A spatiotemporal simulation method for zoonotic diseases based on geospatial big data

Abstract

The present invention provides a spatiotemporal simulation method for zoonoses based on geospatial big data, which is used to reconstruct missing information of spatiotemporal data sets of epidemic cases, including: constructing a data set of epidemic diseases to be tested; constructing a spatiotemporal data set of epidemiological transmission factors to be tested; constructing a prediction model for missing data of epidemiological diseases to be tested; inputting the spatiotemporal data set of epidemiological transmission factors to be tested and the data set of epidemic diseases to be tested into the prediction model for missing data of epidemiological diseases to be tested, and obtaining a feature matrix; and obtaining a prediction matrix according to the feature matrix. The present invention solves the problem of low computational efficiency of spatiotemporal data of multiple transmission factors of epidemics in the prior art and the problem of low accuracy of prediction of missing data of epidemics by generative adversarial networks due to failure to combine external factors.

Description

Technical Field

[0001] The present invention relates to the technical field of missing data prediction, and in particular to a spatiotemporal simulation method of zoonoses based on geospatial big data. Background Art

[0002] Global climate change, land use change and population growth have led to the continuous expansion of the geographical range of zoonoses. Collecting historical epidemiology to be tested, generating spatiotemporal data sets of epidemics at different spatial unit scales (such as districts, cities, provinces) and time scales (such as monthly, weekly), and analyzing the spatiotemporal pattern of epidemic spread are crucial for public health departments to formulate prevention and control strategies and strengthen the allocation of medical resources. In recent years, many countries have established national and regional disease surveillance systems. However, due to insufficient monitoring and lack of human resources, especially in remote or poor areas, the lack of epidemic data is still common. The lack of epidemic monitoring in some areas poses a challenge to the analysis of epidemic data for the entire country or region, and hinders the analysis of the spatial pattern of the epidemic, seasonal fluctuations, etc. In addition, public health departments also need to collect climate and environmental indicator data at different spatiotemporal scales, and combine the epidemic spatiotemporal data sets to conduct risk assessment and prediction of epidemic spread.

[0003] Missing data prediction for spatiotemporal datasets of zoonotic diseases is a difficult problem in the field of epidemiology. Commonly used statistical and machine learning methods interpolate missing values ​​by mining the spatial, temporal or spatiotemporal correlations between data, such as mean filling, kriging interpolation, inverse distance weighting, Gaussian process, autoregressive moving average and k nearest neighbor algorithm. These traditional methods played a certain role in the early days, but the effect is poor when the missing data rate in spatiotemporal datasets is high.

[0004] With the development of artificial intelligence, people have proposed missing data prediction methods based on deep learning, which usually show better results. The core idea of ​​existing data completion methods based on deep learning is to use known data to train the model, so that the model can learn the real distribution of the data and predict the missing data, such as autoencoders and generative adversarial networks.

[0005] However, existing methods based on generative adversarial networks to predict missing information in spatiotemporal data of infectious diseases do not take the external driving factors of the disease as model input features, that is, existing methods cannot simultaneously model the spatial correlation and external correlation of infectious diseases. Taking dengue fever as an example, it usually shows certain spatiotemporal aggregation characteristics. Its transmission and outbreak are affected by factors such as temperature and rainfall, and its spatial aggregation pattern is affected by factors such as local population density, vegetation distribution, elevation, and epidemics in neighboring regions. Therefore, in order to simulate the missing information of infectious spatiotemporal data sets, generative adversarial networks should consider the impact of external factors such as climate, environment, and epidemics in neighboring regions on local epidemics. Summary of the invention

[0006] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a spatiotemporal simulation method for zoonosis risk based on geospatial big data. The present invention solves the problem of low efficiency in calculating spatiotemporal data of multiple transmission factors of the epidemic in the prior art and the problem of low accuracy in predicting missing data due to the current generative adversarial network not combining the impact of external factors such as climate, environment and epidemics in neighboring areas on the local epidemic.

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

[0008] A spatiotemporal simulation method for zoonotic diseases based on geospatial big data, including:

[0009] Construct the disease dataset to be tested;

[0010] Constructing a spatiotemporal dataset of epidemiological transmission factors to be tested, wherein the spatiotemporal dataset of epidemiological transmission factors to be tested is a matrix with spatial units as rows and epidemiological weeks as columns;

[0011] Constructing a prediction model for missing epidemiological data to be tested, wherein the neural network structure of the prediction model for missing epidemiological data to be tested is a CNN-GAIN network structure;

[0012] Inputting the spatiotemporal data set of epidemiological transmission factors to be tested and the data set of epidemic diseases to be tested into the epidemiological missing data prediction model to obtain a feature matrix;

[0013] A prediction matrix is ​​obtained according to the feature matrix.

