CN118051845A - Geospatial full coverage data generation method and device based on space variable parameter machine learning - Google Patents

Abstract

The invention provides a geographic space full-coverage data generation method and device based on space variable parameter machine learning, and relates to the technical field of geographic information science. The method comprises the following steps: gradually partitioning a target area, and calculating the spatial layering heterogeneity of the relationship between various auxiliary variables and target variables in the current partitioning state based on various auxiliary variables and target variables in each observation site in the target area after each partitioning; determining a target partition state based on the spatial hierarchical heterogeneity; under a target partition state, respectively constructing a space variable parameter machine learning model aiming at each subarea in the target area; and respectively carrying out interpolation prediction on target variables of each preset point to be predicted in the target region based on each space variable parameter machine learning model corresponding to each sub-region to obtain an interpolation prediction result and an uncertainty analysis result. According to the method, the spatial distribution map corresponding to the accurate geospatial full coverage data can be interpolated according to the limited observation site data.

Description

Method and device for generating geospatial full coverage data based on spatially variable parameter machine learning

Technical Field

The present invention relates to the field of geographic information science and technology, and in particular to a method and device for generating geographic space full coverage data based on space-varying parameter machine learning.

Background technique

Spatial interpolation is often used to convert the measurement data of discrete points into continuous data surfaces for comparison with the distribution patterns of other spatial phenomena. The establishment of meteorological and soil observation sites is usually costly and limited in number, and it is impossible to obtain continuous surface data in the study area. Therefore, spatial interpolation technology is often used to extrapolate the data of a limited number of observation sites to the entire study area. With the continuous promotion of spatial data interpolation technology, its application has been widely developed, which has led to higher requirements for the accuracy of this technology. However, due to the complex and changeable geographical environment, when performing large-scale high-spatial-resolution interpolation, when the target variable is affected by multiple auxiliary variables, and when there is strong spatial heterogeneity in the relationship between the target variable and the auxiliary variables, the accuracy of the interpolation results obtained by traditional spatial interpolation methods is low.

Under the premise of assuming that the local spatial relationship between the target variable and the auxiliary variable is stable, the researchers developed a geographically weighted regression model and a Bayesian spatially variable coefficient model to use spatial autocorrelation to simulate spatial non-stationary relationships. However, the former relies heavily on a predefined spatial kernel function and is easily affected by collinearity, which in turn affects the accuracy of the interpolation results; the latter requires a predefined distribution for each coefficient of the model and sets a mutual covariance function for the spatial random part of the coefficient, which is difficult to set correctly, and incorrect settings will affect the accuracy of the interpolation results.

Therefore, in the context of large-scale high spatial resolution interpolation, how to interpolate accurate spatial distribution maps of meteorological or soil data based on limited observation site data is an urgent problem to be solved.

Summary of the invention

The present invention provides a method and device for generating geographic space full coverage data based on spatial variable parameter machine learning, so as to solve the defect of low precision of spatial interpolation results of meteorological or soil data in geographic space in the prior art, and realize spatial interpolation of high-precision geographic space full coverage data.

The present invention provides a method for generating geographic space full coverage data based on space-varying parameter machine learning, comprising:

The target area is partitioned step by step, and after each partition, based on various auxiliary variables and target variables in each observation station in the target area, the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and the target variables in the current partition state is calculated;

Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, determine the target partition state, the target partition state corresponds to multiple sub-regions of the target region;

In the target partition state, a spatially varying parameter machine learning model is constructed for each sub-region in the target region, each of which is obtained by training based on the target variables and auxiliary variables of each observation site within the corresponding spatial range, and the distance from each observation site to the corresponding spatially varying parameter machine learning model; each of the spatially varying parameter machine learning models includes position information and spatial range information, the position information is the center coordinates of the sub-region corresponding to the spatially varying parameter machine learning model, and the spatial range information is the size of the sub-region corresponding to the spatially varying parameter machine learning model;

Based on the spatially variable parameter machine learning models corresponding to the sub-regions, interpolation prediction is performed on the target variables of the preset points to be predicted in the target area to obtain target interpolation results and uncertainty analysis results.

According to a method for generating geographic space full coverage data based on space-varying parameter machine learning provided by the present invention, after each partition, based on various auxiliary variables and target variables in each observation station in the target area, the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and the target variable in the current partition state is calculated, including:

For each observation site in the target area, calculating the bivariate local spatial autocorrelation coefficient between the target variable in the observation site and each type of auxiliary variable;

After the target area is partitioned each time, the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the current partition state is calculated based on the local autocorrelation coefficient of each bivariate.

According to a method for generating geographic space full coverage data based on space-varying parameter machine learning provided by the present invention, for each observation site in the target area, calculating the bivariate local spatial autocorrelation coefficient between the target variable in the observation site and each type of auxiliary variable, respectively, includes:

For each of the observation sites in the target area, based on the auxiliary variables of the observation site and the target variables of a first preset number of neighboring observation sites of the observation site, the bivariate local spatial autocorrelation coefficient corresponding to the observation site is calculated.

According to a method for generating geographic space full coverage data based on spatially varying parameter machine learning provided by the present invention, each time the target area is partitioned, the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the current partition state is calculated based on the local autocorrelation coefficient of each bivariate, including:

After the target area is partitioned each time, based on the local spatial autocorrelation coefficients of the two variables corresponding to all the observation sites in each partition space under the current partition state, the local autocorrelation index variance values of the relationships between various auxiliary variables and the target variables in each partition space are calculated;

Based on the bivariate local spatial autocorrelation coefficients corresponding to all the observation sites in the target area, the global autocorrelation index variance values of the relationships between various auxiliary variables and the target variable in the target area are calculated;

The spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the target area under the current partition state is calculated based on the local autocorrelation index variance value and the global autocorrelation index variance value.

According to a method for generating geographic space full coverage data based on spatially variable parameter machine learning provided by the present invention, the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and the target variable after each partition is used to determine the target partition state, including:

Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, the average spatial hierarchical heterogeneity corresponding to the target area after each partition is calculated;

Constructing a first change curve based on the average spatial stratified heterogeneity corresponding to the target area after each partition, or constructing a second change curve based on the difference between the average spatial stratified heterogeneity after two adjacent partitions;

A partition state corresponding to an inflection point of the first change curve or an inflection point of the second change curve is determined as a target partition state.

