Geographic element prediction method and device based on non-parametric weight function model, and medium

Abstract

The invention relates to the technical field of geographic information, and discloses a geographic element prediction method and device based on a non-parametric weight function model and a medium, and the method comprises the steps: determining a characteristic quantity type, a geographic element prediction model and a space-time neighborhood control parameter which are needed by target geographic element prediction; a non-parametric weight function model is determined according to the space-time neighborhood control parameters, the input of the model is the distance from a sampling point or a discrete grid point to a group of control points in the space-time neighborhood, and the output of the model is the weight of the position of the control points in the space-time neighborhood; feature quantities required by model training and geographic elements of sampling points are obtained, and a geographic element prediction model is combined to train the model; obtaining feature quantities required by target geographic element prediction and geographic elements of the sampling points, and interpolating part or all of the feature quantities into the required discrete grid; and a geographic element prediction model and a non-parametric weight function model are adopted to predict the target geographic element of any discrete grid point, so that the automation level and the prediction precision are improved.

Description

Technical Field

[0001] The present invention relates to the field of geographic information technology, and in particular to a geographic element prediction method, device and medium based on a non-parametric weight function model. Background Art

[0002] According to the basic principles of geography, all things are connected, and similar things are more closely related. Therefore, the classic regression model is not particularly suitable for modeling the relationship between geographical elements that change with time and space. Therefore, the spatiotemporal local modeling method used to model the relationship between geographical elements that change with time and space came into being. The spatiotemporal local modeling method has been widely used in the fields of soil science, meteorology, geography, environmental science, forestry, etc. The weight function model is the core of the spatiotemporal local modeling method, but the existing spatiotemporal local modeling method usually uses a parameterized function model as the weight function model. Although the parameterized weight function model is easy to fit and calculate, the parameterized weight function model has limited expression ability and usually requires manual selection, which affects both the automation level of the spatiotemporal local modeling method and the prediction accuracy of the spatiotemporal local modeling method.

[0003] Therefore, how to provide a technical solution for geographic element prediction based on a non-parametric weight function model is a technical problem that needs to be urgently solved by those skilled in the art. Summary of the invention

[0004] The purpose of the present invention is to provide a geographical element prediction method, device and medium based on a non-parametric weight function model, which has a higher automation level and prediction accuracy.

[0005] In order to solve the above technical problems, the present invention provides a geographical element prediction method based on a non-parametric weight function model, the method comprising:

[0006] Determine the types of characteristic quantities, geographic element prediction models and spatiotemporal neighborhood control parameters required for target geographic element prediction;

[0007] Determine the non-parametric weight function model required for the prediction of the target geographic element according to the spatiotemporal neighborhood control parameters; the input of the non-parametric weight function model is the distance from the sampling point or discrete grid point to a group of control points in its spatiotemporal neighborhood, and the output is the weight of the control point position in the spatiotemporal neighborhood;

[0008] Acquire the feature quantities and geographical elements of the sampling points required for training the non-parametric weight function model, construct a first sample set, and train the non-parametric weight function model using the first sample set and the geographical element prediction model;

[0009] Acquire the feature quantities required for prediction of the target geographic elements and the geographic elements of the sampling points, construct a second sample set, and interpolate part or all of the feature quantities to the required discrete grids;

[0010] The first sample set or / and the second sample set, the geographic element prediction model and the trained non-parametric weight function model are used to predict the target geographic element of any discrete grid point.

[0011] In a first aspect, in the above-mentioned geographic element prediction method based on the non-parametric weight function model provided by the present invention, it also includes:

[0012] When there are multiple continuous target geographic elements that need to be predicted, the error weight of each continuous target geographic element is determined to simultaneously model and predict the multiple continuous target geographic elements;

[0013] When there are multiple categorized target geographic elements that need to be predicted, the multiple categorized target geographic elements are combined into new categorized target geographic elements to simultaneously model and predict the multiple categorized target geographic elements.