[0014] Preferably, the step of constructing a spatiotemporal dataset of epidemiological transmission factors to be tested according to the epidemiology to be tested includes:

[0015] Get the start and end dates of the epidemiological week to be tested;

[0016] Determine the grid layers corresponding to the various transmission factors of the epidemiology to be tested according to the start date, end date and transmission factors of the epidemiology week to be tested, the transmission factors including: temperature, rainfall, relative humidity, population density, vegetation distribution and elevation;

[0017] Constructing a space-time matrix corresponding to each transmission factor according to the grid layers corresponding to each transmission factor of the epidemiology to be tested;

[0018] A spatiotemporal data set of epidemiological transmission factors to be tested is constructed based on the spatiotemporal matrices corresponding to the various transmission factors.

[0019] Preferably, the step of constructing a space-time matrix corresponding to each of the transmission factors according to the grid layers corresponding to each of the transmission factors of the epidemiology to be tested includes:

[0020] Taking the epidemiological week as the time unit, performing temporal synthesis on the temperature, rainfall, relative humidity and vegetation distribution and taking the population density data as the weight in the spatial unit, performing population-weighted averaging on the temperature, rainfall, relative humidity and vegetation distribution to obtain a first spatiotemporal matrix, wherein the first spatiotemporal matrix is ​​a spatiotemporal matrix with spatial units as rows and epidemiological weeks as columns;

[0021] Performing population-weighted averaging on the elevation with population density as weight to obtain a second space-time matrix, where the second space-time matrix is ​​a space-time matrix with spatial units as rows and only one column;

[0022] Taking year as the time unit, calculating the average value of population density in each spatial unit to obtain a third space-time matrix, wherein the third space-time matrix is ​​a space-time matrix with spatial units as rows and years as columns;

[0023] The space-time matrices corresponding to the various propagation factors are constructed according to the first space-time matrix, the second space-time matrix and the third space-time matrix.

[0024] Preferably, the spatiotemporal data set of epidemiological transmission factors to be tested and the data set of epidemic diseases to be tested are input into the epidemiological missing data prediction model to obtain a feature matrix, including:

[0025] Obtain a noise matrix, a mask matrix, an epidemiological transmission factor matrix in the spatiotemporal data set of epidemiological transmission factors to be tested, and an epidemic disease matrix in the epidemic disease data set to be tested;

[0026] Determine the missing items of the disease matrix in the disease data set to be tested and initialize the missing items using different noise matrices to obtain an initialized disease matrix;

[0027] The initialized disease matrix, noise matrix, mask matrix and epidemiological transmission factor matrix are input into the epidemiological missing data prediction model to be tested to obtain a feature matrix.

[0028] Preferably, obtaining a prediction matrix according to the feature matrix comprises:

[0029] The feature matrix is ​​discriminated according to the discriminant function to obtain a missing matrix and a non-missing matrix;

[0030] Calculate the missing matrix using the first loss function to obtain a first output matrix;

[0031] Calculate the missing matrix using the second loss function to obtain a second output matrix;

[0032] Based on the training function, a prediction matrix is ​​obtained according to the first output matrix and the second output matrix.

[0033] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0034] The present invention provides a spatiotemporal simulation method for zoonoses based on geospatial big data. The network structure is a CNN-GAIN epidemiological missing data prediction model, epidemiological transmission factors are considered, a spatial matrix is ​​constructed, and the matrix is ​​input into the prediction model to realize the prediction of missing data. The present invention improves the prediction effect when the data missing rate in the spatiotemporal dataset is high and when the generative adversarial network has a greater impact on the local epidemic due to the failure to combine external factors such as climate, environment and epidemics in neighboring areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 A flow chart of a spatiotemporal simulation method for zoonotic diseases based on geospatial big data provided by an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of a technical route for completing missing values ​​in an epidemic spatiotemporal dataset based on geospatial big data cloud computing and deep learning provided by an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of feature learning based on CNN provided in an embodiment of the present invention;

[0039] Figure 4 A schematic diagram of a coupled CNN and GAIN model provided by an embodiment of the present invention;

[0040] Figure 5 A comparison chart of the prediction effects of the model of the present invention and the benchmark model provided in the embodiment of the present invention. DETAILED DESCRIPTION

[0041] 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.