According to a method for generating geographic space full coverage data based on spatially varying parameter machine learning provided by the present invention, each of the spatially varying parameter machine learning models corresponding to each of the sub-regions is trained based on the following method:

Obtain target variables and auxiliary variables of all the observation sites in the sub-region;

Determine the distances from each of the observation sites in the sub-area to the center point of the sub-area;

A random forest model is trained based on the target variables and auxiliary variables of all the observation sites in the sub-region, and the distances from each observation site in the sub-region to the center point of the sub-region, and the trained random forest model is used as the spatially varying parameter machine learning model corresponding to the sub-region.

According to a method for generating geographic space full coverage data based on space-varying parameter machine learning provided by the present invention, the target variables of each to-be-predicted point preset in the target area are interpolated and predicted respectively based on each space-varying parameter machine learning model corresponding to each sub-area, and a target interpolation result and an uncertainty analysis result are obtained, including:

Using multiple spatially varying parameter machine learning models to respectively predict the target variable of the point to be predicted, and obtaining multiple interpolation prediction results, wherein the multiple spatially varying parameter machine learning models include a second preset number of spatially varying parameter machine learning models adjacent to the point to be predicted;

Based on the inverse distance weighted method, the target interpolation result and the uncertainty analysis result of the target variable of the point to be predicted are determined according to the multiple interpolation prediction results.

The present invention also provides a device for generating geographic space full coverage data based on space-varying parameter machine learning, comprising:

A partitioning module is used to partition the target area step by step, and after each partition, based on various auxiliary variables and target variables in each observation station in the target area, calculate the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and target variables in the current partition state;

A determination module, configured to determine a target partition state based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, wherein the target partition state corresponds to a plurality of sub-regions of the target region;

A modeling module is used to construct a spatially varying parameter machine learning model for each sub-region in the target region under the target partition state, each of the spatially varying parameter machine learning models is obtained by training based on the target variables and auxiliary variables of each observation site within the corresponding spatial range, and the distance from each observation site to the corresponding spatially varying parameter machine learning model; each of the spatially varying parameter machine learning models includes position information and spatial range information, the position information is the center coordinates of the sub-region corresponding to the spatially varying parameter machine learning model, and the spatial range information is the size of the sub-region corresponding to the spatially varying parameter machine learning model;

The prediction module is used to perform interpolation prediction on the target variables of each preset point to be predicted in the target area based on the spatially variable parameter machine learning models corresponding to each sub-area, so as to obtain the target interpolation result and uncertainty analysis result.

The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any of the above-described methods for generating geographic space full coverage data based on spatially variable parameter machine learning.

The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for generating geographic space full coverage data based on spatially variable parameter machine learning as described in any of the above is implemented.

The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for generating geographic space full coverage data based on spatially variable parameter machine learning.

The spatial interpolation method and device of spatially variable parameter machine learning provided by the present invention, for the case where there are attribute similarity and spatial autocorrelation between the target variable and various auxiliary variables, and the heterogeneity of the relationship between the target variable and various auxiliary variables is strong, the target area is partitioned based on spatial hierarchical heterogeneity to obtain multiple sub-areas, and the target area prediction points are predicted based on the spatially variable parameter machine learning model corresponding to each sub-area to obtain the target interpolation result. Since each spatially variable parameter machine learning model constructed for each sub-area has a specific position and range, the spatial non-stationary relationship between the target variable and the auxiliary variable can be effectively simulated. At the same time, the spatially variable parameter machine learning model is established based on the target variable, the auxiliary variable and the distance of the observation site from the spatially variable parameter machine learning model position of the observation site, so the local spatial changes in the relationship between the target variable and the auxiliary variable can be automatically modeled, and then the spatial correlation and heterogeneity of the multi-dimensional mixed type auxiliary variables are fully utilized. In the large-scale high spatial resolution interpolation scenario, the target variables of each point to be predicted can be accurately interpolated according to the limited observation site data, and then the accurate spatial distribution map corresponding to the full coverage data of the geographic space is obtained.

BRIEF DESCRIPTION OF THE DRAWINGS

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

FIG1 is a flow chart of a method for generating geographic space full coverage data based on space-varying parameter machine learning according to an embodiment of the present invention;

FIG2 is a second flow chart of a method for generating geographic space full coverage data based on space-varying parameter machine learning provided by an embodiment of the present invention;

3 is a schematic diagram of a first curve and a second curve constructed based on average spatial hierarchical heterogeneity provided by an embodiment of the present invention;

FIG4 is a schematic diagram of each sub-area in a target partition state provided by an embodiment of the present invention;

5 is a schematic diagram of the structure of a device for generating geographic space full coverage data based on space-varying parameter machine learning provided by an embodiment of the present invention;

FIG. 6 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

Aiming at the situation where both attribute similarity and spatial autocorrelation exist between target variables and auxiliary variables, and the relationship heterogeneity between the target variables and auxiliary variables is strong, the present invention studies a method for generating geographic space full coverage data based on spatially varying parameter machine learning, which can fully utilize the spatial correlation and heterogeneity laws of multi-dimensional mixed-type auxiliary variables, so as to improve the accuracy of large-scale high spatial resolution interpolation mapping.

In view of the technical problems existing in the prior art, the embodiment of the present application provides a method for generating geographic space full coverage data based on space-varying parameter machine learning. FIG1 is one of the flow diagrams of the method for generating geographic space full coverage data based on space-varying parameter machine learning provided by an embodiment of the present invention. As shown in FIG1, the method includes:

Step 110: partition the target area step by step, and after each partition, based on various auxiliary variables and target variables in each observation station in the target area, calculate the spatial hierarchical heterogeneity of the relationship between each auxiliary variable and the target variable in the current partition state.

