0031] Furthermore, in the importance ranking module, the support vector machine classifier adopts a radial basis function kernel function, with its penalty parameter C set to 10 and the kernel function parameter gamma set to the reciprocal of the dimension of the input variable; the recursive feature elimination process is executed iteratively, eliminating the last 10% of variables in the importance ranking in each iteration until the number of remaining variables is 1.

[0032] Furthermore, in the optimal variable subset determination module, the determination process of N specifically includes: training the MaxEnt model with the set of variables remaining in each step of the recursive feature elimination process, and calculating the average area under the receiver operating characteristic curve using five-fold cross-validation; performing first-order differencing on the curve of the area under the receiver operating characteristic curve changing with the number of variables, and determining that the curve has entered a plateau period when the absolute value of three consecutive differencing values ​​is less than 0.015, and taking the number of variables corresponding to the starting point of the plateau period as N.

[0033] Furthermore, in the MaxEnt model training module, the number of background points is set to ten times the number of historical plague occurrence points, the regularization multiplier is set to 1.5, the maximum number of iterations is set to 5000, and the convergence threshold is set to 0.00001.

[0034] Furthermore, in the risk level classification module, the preset risk level threshold is determined based on the probability score distribution of historical plague occurrence points. The high-risk area threshold is set to the 90th percentile of the probability score of historical occurrence points, the medium-risk area threshold is set to the 50th percentile, and areas below the 50th percentile are classified as low-risk areas.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] 1. This invention effectively overcomes the subjective bias and redundant interference caused by the reliance on manual experience in screening environmental variables in the traditional MaxEnt model by introducing a three-stage automated variable screening mechanism. The first stage uses a variance threshold method to eliminate invalid variables with weak spatial variability and low information content; the second stage uses a dual criterion of Pearson correlation coefficient and significance test of plague occurrence points to accurately identify and eliminate highly collinear variables, while retaining variables with stronger association with the plague niche; the third stage uses recursive feature elimination and support vector machine importance ranking to achieve quantitative evaluation of variable contribution and objective determination of the optimal subset.

[0037] 2. This invention significantly reduces the dimensionality of the model input, suppresses overfitting, and improves the model's generalization ability. Simultaneously, since the variables ultimately input into the MaxEnt model are all key driving factors that have undergone rigorous statistical testing and machine learning validation, the niche probability distribution output by the model has higher biological interpretability. Furthermore, the risk level threshold is dynamically set based on the probability distribution of historical occurrence points, avoiding false positives or false negatives caused by fixed thresholds, making the risk assessment results closer to actual epidemiological patterns.

[0038] 3. While ensuring the accuracy of prediction, this invention significantly improves the objectivity, stability and practical value of plague risk assessment, providing reliable technical support for early warning and precise prevention and control by public health departments. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the overall technical architecture of the plague risk assessment method and system combining the MaxEnt model proposed in this invention.

[0040] Figure 2 This is a schematic diagram of the core principle framework of the automated variable processing mechanism based on multi-stage feature selection and dynamic weight optimization in this invention;

[0041] Figure 3 This is a flowchart illustrating the standardization of the original environmental variable dataset and the logical process of three-stage variable selection in this invention.

[0042] Figure 4 This is a flowchart illustrating the logical process of determining the optimal variable subset based on recursive feature elimination and support vector machine importance ranking in this invention.

[0043] Figure 5 This is a flowchart illustrating the logical process of training the MaxEnt model and generating the probability distribution of the plague niche in this invention.

[0044] Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the terminal and the cloud in this invention. Detailed Implementation

[0045] Please refer to Figures 1 to 6 This invention provides a method and system for plague risk assessment combining the MaxEnt model, aiming to solve the technical problems of overfitting, poor interpretability, and high false alarm rate in existing MaxEnt models caused by the high subjectivity of manual selection of environmental variables and the high proportion of redundant variables. This embodiment describes the specific implementation process of the method in detail and discloses the system structure as necessary.

[0046] The method is based on the spatial coordinate data of historical plague outbreaks and multi-source environmental variable data covering the target area. It constructs an optimal variable subset through a three-stage automated variable selection mechanism, then trains the MaxEnt model to generate the plague niche probability distribution, and finally outputs a risk level classification map. The entire process strictly follows the preset S-step numbering and sequence.