[0042] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0043] The terms "first", "second", "third" and "fourth" in the specification and claims of the present application and the drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a series of steps, processes, methods, etc. are not limited to the listed steps, but may optionally include steps that are not listed, or may optionally include other step elements inherent to these processes, methods, products or devices.

[0044] The purpose of the present invention is to provide a spatiotemporal simulation method for zoonoses based on geospatial big data. The present invention solves the problem of poor prediction effect in the prior art due to a high data missing rate in spatiotemporal data sets and the problem of low accuracy in predicting missing data due to the current generative adversarial network not combining the impact of external factors such as climate, environment and epidemics in neighboring areas on local epidemics.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] like Figure 1 As shown, the present invention provides a spatiotemporal simulation method of zoonotic diseases based on geospatial big data, comprising:

[0047] Step 100: construct a disease dataset to be tested;

[0048] Step 200: constructing a spatiotemporal dataset of epidemiological transmission factors to be tested, wherein the spatiotemporal dataset of epidemiological transmission factors to be tested is a matrix with spatial units as rows and epidemiological weeks as columns;

[0049] Step 300: constructing a prediction model for missing epidemiological data to be tested, wherein the neural network structure of the prediction model for missing epidemiological data to be tested is a CNN-GAIN network structure;

[0050] Step 400: inputting the spatiotemporal data set of epidemiological transmission factors to be tested and the data set of epidemic diseases to be tested into the epidemiological missing data prediction model to obtain a feature matrix;

[0051] Step 500: Obtain a prediction matrix according to the feature matrix.

[0052] Further, such as Figure 2 As shown, the spatiotemporal data set of epidemiological transmission factors to be tested is constructed according to the epidemiology to be tested, including:

[0053] Get the start and end dates of the epidemiological week to be tested;

[0054] Determine the grid layers corresponding to the various transmission factors of the epidemiology to be tested according to the start date, end date and transmission factors of the epidemiology week to be tested, the transmission factors including: temperature, rainfall, relative humidity, population density, vegetation distribution and elevation;

[0055] Constructing a space-time matrix corresponding to each transmission factor according to the grid layers corresponding to each transmission factor of the epidemiology to be tested;

[0056] A spatiotemporal data set of epidemiological transmission factors to be tested is constructed based on the spatiotemporal matrices corresponding to the various transmission factors.

[0057] Furthermore, the step of constructing a space-time matrix corresponding to each of the transmission factors according to the grid layers corresponding to each of the transmission factors of the epidemiology to be tested includes:

[0058] Taking the epidemiological week as the time unit, performing temporal synthesis on the temperature, rainfall, relative humidity and vegetation distribution and taking the population density data as the weight in the spatial unit, performing population-weighted averaging on the temperature, rainfall, relative humidity and vegetation distribution to obtain a first spatiotemporal matrix, wherein the first spatiotemporal matrix is ​​a spatiotemporal matrix with spatial units as rows and epidemiological weeks as columns;

[0059] Performing population-weighted averaging on the elevation with population density as weight to obtain a second space-time matrix, where the second space-time matrix is ​​a space-time matrix with spatial units as rows and only one column;

[0060] Taking year as the time unit, calculating the average population density in each spatial unit to obtain a third space-time matrix, wherein the third space-time matrix is ​​a space-time matrix with spatial units as rows and years as columns;

[0061] The space-time matrices corresponding to the various propagation factors are constructed according to the first space-time matrix, the second space-time matrix and the third space-time matrix.

[0062] Specifically, taking the dengue fever case as an example, the method includes:

[0063] Get all epidemiological weeks in the study period, and then get the start and end dates of each week;

[0064] Table 1 shows the driving factors of dengue fever transmission and their data sources. As shown in Table 1, the four factors of temperature, rainfall, relative humidity and vegetation index have high temporal resolution. Therefore, all raster layers between the start date and the end date of each epidemiological week are first read for each factor, and the corresponding time synthesis method is used to perform temporal synthesis of the raster layers to form an image set with the epidemiological week as the temporal resolution, that is, each factor has a raster layer in each epidemiological week.