Specifically, FIG. 2 is a second flow chart of a method for generating geographic spatial full coverage data based on spatially variable parameter machine learning provided by an embodiment of the present invention. As shown in FIG. 2 , the target area can be recursively grid-divided based on the quadtree recursive principle, that is, the target area is equally divided into four areas, and the four areas are further divided into four blocks in a preset order such as upper left, upper right, lower left, and lower right. Each division of an area into four blocks is a partition of the target area, and the partitioning is performed step by step. After each partitioning, the spatial hierarchical heterogeneity of the relationship between each auxiliary variable and the target variable in the current partition state needs to be calculated, wherein, after each partitioning, a grid area of the target area represents a partition space.

In one embodiment, after each partitioning, based on various auxiliary variables and target variables in each observation station in the target area, the spatial hierarchical heterogeneity of the relationship between each auxiliary variable and the target variable in the current partitioning state is calculated, including:

For each observation site in the target area, calculating the bivariate local spatial autocorrelation coefficient between the target variable in the observation site and each type of auxiliary variable;

After the target area is partitioned each time, the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the current partition state is calculated based on the local autocorrelation coefficient of each bivariate.

Specifically, as shown in Figure 2, the bivariate local spatial autocorrelation coefficient is determined based on the target variable and its influencing factor data, where the influencing factor data refers to the auxiliary variables that affect the target variable. The bivariate local spatial autocorrelation coefficient (bivariate LISA) can measure the local heterogeneity of the correlation between two variables. After calculating the bivariate LISA values between the target variable (Y) and each type of auxiliary variable (X) at all observation sites, the spatial pattern of the relationship between the target variable (Y) and each type of auxiliary variable (X) can be obtained based on the spatial distribution of the bivariate LISA values.

In one embodiment, for each observation site in the target area, calculating the bivariate local spatial autocorrelation coefficient between the target variable in the observation site and each type of auxiliary variable, respectively, includes:

For each of the observation sites in the target area, based on the auxiliary variables of the observation site and the target variables of a first preset number of neighboring observation sites of the observation site, the bivariate local spatial autocorrelation coefficient corresponding to the observation site is calculated.

Specifically, each observation site can have one target variable and multiple types of auxiliary variables. When calculating the bivariate local spatial autocorrelation coefficients between the target variable and each type of auxiliary variable, the target variable and each type of auxiliary variable at each observation site are first standardized. The standardization process is explained using the target variable at an observation site as an example: the target variable of the observation site is subtracted from the mean of the target variable in the target area, and then divided by the standard deviation of the target variable in the target area to obtain the standardized target variable.

After standardizing each target variable and each auxiliary variable, the bivariate local Moran's index is calculated for each observation site, and the bivariate local Moran's index is used as the bivariate local spatial autocorrelation coefficient. For each observation site, the bivariate local Moran's index at the observation site is calculated based on the auxiliary variables at the observation site and the target variables at the first preset number of neighboring observation sites of the observation site. Among them, for one observation site, one type of auxiliary variable and the target variable correspond to one bivariate local Moran's index, that is, when the observation site has 5 types of auxiliary variables, 5 bivariate local Moran's indexes will be determined for the observation site.

Optionally, the bivariate local Moran index at each observation station in the target area It can be calculated by the following formula:

(1)

Among them, in the above formula, Indicates the observation site on the target area B/> The bivariate local Moran index of The value range is [-1, 1], /> <0 means the target variable is negatively correlated with the auxiliary variable, /> >0 means the target variable is positively correlated with the auxiliary variable; /> Indicates observation site/> A certain type of auxiliary variable value on; /> It is an observation site/> The target variable value of Indication and observation sites/> The number of adjacent observation sites that are adjacent in space (i.e., the first preset number, exemplarily, can be obtained from the observation sites/> Select one adjacent observation site from the top, bottom, left and right of the image, that is, the first preset number/> Can be 4); /> Indicates based on observation site/> and observation sites/> The spatial weight matrix weighted by the distance between them.

In one embodiment, after partitioning the target area each time, the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the current partition state is calculated based on the local autocorrelation coefficients of each bivariate, including:

After the target area is partitioned each time, based on the local spatial autocorrelation coefficients of the two variables corresponding to all the observation sites in each partition space under the current partition state, the local autocorrelation index variance values of the relationships between various auxiliary variables and the target variables in each partition space are calculated;

Based on the bivariate local spatial autocorrelation coefficients corresponding to all the observation sites in the target area, the global autocorrelation index variance values of the relationships between various auxiliary variables and the target variable in the target area are calculated;

The spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the target area under the current partition state is calculated based on the local autocorrelation index variance value and the global autocorrelation index variance value.

Specifically, as shown in FIG2 , the spatial hierarchical heterogeneity corresponding to each type of auxiliary variable in the current partition state can be calculated based on the bivariate local autocorrelation coefficient through the geographic detector. It can be seen from the above embodiment that each observation site corresponds to multiple types of auxiliary variables, and each type of auxiliary variable corresponds to a bivariate local spatial autocorrelation coefficient. Based on this, each time the target area is partitioned, for each partition space in the current partition state, the local autocorrelation index variance value of the relationship between each type of auxiliary variable and the target variable in the partition space is calculated according to the bivariate local spatial autocorrelation coefficient corresponding to all the observation sites in the partition space. Each time the target area is partitioned, based on the bivariate local spatial autocorrelation coefficient corresponding to all the observation sites in the target area, the global autocorrelation index variance value of the relationship between each type of auxiliary variable in the target area and the target variable is calculated, that is, each auxiliary variable corresponds to a global autocorrelation index variance value.

After obtaining the local autocorrelation index variance values and global autocorrelation index variance values of the relationship between each type of auxiliary variable and the target variable, the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the target area under the current partition state is calculated based on the local autocorrelation index variance values and the global autocorrelation index variance values. :

(2)

in, Indicates the first/> The spatial stratified heterogeneity corresponding to the auxiliary variables of the class, /> The value range of is [0, 1], Indicates that there is no obvious spatial pattern in the relationship between the target variable and the auxiliary variable, and they are randomly assigned to spatial locations. /> It means that the relationship between the target variable and the auxiliary variable in the same partition space is consistent, while the relationship between the target variable and the auxiliary variable in different partition spaces is different;/> Represents partition space; /> Indicates the number of partition spaces in the target area; /> Represents partition space/> The number of observation sites in; /> Indicates the number of observation sites in the target area; /> Represents partition space/> Middle/> The local autocorrelation index variance value corresponding to the auxiliary variable of the class; /> Indicates the target area. The global autocorrelation index variance value corresponding to the class auxiliary variable.