[0047] First, perform step S1: Obtain the raw environmental variable dataset covering the target area. This raw environmental variable dataset includes temperature, precipitation, vegetation index, host animal population density, soil type, altitude, land use type, and human activity intensity indicators. The temperature data is sourced from meteorological satellite remote sensing products, with a spatial resolution of one kilometer and a temporal resolution of daily.

[0048] Precipitation data were derived from the Global Precipitation Measurement Project fusion product, with a spatial resolution of 10 km, and were uniformly reduced to 1 km after spatial interpolation; vegetation indices were obtained using the Normalized Difference Vegetation Index (NDVI), calculated from Landsat satellite imagery, with a spatial resolution of 30 m, and were resampled to 1 km; host animal population density data were derived from long-term monitoring records and field survey reports from local disease control centers, and were used to generate continuous surfaces through Kriging interpolation, with a spatial resolution of 1 km; soil type data were derived from the World Soil Database, were categorical variables, and were converted into multiple binary raster layers through unique thermal encoding.

[0049] Elevation data were derived from the digital elevation model of the Space Shuttle Radar Topographic Mapping Mission, with a spatial resolution of 90 meters, and resampled to one kilometer. Land use type data were derived from the global land cover dataset, also as categorical variables, and processed using one-hot encoding. The human activity intensity index was calculated by combining nighttime light index, road density, and population density data, with each sub-item uniformly scaled to one kilometer resolution before being weighted and summed. All data were stored in GeoTIFF format and had a unified geographic projection coordinate system, typically WGS84 or UTM projection coordinate system.

[0050] The following step, S2, involves spatially unifying the original environmental variable dataset and aligning it with geographic coordinates to form a standardized environmental variable matrix. This step first determines the minimum bounding rectangle of the target analysis area and then clips all environmental variable raster layers to this bounding rectangle. Next, all layers are resampled to a uniform spatial resolution, determined by the highest resolution data in the original dataset, which in this embodiment is one kilometer.

[0051] The resampling method is selected based on the variable type: for continuous variables such as temperature, precipitation, vegetation index, altitude and human activity intensity indicators, bilinear interpolation is used; for categorical variables such as soil type and land use type, nearest neighbor interpolation is used for each binary layer after one-heat encoding.

[0052] After resampling, all raster layers have the same number of rows and columns, with each pixel corresponding to a unique geographic coordinate. Finally, all raster layers are stacked by band to form a three-dimensional array with dimensions H×W×C, where H and W are the height and width of the image (i.e., the number of rows and columns), and C is the total number of environmental variables. This three-dimensional array is the standardized environmental variable matrix, serving as the unified input data structure for subsequent processing.

[0053] Next, step S3 is executed: The standardized environmental variable matrix undergoes initial variable screening in the first stage, using a variance thresholding method to remove environmental variables with variances below a preset threshold. This step first iterates through each band (i.e., each environmental variable) of the standardized environmental variable matrix, calculating the variance of all pixel values ​​in that band. The formula for calculating variance is... Where N is the total number of pixels, For the first The value of a pixel, This is the mean of all pixels in the band. After calculation, a variance vector of length C is obtained. Subsequently, the variance vector is sorted, and its tenths value is determined, which is the preset variance threshold. In this embodiment, if the original environmental variable dataset contains 20 variables, 20 variance values ​​are calculated, and the third smallest value (rounded up) is taken as the threshold.

[0054] If the variance of any environmental variable is less than this threshold, it is considered to lack significant spatial variability within the entire target area, has insufficient information, and cannot effectively distinguish the ecological characteristics of different areas, and is therefore eliminated. After this step, the remaining environmental variables constitute the initial set of environmental variables.