[0065] Since dengue fever mainly occurs in human-dominated urbanized and semi-urbanized areas, the weekly image sets of temperature, rainfall, relative humidity, and vegetation index synthesized with epidemiological weeks as the time unit were population-weighted averaged within the spatial unit using population density data as weights to form a spatiotemporal matrix with spatial units as rows and epidemiological weeks as columns for temperature, rainfall, relative humidity, and vegetation index, respectively.

[0066] Elevation data does not change over time, only spatial differences exist, so no temporal synthesis is performed. Also considering that dengue fever mainly occurs in human-dominated areas, the elevation data is only weighted by population density, forming a space-time matrix with spatial units as rows and only one column.

[0067] The temporal resolution of population density is annual, that is, the population density grid data of all epidemiological weeks in each year is unchanged. Therefore, with year as the time unit, the average population density in each spatial unit is calculated. Finally, a spatiotemporal matrix with spatial units as rows and years as columns is formed for population density.

[0068] The above spatiotemporal matrices of temperature, rainfall, relative humidity, vegetation index, elevation, and population density have different column sizes. By duplicating the columns of the elevation matrix and duplicating the columns of population density within the year, the matrices of all factors are adjusted to a matrix with the number of spatial units as the number of rows and the number of epidemiological weeks as the number of columns. Table 1 is shown below:

[0069] Table 1 Drivers of dengue transmission and their data sources

[0070]

[0071] Furthermore, the spatiotemporal data set of epidemiological transmission factors to be tested and the data set of epidemic diseases to be tested are input into the epidemiological missing data prediction model to obtain a feature matrix, including:

[0072] Obtain a noise matrix, a mask matrix, an epidemiological transmission factor matrix in the spatiotemporal data set of epidemiological transmission factors to be tested, and an epidemic disease matrix in the epidemic disease data set to be tested;

[0073] Determine the missing items of the disease matrix in the disease data set to be tested and initialize the missing items using different noise matrices to obtain an initialized disease matrix;

[0074] The initialized disease matrix, noise matrix, mask matrix and epidemiological transmission factor matrix are input into the epidemiological missing data prediction model to be tested to obtain a feature matrix.

[0075] Furthermore, obtaining a prediction matrix according to the feature matrix includes:

[0076] The feature matrix is ​​discriminated according to the discriminant function to obtain a missing matrix and a non-missing matrix;

[0077] Calculate the missing matrix using the first loss function to obtain a first output matrix;

[0078] Calculate the missing matrix using the second loss function to obtain a second output matrix;

[0079] Based on the training function, a prediction matrix is ​​obtained according to the first output matrix and the second output matrix.

[0080] This example takes dengue fever cases as an example. First, the dengue fever case data is represented as a matrix. A data matrix X containing dengue fever cases is defined, where a row represents a spatial unit and a column represents a timestamp (an epidemiological week). An entry X ij Refers to the value of the i-th spatial unit at the j-th timestamp. Since the epidemic monitoring system of some spatial units (some areas) is imperfect or missing, the spatiotemporal set of cases is incomplete. The task of this article is to complete the missing values ​​in the disease matrix. Therefore, in the disease matrix X, this embodiment calls the missing areas missing items, and the areas with available (original) values ​​are called observation items. Then this embodiment defines a mask matrix M, where M is a 01 matrix, indicating whether the value at the corresponding position of X is missing, and is defined as follows:

[0081]

[0082] As for the driving factor data, this embodiment can also define a data set Y={Y 1 ,Y 2 ,…,Y S}, where S represents different driving factors, and the dimension of each Y matrix is ​​the same as that of the X matrix, which are matrices reconstructed based on the distance space between spatial units.

[0083] like Figure 3As shown, CNN-based feature learning. The purpose of introducing CNN is to extract the spatiotemporal correlation within the disease and the correlation between the disease and environmental factor data. First, the missing items of each disease matrix are initialized with different noise variables, which are sampled from a normal distribution or a uniform distribution. Then, this embodiment can regard the disease matrix, mask matrix, noise matrix, and driver variable matrix as an N-channel image, where the two dimensions of the image represent time and space, and the number of channels represents different data types. Finally, this embodiment can input this image into a CNN-based representation, namely C, to obtain a feature matrix X'.