Step 120: Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, determine the target partition state, the target partition state corresponding to the multiple sub-regions of the target region.

Specifically, after each partitioning, the target partitioning state is determined based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable. That is, after it is determined that no further partitioning is required, the current partitioning state is determined as the final target partitioning state, and the target partitioning state corresponds to multiple sub-areas of the target area.

In one embodiment, the determining of the target partition state based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition includes:

Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, the average spatial hierarchical heterogeneity corresponding to the target area after each partition is calculated;

Constructing a first change curve based on the average spatial stratified heterogeneity corresponding to the target area after each partition, or constructing a second change curve based on the difference between the average spatial stratified heterogeneity after two adjacent partitions;

A partition state corresponding to an inflection point of the first change curve or an inflection point of the second change curve is determined as a target partition state.

Specifically, after each partition, the average spatial stratified heterogeneity corresponding to the target area after each partition is calculated based on the spatial stratified heterogeneity calculated for each type of auxiliary variable. For example, when there are 5 types of auxiliary variables corresponding to each observation station in the target area, based on the above embodiment, the spatial stratified heterogeneity of the 5 types of auxiliary variables can be calculated respectively. 、/> 、/> 、/> and/> , and then calculate and determine the average spatial stratified heterogeneity/> .

FIG3 is a schematic diagram of a first curve and a second curve constructed based on average spatial stratified heterogeneity provided by an embodiment of the present invention. As shown in FIG3, based on the average spatial stratified heterogeneity calculated after each partition, a first change curve can be constructed, and the first change curve represents the change of the average spatial stratified heterogeneity after each partition. Alternatively, based on the difference between the average spatial stratified heterogeneity after two adjacent partitions, , a second change curve can be constructed, which represents the difference in average spatial stratified heterogeneity corresponding to two adjacent partitions. The partition state corresponding to the inflection point of the first change curve or the inflection point of the second change curve is determined as the target partition state. The inflection point of the first change curve, that is, the average spatial stratified heterogeneity of the partition state after the partition is carried out tends to be stable, and the inflection point of the second change curve, that is, the difference in average spatial stratified heterogeneity after two adjacent partitions, tends to be stable.

In the above-mentioned embodiment, the method for generating geographic space full coverage data based on spatially variable parameter machine learning, the target partition state and the spatial position and spatial range of each partition space under the target partition state are based on the autocorrelation between the target variable and the auxiliary variable, and are recursively determined by the geographic detector model based on the quadtree principle, which can effectively simulate the spatial non-stationary relationship between variables.

Step 130: In the target partition state, a spatially varying parameter machine learning model is constructed for each sub-region in the target region, each of which is respectively obtained by training based on the target variables, auxiliary variables of each observation site within the corresponding spatial range, and the distance from each observation site to the corresponding spatially varying parameter machine learning model; each of the spatially varying parameter machine learning models includes location information and spatial range information, the location information is the center coordinates of the sub-region corresponding to the spatially varying parameter machine learning model, and the spatial range information is the size of the sub-region corresponding to the spatially varying parameter machine learning model.

In one embodiment, each of the spatially variable parameter machine learning models corresponding to each of the sub-regions is trained based on the following method:

Obtain target variables and auxiliary variables of all the observation sites in the sub-region;

Determine the distances from each of the observation sites in the sub-area to the center point of the sub-area;

A random forest model is trained based on the target variables and auxiliary variables of all the observation sites in the sub-region, and the distances from each observation site in the sub-region to the center point of the sub-region, and the trained random forest model is used as the spatially varying parameter machine learning model corresponding to the sub-region.

Specifically, in the target partition state, the target area includes multiple sub-areas. For each sub-area, a spatial edge parameter machine learning model with location information and spatial range is set at the center of the sub-area, and the spatial variable parameter machine learning model corresponding to each sub-area is trained separately. As shown in Figure 2, the specific training process of the spatial variable parameter machine learning model corresponding to each sub-area is as follows:

Calculate the coordinates of the center point of the sub-region, determine the size of the sub-region, determine the center coordinates of the sub-region as the position ( S ) of the corresponding spatially varying parameter machine learning model to be trained, and determine the size of the sub-region as the spatial range ( E ) of the corresponding spatially varying parameter machine learning model to be trained (i.e., random forest model RF ). FIG. 4 is a schematic diagram of each sub-region under the target partition state provided by an embodiment of the present invention. As shown in FIG. 4, each rectangular square in the figure represents the size of each sub-region, that is, the spatial range of the spatially varying parameter machine learning model to be trained; the large dot in each sub-region represents the center point of the sub-region, that is, the position of the spatially varying parameter machine learning model to be trained, and each small dot in FIG. 4 represents the location distribution of the observation site.

Obtain the target variables and auxiliary variables of all observation sites in the sub-region, determine the distance between each observation site in the sub-region and the center point of the sub-region (that is, the location of the spatially varying parameter machine learning model to be trained), use the data of the observation sites covered by the sub-region as samples, and bring the target variables, auxiliary variables of the observation sites, and the distance between the observation sites and the corresponding spatially varying parameter machine learning model to be trained into formula (3) to train the spatially varying parameter machine learning model:

(3)

In the above formula, Y represents the value of the target variable; and/> Respectively represent the location information and spatial range information of the space-varying parameter machine learning model to be trained; /> Represents auxiliary variables that affect Y; /> Represents the spatial extent of the spatially varying parameter machine learning model to be trained/> The location of each observation station under coverage to this spatially variable parameter machine learning model/> distance.

Add to As training data, it is intended to simulate the local spatial variation of the relationship between the target variable and the auxiliary variables, which means that two points with the same values of all auxiliary variables may have different target variable values if they are at different distances from the spatial machine learning model, and this possible difference is predicted by the spatially variable parameter machine learning model based on the observation site data.