[0055] Then, step S4 is performed: The second stage of variable correlation reduction is applied to the initially screened environmental variables. The Pearson correlation coefficient between any two environmental variables is calculated. If its absolute value is greater than or equal to 0.7, the variable with a higher correlation to the spatial distribution of historical plague outbreaks is retained, and the other variable is removed. This step first constructs a correlation matrix with dimensions [missing information]. ,in This represents the number of environmental variables after initial screening. Each element in the matrix... Representing the The variable and the first The Pearson correlation coefficient between the variables is calculated using the following formula:

[0056] That For variables and covariance, and Let these be the standard deviations of the two variables. Iterate through the upper triangular portion of the correlation matrix (excluding the diagonal) and find all elements with an absolute value greater than or equal to 0.7. For each pair of highly correlated variables... Further analysis is needed to determine which factor is statistically more strongly associated with the occurrence of plague. To this end, the geographical coordinates of all plague outbreak locations were extracted from a historical plague outbreak database, and the corresponding environmental variable values ​​were queried from a standardized environmental variable matrix to form a positive sample set. Simultaneously, the global mean of each environmental variable was calculated across the entire target region.

[0057] Next, for each environmental variable, a two-sample t-test was performed to compare whether there was a significant difference between its mean value at the plague outbreak site and the global mean. The null hypothesis of the t-test was that there was no difference between the two, and the alternative hypothesis was that there was a difference. The calculated p-value was used as the criterion; variables with p-values ​​less than 0.05 were considered to be significantly correlated with the spatial distribution of the plague outbreak site.

[0058] In highly correlated pairs of variables, if only one variable satisfies the p-value condition, that variable is retained; if both variables satisfy the condition or neither satisfies it, the variable with the smaller p-value in the t-test is retained. This dual criterion effectively eliminates collinearity between variables while ensuring the biological significance of the retained variables.

[0059] Next, step S5 is executed: The reduced set of environmental variables undergoes a third-stage support vector machine (SVM) importance ranking based on recursive feature elimination. Historical plague outbreak points are used as positive samples, and randomly generated non-plague outbreak points are used as negative samples. The SVM classifier is trained and sorted in descending order based on the absolute weight of each environmental variable in the classification decision function. This step first constructs the training dataset. Positive samples are historical plague outbreak points, labeled +1; negative samples are randomly generated points within the target area, with a number five times that of positive samples, labeled -1.

[0060] For each sample point, all corresponding environmental variable values ​​are extracted from the standardized environmental variable matrix to form a feature vector. Then, a support vector machine classifier is initialized, using a radial basis function kernel, with a penalty parameter C set to 10 and the kernel parameter gamma set to the reciprocal of the dimension of the current input variable. A recursive feature elimination algorithm is then executed on this classifier.

[0061] The algorithm iterates as follows: In each iteration, a support vector machine classifier is trained using all remaining variables; after training, the weight coefficients corresponding to each input variable are extracted from the classifier's decision function (for non-linear kernels, their importance is approximated by calculating the average contribution of each feature to the support vectors); all variables are sorted in descending order according to the absolute value of the weight coefficients; the bottom 10% of variables are removed (rounded down, at least one is removed).

[0062] This process iterates until only one variable remains. In each iteration, the current set of variables and its corresponding importance ranking are recorded. Finally, the ranking results from all iterations are combined to obtain the final importance score of each variable before it is removed, thus forming a global importance descending list.

[0063] Then, step S6 is executed: the top N environmental variables from the ranking results are selected to form the optimal variable subset, where N is the minimum number of variables required for the cross-validation accuracy to plateau for the first time. This step utilizes the intermediate results of recursive feature elimination in step S5. For each set of variables remaining in each step of the recursive feature elimination process, a MaxEnt model is trained independently. The training parameters of the MaxEnt model are fixed: the number of background points is set to 10 times the number of historical plague occurrence points, the regularization multiplier is set to 1.5, the maximum number of iterations is set to 5000, and the convergence threshold is set to 0.00001. After each MaxEnt model is trained, its performance is evaluated using five-fold cross-validation.

[0064] Specifically, historical plague outbreak locations were randomly divided into 5 groups. Four groups were used as the training set and one as the validation set in each iteration. The area under the receiver operating characteristic (AUC) curve for the model on the validation set was calculated. This process was repeated 5 times, and the average AUC was calculated to obtain the average AUC curve for that set of variables. The average AUC curves for all variable sets (from all variables to individual variables) were then sorted in descending order of the number of variables, forming a performance curve.

[0065] Perform first-order difference calculations on the curve, i.e., calculate the slope between two adjacent points. When the absolute value of three consecutive difference values ​​is less than 0.015, the performance curve is considered to have entered a plateau phase and no longer significantly improves with the increase of the number of variables. Take N as the number of variables corresponding to the starting point of the plateau phase, and select the top N variables from the importance ranking list of step S5 to form the optimal variable subset.