[0014] On the other hand, in the above-mentioned geographic element prediction method based on the non-parametric weight function model provided by the present invention, it also includes:

[0015] When the control point is a sampling point in the spatiotemporal neighborhood, the output of the non-parametric weight function model is the weight of each sample in the spatiotemporal neighborhood;

[0016] When the control point is a set of evenly or unevenly distributed points artificially set in the spatiotemporal neighborhood, the weight of the control point position is interpolated to the sampling point position in the spatiotemporal neighborhood by interpolation to obtain the weight of the sample corresponding to the sampling point position.

[0017] On the other hand, in the geographic element prediction method based on the non-parametric weight function model provided by the present invention, the non-parametric weight function model is trained using the first sample set and the geographic element prediction model, including:

[0018] For any sampling point in the first sample set, determine, according to the spatiotemporal neighborhood control parameter, a sample in the spatiotemporal neighborhood and a control point in the spatiotemporal neighborhood corresponding to the sampling point;

[0019] Using the samples and control points in the determined spatiotemporal neighborhood, the geographic element prediction model and the non-parametric weight function model to be trained, the relationship between the target geographic element and the characteristic quantity is constructed;

[0020] By using the relationship between the constructed target geographic elements and the characteristic quantities, as well as the characteristic quantities of the samples corresponding to the sampling points, the predicted values ​​of the target geographic elements corresponding to the sampling points are obtained.

[0021] On the other hand, in the above-mentioned geographic element prediction method based on the non-parametric weight function model provided by the present invention, the relationship between the target geographic element and the characteristic quantity is constructed by using the samples in the determined spatiotemporal neighborhood and the control points in the spatiotemporal neighborhood, the geographic element prediction model and the non-parametric weight function model to be trained, including:

[0022] Calculate the distance from the sampling point to each control point in the determined spatiotemporal neighborhood;

[0023] Calculating the weight of each control point in the determined spatiotemporal neighborhood using the non-parametric weight function model to be trained;

[0024] According to the calculated weight of each control point, the weight of each sample in the determined spatiotemporal neighborhood is obtained;

[0025] The determined samples and corresponding weights in the spatiotemporal neighborhood and the geographic element prediction model are used to construct a relationship between the target geographic element and the characteristic quantity.

[0026] On the other hand, in the geographic element prediction method based on the non-parametric weight function model provided by the present invention, the first sample set or / and the second sample set, the geographic element prediction model and the trained non-parametric weight function model are used to predict the target geographic element of any discrete grid point, including:

[0027] For any discrete grid point, according to the spatiotemporal neighborhood control parameter, the samples and control points in the spatiotemporal neighborhood corresponding to the discrete grid point are determined;

[0028] Using the samples and control points in the determined spatiotemporal neighborhood, the geographic element prediction model and the trained non-parametric weight function model, the relationship between the target geographic element and the characteristic quantity is constructed;

[0029] By using the relationship between the constructed target geographic elements and the characteristic quantities, as well as the characteristic quantities corresponding to the discrete grid points, the predicted values ​​of the target geographic elements corresponding to the discrete grid points are obtained.

[0030] On the other hand, in the above-mentioned geographic element prediction method based on the non-parametric weight function model provided by the present invention, the relationship between the target geographic element and the characteristic quantity is constructed by using the samples in the determined spatiotemporal neighborhood and the control points in the spatiotemporal neighborhood, the geographic element prediction model and the trained non-parametric weight function model, including:

[0031] Calculate the distance from the discrete grid point to each control point in the determined space-time neighborhood;

[0032] The trained non-parametric weight function model is used to calculate the weight of each control point in the determined spatiotemporal neighborhood;

[0033] According to the calculated weight of each control point, the weight of each sample in the determined spatiotemporal neighborhood is obtained;

[0034] The relationship between the target geographic element and the characteristic quantity is constructed by using the determined samples and corresponding weights within the spatiotemporal neighborhood and the geographic element prediction model.

[0035] On the other hand, in the above-mentioned geographic element prediction method based on the non-parametric weight function model provided by the present invention, it also includes:

[0036] When the target geographic element is a continuous geographic element, a normalized or standardized value of the geographic element is generated, and the generated normalized or standardized value of the geographic element is denormalized or destandardized to obtain a final predicted value;

[0037] When the target geographic element is a categorized geographic element, probability values ​​for each category of the geographic element are generated, and based on the generated probability values ​​for each category of the geographic element, the category to which the geographic element belongs is obtained as the final prediction value.