In the specific training process, in order to ensure that the spatially varying parameter machine learning model to be trained has sufficient training samples, the spatial range of the spatially varying parameter machine learning model to be trained can be Zoom in to expand the number of observation sites added to its training. The specific zoom level can be determined according to the preset minimum amount of training data. For example, when the training model requires at least 100 observation sites within its scope, the spatial scope of the spatially variable parameter machine learning model to be trained is expanded. , so that its range includes at least 100 observation sites. This enlargement is reasonable because machine learning can build more complex models than linear models and can model local changes in relationships. In addition, in the spatial variation coefficient model, some overlap of training data is acceptable.

Compared with existing interpolation methods, spatially varying parameter machine learning models can utilize high-dimensional auxiliary variables to simulate the spatial non-stationary relationship between the target variable and its auxiliary variables while avoiding collinearity, thus having the following beneficial effects:

(1) In order to address the problem of spatial non-stationarity in the relationship between the target variable and its auxiliary variables due to the influence of complex terrain, human activities and other conditions, two main parameters, spatial location and range, were set for the spatially varying parameter machine learning model corresponding to each sub-region. In addition, the distance between the spatially varying parameter machine learning model and the observation site was added as a key variable during training to automatically simulate the local spatial variation of the relationship between the target variable and the auxiliary variables, thereby correctly utilizing the spatial autocorrelation law and improving the prediction accuracy of the model.

(2) In order to address the problem of factor collinearity in the multivariate linear regression model and poor interpretability of machine learning, the spatially variable parameter machine learning model in the above embodiment can effectively handle the situation where multidimensional auxiliary variables have high interactions, and can better simulate the nonlinear relationship between the target variable and its auxiliary variables.

Step 140: Based on the spatially variable parameter machine learning models corresponding to the sub-regions, interpolation prediction is performed on the target variables of the preset points to be predicted in the target region to obtain target interpolation results and uncertainty analysis results.

In one embodiment, the interpolation prediction of the target variables of each preset point to be predicted in the target area based on each space-varying parameter machine learning model corresponding to each sub-area is performed to obtain the target interpolation result and the uncertainty analysis result, including:

Using multiple spatially varying parameter machine learning models to respectively predict the target variable of the point to be predicted, and obtaining multiple interpolation prediction results, wherein the multiple spatially varying parameter machine learning models include a second preset number of spatially varying parameter machine learning models adjacent to the point to be predicted;

Based on the inverse distance weighted method, the target interpolation result and the uncertainty analysis result of the target variable of the point to be predicted are determined according to the multiple interpolation prediction results.

Specifically, after determining the target partition state, interpolation prediction is performed on each point to be predicted in the target area based on each spatially variable parameter machine learning model corresponding to each sub-area under the target partition state. The method for obtaining the target interpolation result is specifically as follows:

For each point to be predicted, multiple space-varying parameter machine learning models are used to perform prediction, thereby obtaining multiple interpolation prediction results. The multiple space-varying parameter machine learning models include a second preset number of space-varying parameter machine learning models adjacent to the point to be predicted. For example, the four space-varying parameter machine learning models closest to the point to be predicted can be selected.

Furthermore, based on the inverse distance weighted method, the target interpolation result of the point to be predicted is determined according to multiple interpolation prediction results. In the method for generating geographic spatial full coverage data based on spatially variable parameter machine learning, the target interpolation result of each point to be predicted is a linear combination of the interpolation prediction results of the spatially variable parameter machine learning model of the spatially adjacent points to be predicted, that is, the target interpolation result for:

(4)

Among them, in the above formula Indicates the point to be predicted/> The target interpolation result (that is, the target variable predicted by the model); Indicates the first/> A space-varying parameter machine learning model/> Treat prediction points/> The interpolation prediction result of Indicates the point to be predicted/> The number of spatially varying parameter machine learning models for prediction (i.e., the second preset number);/> Representing spatially varying parameter machine learning models/> The corresponding weight values, i.e., the estimated regression coefficients, of the spatially varying parameter machine learning models used for prediction are spatially different and are determined based on the inverse distance weighting method; Indicates the point to be predicted/> The location coordinates of .

Optionally, multiple kernel functions can be used to determine the weights corresponding to each spatially variable parameter machine learning model in formula (4). As shown in Figure 2, the embodiment of the present application uses five kernel functions for cross-testing to determine the kernel function with smaller error for interpolation prediction. The five kernel functions used in the embodiment of the present application include: nearest neighbor, equal weight (EW), inverse distance weighted (IDW), Gaussian weighted (GW) and adaptive Gaussian weighted (GAW), which are specifically introduced as follows:

(1) Nearest neighbor: The prediction value of the spatially variable parameter machine learning model that is closest to the point to be predicted is used as the final target interpolation result.

(2) EW: Use the nearest A spatially variable parameter machine learning model is used for prediction, and the average value of each interpolation prediction result is taken as the final target interpolation result.

(3) IDW: Use the nearest neighboring points to be predicted The spatially variable parameter machine learning model is used for prediction, and the inverse distance weighting is performed on each interpolation prediction result to obtain the final target interpolation result. The weight of each spatially variable parameter machine learning model is The formula is as follows:

(5)

Among them, in the above formula Represents an arbitrary positive real number, usually set to 2; /> Representing spatially varying parameter machine learning models/> To the predicted point/> Euclidean distance; /> Indicates the point to be predicted/> The number of spatially varying parameter machine learning models used to make predictions.

(4) GW: Use the nearest neighboring point to be predicted The prediction is made by using a spatially variable parameter machine learning model, and the interpolation prediction results are Gaussian weighted. The weight of each spatially variable parameter machine learning model/> The formula is as follows:

(6)

Among them, in the above formula Indicates the point to be predicted/> The location coordinates of Representing spatially varying parameter machine learning models/> To the predicted point/> The Euclidean distance is calculated as/> ,/> is a constant, /> Representing spatially varying parameter machine learning models/> The location coordinates of .

(5) GAW: Use the nearest A spatially variable parameter machine learning model is used for prediction, and Gaussian adaptive weighting is performed on the final target interpolation result. The formula is as follows:

(7)

Among them, in the above formula Indicates the point to be predicted/> The location coordinates of Representing spatially varying parameter machine learning models/> To the predicted point/> Euclidean distance of; /> Represents the optimization parameters.