[0066] Next, step S7 is executed: the optimal subset of variables is input into the MaxEnt model, and combined with the spatial coordinates of historical plague occurrence points, a plague niche probability distribution model is trained. This step uses the optimal subset of variables determined in step S6 to re-extract all environmental variable data, forming a new standardized environmental variable matrix. This matrix, along with the complete historical plague occurrence point coordinate data, is then input into the MaxEnt software or the self-developed MaxEnt algorithm implementation module.

[0067] The model training parameters remained consistent with those used during cross-validation: the number of background points was 10 times the number of historical occurrence points, uniformly and randomly sampled from the target region; the regularization multiplier was 1.5 to control model complexity and prevent overfitting; the maximum number of iterations was 5000; and the convergence threshold was 0.00001. After training, the MaxEnt model output a probability raster map, where each pixel value represents the relative probability of plague occurring at that geographic location, ranging from 0 to 1. This probability raster map is the plague niche probability distribution model.

[0068] Then, step S8 is executed: Based on the plague niche probability distribution model, the plague occurrence probability score is calculated for all geographical units within the target area, and low-risk, medium-risk, and high-risk areas are divided according to a preset risk level threshold. This step directly reads the probability raster map output from step S7. The value of each pixel is the plague occurrence probability score for that geographical unit.

[0069] The risk level threshold is set based on the probability score distribution of historical plague outbreak locations. Specifically, on the probability grid, the probability scores corresponding to the locations of all historical plague outbreak locations are queried to form a score set.

[0070] Calculate the 50th quantile (median) and 90th quantile of the set. Set the 90th quantile as the high-risk zone threshold and the 50th quantile as the medium-risk zone threshold. For any geographic unit within the target area, if its probability score is greater than or equal to the high-risk zone threshold, it is classified as a high-risk zone; if its probability score is greater than or equal to the medium-risk zone threshold but less than the high-risk zone threshold, it is classified as a medium-risk zone; and if its probability score is less than the medium-risk zone threshold, it is classified as a low-risk zone. This dynamic threshold setting method ensures that the risk classification matches the actual distribution characteristics of historical epidemics.

[0071] Finally, step S9 is executed: This step outputs a plague risk assessment map containing the risk level classification results. This step visualizes the risk level classification results obtained in step S8 as a color-coded thematic map. Typically, high-risk areas are represented in red, medium-risk areas in yellow, and low-risk areas in green. The map includes necessary legends, scale bars, a north arrow, and administrative boundaries. The final map product can be output in various formats such as GeoPDF, GeoTIFF, or Shapefile for use by public health decision-making departments, disease control centers, and researchers.

[0072] Based on the above methods, this invention also provides a plague risk assessment system incorporating the MaxEnt model. This system includes the following functional modules:

[0073] The raw environment variable acquisition module is used to execute step S1, which acquires and integrates various types of environmental variable data required from multi-source databases and remote sensing platforms.

[0074] The data standardization module is used to perform step S2, which spatially clips, resamples, projects, and stacks bands on the raw data to generate a standardized environmental variable matrix.

[0075] The variable screening module is used to execute step S3, calculate the variance of each variable, and perform preliminary screening based on the decimal threshold.

[0076] The correlation reduction module is used to perform step S4, calculate the Pearson correlation coefficient between variables, and reduce collinear variables by combining the p-value of the two-sample t-test.

[0077] The importance ranking module is used to execute step S5, build a support vector machine classifier and execute a recursive feature elimination algorithm to generate a variable importance ranking list.

[0078] The optimal variable subset determination module is used to execute step S6, which determines the optimal number of variables N through cross-validation and plateau detection, and selects the top N variables.

[0079] The MaxEnt model training module is used to perform step S7, which trains the MaxEnt model using the optimal subset of variables and historical occurrence data to generate a probability distribution grid.

[0080] The risk level classification module is used to execute step S8, dynamically calculate the risk threshold based on the probability score distribution of historical occurrence points, and classify the risk level of the entire region.

[0081] The risk map output module is used to execute step S9, which generates a standardized thematic map from the risk level classification results and outputs it.