[0084] like Figure 4 As shown in the figure, the GAIN network structure. The obtained feature matrix is ​​input into the generator to obtain the complete matrix after completion. The discriminator, like the traditional GAN ​​framework, will serve as an opponent to train G. But unlike the traditional GAN, the output of the generator has only two options: real or false, that is, the output is a scalar. Under the model of this embodiment, the output is a combination of observed and inferred entries, so it is equivalent to outputting a vector, because D is no longer intended to identify whether the input data is completely real or completely false, but to try to distinguish which entries are observed and which are inferred, which is equivalent to predicting the mask matrix M. Finally, the discriminator is a function D: i th The component corresponding to X'i th The probability of a component being observed. The higher the probability value, the more likely the item is to be observed, that is, the lower the probability that the item is missing.

[0085] For model training, the minimax optimization problem is solved iteratively. The generator G and the discriminator D have the same structure, including multiple convolutional layers and fully connected neural networks. Therefore, for the generator G, its loss function consists of two parts. First, By defining:

[0086]

[0087] For the loss function in the second part, Defined as:

[0088]

[0089] Among them, L M (x ij ,x' ij )=(x' ij -x ij )2 Represents the generator G predicted value x' ij With the true value x ij The square error between them, M is the element in the above mask matrix, The mask matrix predicted by the discriminator D;

[0090] From the definitions of these two loss functions, we can see that Applied to missing entries (M ij =0), Applied to entries without missing items (M ij =1).

[0091] When M ij When is 0, if is close to 1, then That is, when D cannot recognize the completed value as a completed value, is smaller (it incorrectly classifies it as the true value). When the reconstructed feature is close to the actual true feature value, is smaller, that is, the true feature output by the generator G. G is then trained to minimize the weighted sum of the two losses, as follows:

[0092]

[0093] Where α is a hyperparameter of the model.

[0094] This embodiment uses two indicators, root mean square error (RMSE) and mean absolute error (MAE), to evaluate the prediction performance, as shown below:

[0095]

[0096]

[0097] where x' ij is the inferred value, x ij is the true value and n is the number of inferred missing values.

[0098] This embodiment collects the number of dengue cases, temperature, rainfall, relative humidity, vegetation index and elevation of 418 epidemiological weeks in 558 spatial units (micro-regions, one of Brazil's geostatistical units) across Brazil from 2013 to 2020, forming a 558*418 matrix of dengue cases and driving factors. The missing situation of the spatiotemporal data set of the epidemic is simulated in the following way: a given proportion (R%) of units are randomly selected from all spatial units, and all their case data are set to null values, thereby simulating the situation that some areas cannot realize epidemic data monitoring and collection due to limited human and economic resources. In order to illustrate the accuracy of the method of this patent, based on the same data set and missing rate, this patent applies several commonly used benchmark models for comparative analysis, including Average, DME and GAIN. Average is to take the average value of non-missing data at each timestamp to fill in the missing data. The DME (Deep Multimodal Encoding) model is an improvement on the traditional autoencoder model. Multi-channel data can be input into the encoder. It can learn the correlation between different types of data and interact with each other's information, so as to achieve better completion effect. GAIN is a method based on the traditional generative adversarial model. It only uses simple fully connected neural networks in the generator and discriminator, and only considers dengue fever data, ignoring external driving factors.

[0099] Figure 5 The prediction effect comparison between the model of the present invention and the benchmark model is shown in Figure 2. Figure 5 As shown, it can be clearly shown that the prediction accuracy of the model of this embodiment is improved compared with the existing model. Compared with other models, the CNN-GAIN proposed in this patent can more reasonably capture the seasonal changes of dengue fever in the missing areas, the peak of the epidemic and other characteristics, and is closer to the real data. This shows that its completion effect can be reliably applied to the subsequent spatiotemporal analysis of the infectious disease transmission pattern. In addition, from the perspective of model accuracy evaluation, Table 2 is a comparison of the accuracy of the model proposed in this patent and the commonly used completion model. Under different missing rates, the CNN-GAIN proposed in this patent has significant advantages, which illustrates the predictive ability of the model. Table 2 is shown below:

[0100] Table 2 Accuracy comparison between the model proposed in this patent and the commonly used completion model

[0101]

[0102] The beneficial effects of the present invention are as follows:

[0103] Taking dengue fever, a typical zoonosis caused by viruses, as an example, the present invention makes the following improvements to the current prediction technology of missing information in the spatiotemporal dataset of epidemic diseases: (1) Model characteristics: The spatial correlation of epidemics in neighboring regions and the correlation between epidemics and driving factors are simultaneously incorporated into the prediction model of missing information in the spatiotemporal dataset of epidemic diseases; (2) Driving factors: The geospatial big data cloud computing is reasonably applied to the spatiotemporal characterization of multiple driving factors. The present invention improves the prediction effect when the data missing rate of the spatiotemporal dataset of epidemic diseases is high and when the generative adversarial network does not combine the influence of external factors such as climate, environment and epidemics in neighboring regions on the local epidemic.