After cross-testing, the kernel function with the smallest error is the inverse distance weighted (IDW). Therefore, the weight value corresponding to each spatial variable parameter machine learning model is determined based on the inverse distance weighted method.

It is understandable that the spatially varying parameter machine learning model will produce certain errors when making predictions. Therefore, the uncertainty analysis result of the final target interpolation result can be measured according to the uncertainty of each spatially varying parameter machine learning model. That is, the error variance of the final target interpolation result (that is, the uncertainty analysis result) can be converted into a linear combination of the error variances of each spatially varying parameter machine learning model. The formula is as follows:

(8)

Wherein, V in the above formula represents the variance statistic of the interpolation prediction result of the spatially varying parameter machine learning model for the predicted point; Indicates the interpolation prediction result; /> Represents a true value; /> Indicates the number of spatially variable parameter machine learning models used to predict the points to be predicted; /> Indicates the first/> Interpolation prediction results of a spatially variable parameter machine learning model; /> Indicates the first/> The weights of a spatially varying parameter machine learning model.

Since each spatially varying parameter machine learning model is trained and established independently, the prediction errors of different spatially varying parameter machine learning models can be considered independent. Based on this, equation (8) can be further derived as equation (9):

(9)

in, It is the first/> The error variance of the interpolated prediction results of a spatially varying parameter machine learning model.

Furthermore, the confidence interval of the spatial interpolation prediction can be determined based on the prediction results of the spatially varying parameter machine learning model for each point to be predicted in the target area.

By estimating the uncertainty of the prediction results, giving the error variance of the prediction results, and giving the confidence interval of the spatial interpolation prediction through the prediction results, the interpretability of the spatially varying parameter machine learning model is further improved.

The spatial interpolation method of spatially variable parameter machine learning provided by the present invention is aimed at the situation that there are attribute similarity and spatial autocorrelation between the target variable and various auxiliary variables at the same time, and the relationship heterogeneity between the target variable and various auxiliary variables is strong. By partitioning the target area based on spatial hierarchical heterogeneity, multiple sub-areas are obtained, and the target area prediction points are predicted based on the spatially variable parameter machine learning model corresponding to each sub-area to obtain the target interpolation result. Since each spatially variable parameter machine learning model constructed for each sub-area has a specific position and range, the spatial non-stationary relationship between the target variable and the auxiliary variable can be effectively simulated. At the same time, the spatially variable parameter machine learning model is established based on the target variable, the auxiliary variable and the distance of the observation site from the spatially variable parameter machine learning model position of the observation site, so the local spatial changes in the relationship between the target variable and the auxiliary variable can be automatically modeled, and then the spatial correlation and heterogeneity of the multi-dimensional mixed type auxiliary variables are fully utilized. In the large-scale high spatial resolution interpolation scenario, the target variables of each point to be predicted can be accurately interpolated according to the limited observation site data, and then the accurate spatial distribution map corresponding to the full coverage data of the geographic space is obtained.

The following is a description of the device for generating geographic space full coverage data based on spatially varying parameter machine learning provided by the present invention. The device for generating geographic space full coverage data based on spatially varying parameter machine learning described below and the method for generating geographic space full coverage data based on spatially varying parameter machine learning described above can be referenced to each other.

FIG5 is a schematic diagram of the structure of a device for generating geographic space full coverage data based on space-varying parameter machine learning according to an embodiment of the present invention. As shown in FIG5 , the device 500 for generating geographic space full coverage data based on space-varying parameter machine learning includes: a partitioning module 510, a determination module 520, a modeling module 530, and a prediction module 540;

Partitioning module 510, for partitioning the target area step by step, and after each partitioning, based on various auxiliary variables and target variables in each observation station in the target area, calculating the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and target variables in the current partitioning state;

A determination module 520, configured to determine a target partition state based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, wherein the target partition state corresponds to a plurality of sub-regions of the target region;

Modeling module 530, for constructing a spatially varying parameter machine learning model for each of the sub-regions in the target region in the target partition state, each of the spatially varying parameter machine learning models is obtained by training based on the target variables and auxiliary variables of each observation site within the corresponding spatial range, and the distance from each observation site to the corresponding spatially varying parameter machine learning model; each of the spatially varying parameter machine learning models includes position information and spatial range information, the position information is the center coordinates of the sub-region corresponding to the spatially varying parameter machine learning model, and the spatial range information is the size of the sub-region corresponding to the spatially varying parameter machine learning model;

The prediction module 540 is used to perform interpolation prediction on the target variables of each preset point to be predicted in the target area based on the spatially variable parameter machine learning models corresponding to each sub-area, so as to obtain the target interpolation result and the uncertainty analysis result.

The spatial interpolation device of spatially variable parameter machine learning provided by the present invention, for the case where there are attribute similarity and spatial autocorrelation between the target variable and various auxiliary variables, and the heterogeneity of the relationship between the target variable and various auxiliary variables is strong, by partitioning the target area based on spatial hierarchical heterogeneity, a plurality of sub-areas are obtained, and the target interpolation result is obtained by predicting the target area to be predicted points based on the spatially variable parameter machine learning model corresponding to each sub-area. Since each spatially variable parameter machine learning model constructed for each sub-area has a specific position and range, the spatial non-stationary relationship between the target variable and the auxiliary variable can be effectively simulated. At the same time, the spatially variable parameter machine learning model is established based on the target variable, the auxiliary variable and the distance of the observation site from the spatially variable parameter machine learning model position of the observation site, so the local spatial changes in the relationship between the target variable and the auxiliary variable can be automatically modeled, and then the spatial correlation and heterogeneity of the multi-dimensional mixed type auxiliary variables are fully utilized. In the large-scale high spatial resolution interpolation scenario, the target variables of each point to be predicted can be accurately interpolated according to the limited observation site data, and then the accurate spatial distribution map corresponding to the full coverage data of the geographic space is obtained.