[0082] All modules communicate via a unified data bus, sharing a standardized environmental variable matrix, historical event data, and intermediate calculation results. The system can be deployed on high-performance computing servers or cloud computing platforms to process large-scale geospatial data.

Claims

1. A plague risk assessment method combining the MaxEnt model, characterized in that, include: Obtain the original environmental variable dataset covering the target area; The original environmental variable dataset is spatially unified and aligned with geographic coordinates to form a standardized environmental variable matrix. The first stage of variable screening is performed on the standardized environmental variable matrix, and environmental variables with variances lower than a preset threshold are removed using the variance threshold method. The second stage of variable correlation reduction was performed on the environmental variables after the initial screening. The Pearson correlation coefficient between any two environmental variables was calculated. If the absolute value was greater than or equal to 0.7, the variable with a higher correlation to the spatial distribution of historical plague occurrence points was retained, and the other variable was removed. The third stage of importance ranking based on recursive feature elimination is performed on the reduced set of environmental variables. The support vector machine classifier is trained with historical plague occurrence points as positive samples and randomly generated non-occurrence points as negative samples. The environmental variables are then sorted in descending order according to the absolute value of their weights in the classification decision function. Select the top N environmental variables from the ranking results to form the optimal variable subset, where N is the minimum number of variables that makes the cross-validation accuracy reach the plateau period for the first time; The optimal subset of variables is input into the MaxEnt model, and combined with the spatial coordinates of historical plague occurrence points, a plague niche probability distribution model is trained. Based on the plague niche probability distribution model, the probability score of plague occurrence is calculated for all geographical units in the target area, and low-risk, medium-risk and high-risk areas are divided according to the preset risk level threshold. Output a plague risk assessment map that includes the risk level classification results.

2. The plague risk assessment method combining the MaxEnt model according to claim 1, characterized in that, The original environmental variable dataset includes temperature, precipitation, vegetation index, host animal population density, soil type, altitude, land use type, and human activity intensity indicators.

3. The plague risk assessment method combining the MaxEnt model according to claim 2, characterized in that, The standardized environmental variable matrix is ​​subjected to a first-stage variable screening, which involves using a variance thresholding method to remove environmental variables with variances below a preset threshold. This includes: Calculate the variance of each environmental variable in the standardized environmental variable matrix; The variances of all environmental variables are sorted in ascending order, and the tenths place value is taken as the preset threshold. Environmental variables with variances less than the preset threshold are removed to form a set of environmental variables after initial screening.

4. The plague risk assessment method combining the MaxEnt model according to claim 3, characterized in that, A second stage of correlation reduction was performed on the environmental variables after the initial screening, including: Construct the Pearson correlation coefficient matrix among environmental variables after initial screening; Traverse the Pearson correlation coefficient matrix to identify any two environmental variables whose absolute correlation coefficient is greater than or equal to 0.7; For each pair of variables, the significance level of the difference between the mean value at the location of the historical plague outbreak and the global mean was calculated, and the p-value was obtained by a two-sample t-test. Variables with p-values ​​less than 0.05 are retained. If both p-values ​​are less than 0.05 or neither is less than 0.05, the variable with the smaller p-value is retained.

5. The plague risk assessment method combining the MaxEnt model according to claim 4, characterized in that, The reduced set of environmental variables is then subjected to a third-stage support vector machine importance ranking based on recursive feature elimination, including: A training set was constructed using historical plague outbreak locations as positive samples and randomly generated non-outbreak locations as negative samples. Initialize the support vector machine classifier using the radial basis function kernel, set the penalty parameter C to 10, and set the kernel function parameter gamma to the reciprocal of the dimension of the current input variable; Recursive feature elimination is performed on the support vector machine classifier, and in each iteration, the last 10% of variables in importance are eliminated until the number of remaining variables is 1. Record the absolute weight of each environmental variable in the classification decision function in each iteration, and integrate all iteration results to generate a global importance descending sequence list.

6. The plague risk assessment method combining the MaxEnt model according to claim 5, characterized in that, The top N environmental variables are selected from the ranking results to form the optimal variable subset, where N is the minimum number of variables required to make the cross-validation accuracy reach a plateau for the first time. This subset includes: The MaxEnt model is trained for the set of variables remaining at each step in the recursive feature elimination process. Five-fold cross-validation was used to calculate the average area under the receiver operating characteristic curve for each MaxEnt model. Perform a first-order difference on the curve of the area under the receiver operating characteristic curve as a function of the number of variables; When the absolute value of the difference is less than 0.015 for three consecutive times, the judgment curve enters the plateau period, and the number of variables corresponding to the starting point of the plateau period is N.