[0104] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0105] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A spatiotemporal simulation method for zoonotic diseases based on geospatial big data, characterized in that: include: Construct the disease dataset to be tested; Constructing a spatiotemporal dataset of epidemiological transmission factors to be tested, wherein the spatiotemporal dataset of epidemiological transmission factors to be tested is a matrix with spatial units as rows and epidemiological weeks as columns; Constructing a prediction model for missing epidemiological data to be tested, wherein the neural network structure of the prediction model for missing epidemiological data to be tested is a CNN-GAIN network structure; Inputting the spatiotemporal data set of epidemiological transmission factors to be tested and the data set of epidemic diseases to be tested into the epidemiological missing data prediction model to obtain a feature matrix; Obtaining a prediction matrix according to the feature matrix; According to the epidemiology to be tested, a spatiotemporal dataset of epidemiological transmission factors to be tested is constructed, including: Get the start and end dates of the epidemiological week to be tested; Determine the grid layers corresponding to the various transmission factors of the epidemiology to be tested according to the start date, end date and transmission factors of the epidemiology week to be tested, the transmission factors including: temperature, rainfall, relative humidity, population density, vegetation distribution and elevation; Constructing a space-time matrix corresponding to each transmission factor according to the grid layers corresponding to each transmission factor of the epidemiology to be tested; A spatiotemporal data set of epidemiological transmission factors to be tested is constructed based on the spatiotemporal matrices corresponding to the various transmission factors.

2. The spatiotemporal simulation method of zoonotic diseases based on geospatial big data according to claim 1, characterized in that: The step of constructing a space-time matrix corresponding to each of the transmission factors according to the grid layers corresponding to each of the transmission factors of the epidemiology to be tested includes: Taking the epidemiological week as the time unit, performing temporal synthesis on the temperature, rainfall, relative humidity and vegetation distribution and taking the population density data as the weight in the spatial unit, performing population-weighted averaging on the temperature, rainfall, relative humidity and vegetation distribution to obtain a first spatiotemporal matrix, wherein the first spatiotemporal matrix is ​​a spatiotemporal matrix with spatial units as rows and epidemiological weeks as columns; Performing population-weighted averaging on the elevation with population density as weight to obtain a second space-time matrix, where the second space-time matrix is ​​a space-time matrix with spatial units as rows and only one column; Taking year as the time unit, calculating the average value of population density in each spatial unit to obtain a third space-time matrix, wherein the third space-time matrix is ​​a space-time matrix with spatial units as rows and years as columns; The space-time matrices corresponding to the various propagation factors are constructed according to the first space-time matrix, the second space-time matrix and the third space-time matrix.

3. The spatiotemporal simulation method of zoonotic diseases based on geospatial big data according to claim 1, characterized in that: The epidemiological missing data prediction model to be tested is input with the spatiotemporal data set of the epidemiological transmission factors to be tested and the disease data set to be tested to obtain a feature matrix, including: Obtain a noise matrix, a mask matrix, an epidemiological transmission factor matrix in the spatiotemporal data set of epidemiological transmission factors to be tested, and an epidemic disease matrix in the epidemic disease data set to be tested; Determine the missing items of the disease matrix in the disease data set to be tested and initialize the missing items using different noise matrices to obtain an initialized disease matrix; The initialized disease matrix, noise matrix, mask matrix and epidemiological transmission factor matrix are input into the epidemiological missing data prediction model to be tested to obtain a feature matrix.

4. The spatiotemporal simulation method of zoonotic diseases based on geospatial big data according to claim 1, characterized in that: The step of obtaining a prediction matrix according to the feature matrix comprises: The feature matrix is ​​discriminated according to a discriminant function to obtain a missing matrix and a non-missing matrix; Calculate the missing matrix using the first loss function to obtain a first output matrix; Calculate the missing matrix using the second loss function to obtain a second output matrix; Based on the training function, a prediction matrix is ​​obtained according to the first output matrix and the second output matrix.

Images