In one embodiment, the partition module 510 is specifically used for:

For each observation site in the target area, calculate the bivariate local spatial autocorrelation coefficient between the target variable in the observation site and each type of auxiliary variable;

After the target area is partitioned each time, the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the current partition state is calculated based on the local autocorrelation coefficient of each bivariate.

In one embodiment, the partition module 510 is specifically used for:

For each of the observation sites in the target area, based on the auxiliary variables of the observation site and the target variables of a first preset number of neighboring observation sites of the observation site, the bivariate local spatial autocorrelation coefficient corresponding to the observation site is calculated.

In one embodiment, the partition module 510 is specifically used for:

After the target area is partitioned each time, based on the local spatial autocorrelation coefficients of the two variables corresponding to all the observation sites in each partition space under the current partition state, the local autocorrelation index variance values corresponding to each type of auxiliary variable in each partition space are calculated;

Based on the bivariate local spatial autocorrelation coefficients corresponding to all the observation sites in the target area, the global autocorrelation index variance values of the relationships between various auxiliary variables and the target variable in the target area are calculated;

The spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the target area under the current partition state is calculated based on the local autocorrelation index variance value and the global autocorrelation index variance value.

In one embodiment, the determining module 520 is specifically configured to:

Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, the average spatial hierarchical heterogeneity corresponding to the target area after each partition is calculated;

Constructing a first change curve based on the average spatial stratified heterogeneity corresponding to the target area after each partition, or constructing a second change curve based on the difference between the average spatial stratified heterogeneity after two adjacent partitions;

A partition state corresponding to an inflection point of the first change curve or an inflection point of the second change curve is determined as a target partition state.

In one embodiment, the space-varying parameter machine learning models corresponding to the target partition spaces are respectively trained based on the following methods:

Obtain target variables and auxiliary variables of all the observation sites in the sub-region;

Determine the distances from each of the observation sites in the sub-area to the center point of the sub-area;

A random forest model is trained based on the target variables and auxiliary variables of all the observation sites in the sub-region, and the distances from each observation site in the sub-region to the center point of the sub-region, and the trained random forest model is used as the spatially varying parameter machine learning model corresponding to the sub-region.

In one embodiment, the prediction module 540 is specifically used to:

Using multiple spatially varying parameter machine learning models to respectively predict the target variable of the point to be predicted, and obtaining multiple interpolation prediction results, wherein the multiple spatially varying parameter machine learning models include a second preset number of spatially varying parameter machine learning models adjacent to the point to be predicted;

Based on the inverse distance weighted method, the target interpolation result and the uncertainty analysis result of the target variable of the point to be predicted are determined according to the multiple interpolation prediction results.

FIG6 illustrates a schematic diagram of the physical structure of an electronic device. As shown in FIG6 , the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute a method for generating geographic space full coverage data based on space-varying parameter machine learning, the method comprising:

The target area is partitioned step by step, and after each partition, based on various auxiliary variables and target variables in each observation station in the target area, the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and the target variables in the current partition state is calculated;

Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, determine the target partition state, the target partition state corresponds to multiple sub-regions of the target region;

In the target partition state, a spatially varying parameter machine learning model is constructed for each sub-region in the target region, each of which is obtained by training based on the target variables and auxiliary variables of each observation site within the corresponding spatial range, and the distance from each observation site to the corresponding spatially varying parameter machine learning model; each of the spatially varying parameter machine learning models includes position information and spatial range information, the position information is the center coordinates of the sub-region corresponding to the spatially varying parameter machine learning model, and the spatial range information is the size of the sub-region corresponding to the spatially varying parameter machine learning model;

Based on the spatially variable parameter machine learning models corresponding to the sub-regions, interpolation prediction is performed on the target variables of the preset points to be predicted in the target region to obtain target interpolation results and uncertainty analysis results.

In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

On the other hand, the present invention also provides a computer program product, the computer program product includes a computer program, the computer program can be stored on a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the method for generating geographic space full coverage data based on space-varying parameter machine learning provided by the above methods, the method includes:

The target area is partitioned step by step, and after each partition, based on various auxiliary variables and target variables in each observation station in the target area, the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and the target variables in the current partition state is calculated;

Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, determine the target partition state, the target partition state corresponds to multiple sub-regions of the target region;

In the target partition state, a spatially varying parameter machine learning model is constructed for each sub-region in the target region, each of which is obtained by training based on the target variables and auxiliary variables of each observation site within the corresponding spatial range, and the distance from each observation site to the corresponding spatially varying parameter machine learning model; each of the spatially varying parameter machine learning models includes position information and spatial range information, the position information is the center coordinates of the sub-region corresponding to the spatially varying parameter machine learning model, and the spatial range information is the size of the sub-region corresponding to the spatially varying parameter machine learning model;

Based on the spatially variable parameter machine learning models corresponding to the sub-regions, interpolation prediction is performed on the target variables of the preset points to be predicted in the target region to obtain target interpolation results and uncertainty analysis results.

In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the method for generating geographic space full coverage data based on space-varying parameter machine learning provided by the above methods, the method comprising:

The target area is partitioned step by step, and after each partition, based on various auxiliary variables and target variables in each observation station in the target area, the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and the target variables in the current partition state is calculated;

Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, determine the target partition state, the target partition state corresponds to multiple sub-regions of the target region;

In the target partition state, a spatially varying parameter machine learning model is constructed for each sub-region in the target region, each of which is obtained by training based on the target variables and auxiliary variables of each observation site within the corresponding spatial range, and the distance from each observation site to the corresponding spatially varying parameter machine learning model; each of the spatially varying parameter machine learning models includes position information and spatial range information, the position information is the center coordinates of the sub-region corresponding to the spatially varying parameter machine learning model, and the spatial range information is the size of the sub-region corresponding to the spatially varying parameter machine learning model;

Based on the spatially variable parameter machine learning models corresponding to the sub-regions, interpolation prediction is performed on the target variables of the preset points to be predicted in the target region to obtain target interpolation results and uncertainty analysis results.