7. The plague risk assessment method combining the MaxEnt model according to claim 6, characterized in that, The optimal subset of variables is input into the MaxEnt model, and combined with the spatial coordinates of historical plague occurrence points, to train a plague niche probability distribution model, including: Extract the data corresponding to the optimal subset of variables from the standardized environment variable matrix and reconstruct it into a new input matrix; The MaxEnt model is set to have 10 times the number of historical plague occurrences, a regularization multiplier of 1.5, a maximum number of iterations of 5000, and a convergence threshold of 0.00001. The MaxEnt model is trained based on the new input matrix and the coordinates of historical plague occurrence points, and the output is a raster map of the plague occurrence probability covering the target area as the plague niche probability distribution model.

8. The plague risk assessment method combining the MaxEnt model according to claim 7, characterized in that, Based on the plague niche probability distribution model, plague occurrence probability scores are calculated for all geographical units within the target area, and low-risk, medium-risk, and high-risk areas are divided according to preset risk level thresholds, including: Extract the probability scores corresponding to the locations of all historical plague occurrences from the plague niche probability distribution model; Calculate the 50th and 90th percentiles of the probability scores; The 90th percentile was set as the threshold for the high-risk zone, and the 50th percentile was set as the threshold for the medium-risk zone. For any geographical unit within the target area, if its probability score is greater than or equal to the high-risk area threshold, it is classified as a high-risk area; if its probability score is greater than or equal to the medium-risk area threshold but less than the high-risk area threshold, it is classified as a medium-risk area; if its probability score is less than the medium-risk area threshold, it is classified as a low-risk area.

9. The plague risk assessment method combining the MaxEnt model according to claim 8, characterized in that, Output a plague risk assessment map containing risk level classification results, including: High-risk areas, medium-risk areas, and low-risk areas are visualized using red, yellow, and green colors, respectively. Embed legends, scale bars, north arrows, and administrative boundaries into the map; Output the plague risk assessment map in GeoPDF, GeoTIFF, or Shapefile format.

10. A plague risk assessment system combining the MaxEnt model, characterized in that, include: The original environmental variable acquisition module is used to acquire the original environmental variable dataset covering the target area. The original environmental variable dataset includes temperature, precipitation, vegetation index, host animal population density, soil type, altitude, land use type, and human activity intensity index. The data standardization module is used to perform spatial resolution unification and geographic coordinate alignment on the original environmental variable dataset to form a standardized environmental variable matrix. The variable screening module is used to perform the first stage of variable screening on the standardized environmental variable matrix, and to remove environmental variables with variances lower than a preset threshold by using the variance threshold method. The correlation reduction module is used to perform the second stage of variable correlation reduction on the environmental variables after the initial screening. It calculates the Pearson correlation coefficient between any two environmental variables. If the absolute value is greater than or equal to 0.7, the variable with a higher correlation to the spatial distribution of historical plague occurrence points is retained, and the other variable is removed. The importance ranking module is used to perform the third stage of support vector machine importance ranking based on recursive feature elimination on the reduced set of environmental variables. It uses historical plague occurrence points as positive samples and randomly generated non-occurrence points as negative samples to train the support vector machine classifier and sorts them in descending order according to the absolute value of the weight of each environmental variable in the classification decision function. The optimal variable subset determination module is used to select the top N environmental variables from the ranking results to form the optimal variable subset, where N is the minimum number of variables that makes the cross-validation accuracy reach the plateau period for the first time. The MaxEnt model training module is used to input the optimal subset of variables into the MaxEnt model and, in combination with the spatial coordinates of historical plague occurrence points, train a plague niche probability distribution model. The risk level classification module is used to calculate the probability score of plague occurrence for all geographical units in the target area based on the plague niche probability distribution model, and to classify low-risk areas, medium-risk areas and high-risk areas according to the preset risk level threshold. The risk map output module is used to output a plague risk assessment map that includes risk level classification results.

Images