The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM/RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A method for generating geographic space full coverage data based on spatially variable parameter machine learning, characterized by comprising: The target area is partitioned step by step, and after each partition, based on various auxiliary variables and target variables in each observation station in the target area, the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and the target variables in the current partition state is calculated; Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, determine the target partition state, the target partition state corresponds to multiple sub-regions of the target region; In the target partition state, a spatially varying parameter machine learning model is constructed for each sub-region in the target region, each of which is obtained by training based on the target variables and auxiliary variables of each observation site within the corresponding spatial range, and the distance from each observation site to the corresponding spatially varying parameter machine learning model; each of the spatially varying parameter machine learning models includes position information and spatial range information, the position information is the center coordinates of the sub-region corresponding to the spatially varying parameter machine learning model, and the spatial range information is the size of the sub-region corresponding to the spatially varying parameter machine learning model; Based on the spatially variable parameter machine learning models corresponding to the sub-regions, interpolation prediction is performed on the target variables of the preset points to be predicted in the target region to obtain target interpolation results and uncertainty analysis results. 2. The method for generating geographic space full coverage data based on spatially variable parameter machine learning according to claim 1 is characterized in that after each partitioning, based on various auxiliary variables and target variables in each observation station in the target area, the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and the target variable in the current partition state is calculated, including: For each of the observation sites in the target area, calculating the bivariate local spatial autocorrelation coefficient between the target variable in the observation site and each type of the auxiliary variables; After the target area is partitioned each time, the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the current partition state is calculated based on the local autocorrelation coefficient of each bivariate. 3. The method for generating geographic space full coverage data based on space-varying parameter machine learning according to claim 2 is characterized in that, for each observation site in the target area, the bivariate local spatial autocorrelation coefficient between the target variable in the observation site and each type of auxiliary variable is calculated, including: For each of the observation sites in the target area, based on the auxiliary variables of the observation site and the target variables of a first preset number of neighboring observation sites of the observation site, the bivariate local spatial autocorrelation coefficient corresponding to the observation site is calculated. 4. The method for generating geographic space full coverage data based on spatially varying parameter machine learning according to claim 2 is characterized in that after partitioning the target area each time, the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the current partition state is calculated based on the local autocorrelation coefficient of each bivariate, including: After the target area is partitioned each time, based on the local spatial autocorrelation coefficients of the two variables corresponding to all the observation sites in each partition space under the current partition state, the local autocorrelation index variance values of the relationships between various auxiliary variables and the target variables in each partition space are calculated; Based on the bivariate local spatial autocorrelation coefficients corresponding to all the observation sites in the target area, the global autocorrelation index variance values of the relationships between various auxiliary variables and the target variable in the target area are calculated; The spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable in the target area under the current partition state is calculated based on the local autocorrelation index variance value and the global autocorrelation index variance value. 5. The method for generating geographic space full coverage data based on spatially varying parameter machine learning according to any one of claims 1 to 4, characterized in that the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition is used to determine the target partition state, comprising: Based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, the average spatial hierarchical heterogeneity corresponding to the target area after each partition is calculated; Constructing a first change curve based on the average spatial stratified heterogeneity corresponding to the target area after each partition, or constructing a second change curve based on the difference between the average spatial stratified heterogeneity after two adjacent partitions; A partition state corresponding to an inflection point of the first change curve or an inflection point of the second change curve is determined as a target partition state. 6. The method for generating geographic space full coverage data based on spatially varying parameter machine learning according to any one of claims 1 to 4, characterized in that each of the spatially varying parameter machine learning models corresponding to each of the sub-regions is trained based on the following method: Obtain target variables and auxiliary variables of all the observation sites in the sub-region; Determine the distances from each of the observation sites in the sub-area to the center point of the sub-area; A random forest model is trained based on the target variables and auxiliary variables of all the observation sites in the sub-region, and the distances from each observation site in the sub-region to the center point of the sub-region, and the trained random forest model is used as the spatially varying parameter machine learning model corresponding to the sub-region. 7. The method for generating geographic space full coverage data based on space-varying parameter machine learning according to any one of claims 1 to 4, characterized in that the target variables of each preset point to be predicted in the target area are interpolated and predicted based on each space-varying parameter machine learning model corresponding to each sub-area, respectively, to obtain a target interpolation result and an uncertainty analysis result, including: Using multiple spatially varying parameter machine learning models to respectively predict the target variable of the point to be predicted, and obtaining multiple interpolation prediction results, wherein the multiple spatially varying parameter machine learning models include a second preset number of spatially varying parameter machine learning models adjacent to the point to be predicted; Based on the inverse distance weighted method, the target interpolation result and the uncertainty analysis result of the target variable of the point to be predicted are determined according to the multiple interpolation prediction results. 8. A device for generating geographic space full coverage data based on space-varying parameter machine learning, characterized by comprising: A partitioning module is used to partition the target area step by step, and after each partition, based on various auxiliary variables and target variables in each observation station in the target area, calculate the spatial hierarchical heterogeneity of the relationship between various auxiliary variables and target variables in the current partition state; A determination module, configured to determine a target partition state based on the spatial hierarchical heterogeneity of the relationship between each type of auxiliary variable and the target variable after each partition, wherein the target partition state corresponds to a plurality of sub-regions of the target region; A modeling module is used to construct a spatially varying parameter machine learning model for each sub-region in the target region under the target partition state, each of the spatially varying parameter machine learning models is respectively obtained by training based on the target variables and auxiliary variables of each observation site within the corresponding spatial range, and the distance from each observation site to the corresponding spatially varying parameter machine learning model; each of the spatially varying parameter machine learning models includes position information and spatial range information, the position information is the center coordinates of the sub-region corresponding to the spatially varying parameter machine learning model, and the spatial range information is the size of the sub-region corresponding to the spatially varying parameter machine learning model; The prediction module is used to perform interpolation prediction on the target variables of each preset point to be predicted in the target area based on the spatially variable parameter machine learning models corresponding to each sub-area, so as to obtain the target interpolation result and uncertainty analysis result. 9. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for generating geographic spatial full coverage data based on spatially variable parameter machine learning as described in any one of claims 1 to 7 is implemented. 10. A non-transitory computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for generating geographic spatial full coverage data based on spatially variable parameter machine learning as described in any one of claims 1 to 7 is implemented.