[0038] In order to solve the above technical problems, the present invention also provides a geographical element prediction device based on a non-parametric weight function model, the device comprising:

[0039] Memory for storing computer programs;

[0040] A processor is used to implement the steps of the above-mentioned geographic element prediction method based on the non-parametric weight function model when executing the computer program.

[0041] In order to solve the above technical problems, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned geographic element prediction method based on the non-parametric weight function model are implemented.

[0042] It can be seen from the above technical scheme that the present invention provides a method for predicting geographic elements based on a non-parametric weight function model, the method comprising: determining the type of feature quantity, geographic element prediction model and spatiotemporal neighborhood control parameters required for the prediction of the target geographic element; determining the non-parametric weight function model required for the prediction of the target geographic element according to the spatiotemporal neighborhood control parameters; the input of the non-parametric weight function model is the distance from the sampling point or discrete grid point to a group of control points in its spatiotemporal neighborhood, and the output is the weight of the control point position in the spatiotemporal neighborhood; obtaining the feature quantity and the geographic elements of the sampling point required for the training of the non-parametric weight function model, constructing a first sample set, and using the first sample set and the geographic element prediction model to train the non-parametric weight function model; obtaining the feature quantity and the geographic elements of the sampling point required for the prediction of the target geographic element, constructing a second sample set, and interpolating part or all of the feature quantities in the feature quantity to the required discrete grid; using the first sample set and / or the second sample set, the geographic element prediction model and the trained non-parametric weight function model to predict the target geographic element of any discrete grid.

[0043] The beneficial effect of the present invention is that the above-mentioned geographic element prediction method based on the non-parametric weight function model provided by the present invention first determines the type of feature quantity, geographic element prediction model and spatiotemporal neighborhood control parameters required for the prediction of the target geographic element; then determines the non-parametric weight function model required for the prediction of the target geographic element according to the spatiotemporal neighborhood control parameters; the input of the model is the distance from the sampling point or discrete grid point to a group of control points in its spatiotemporal neighborhood, and the output is the weight of the control point position in the spatiotemporal neighborhood; then obtains the feature quantity required for the training of the non-parametric weight function model and the geographic elements of the sampling point, constructs a first sample set, and uses the first sample set and the geographic element prediction model to train the non-parametric weight function model; in addition, obtains the feature quantity required for the prediction of the target geographic element and the geographic elements of the sampling point, constructs a second sample set, and interpolates part or all of the feature quantities in the feature quantity to the required discrete grid; finally, uses the first sample set or / and the second sample set, the geographic element prediction model and the trained non-parametric weight function model to predict the target geographic element of any discrete grid. The non-parametric weight function model used in the present invention has a stronger expression ability, requires less human intervention, and enables the entire prediction method to have a higher level of automation and prediction accuracy.

[0044] In addition, the present invention also provides a corresponding geographic element prediction device based on a non-parametric weight function model and a computer-readable storage medium for the geographic element prediction method based on a non-parametric weight function model, which has the same or corresponding technical features as the above-mentioned geographic element prediction method based on a non-parametric weight function model, and the effect is the same as above. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. 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 work.

[0046] Figure 1 A flowchart of a geographic element prediction method based on a non-parametric weight function model provided by an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of the structure of a geographic element prediction device based on a non-parametric weight function model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] Spatiotemporal local modeling methods have been widely used in fields such as soil science (such as soil organic carbon prediction), meteorology (such as modeling the relationship between surface temperature and environmental factors, precipitation interpolation, temperature interpolation, radar-rain gauge combined precipitation estimation, multi-source data fusion quantitative precipitation estimation, and interpretation of numerical model forecast products), geography (such as research on factors affecting urban housing prices), environmental science (such as ozone concentration estimation, PM2.5 estimation), and forestry (such as forest carbon storage estimation). The weight function (also known as weight function, kernel function) model is the core of the spatiotemporal local modeling method. However, existing spatiotemporal local modeling methods usually use a few parameterized function models such as distance threshold model, Gaussian model, exponential model, and bisquare model as weight function models. Parameterized weight function models are easy to fit and calculate and have been widely used, but parameterized weight function models have limited expression capabilities and usually require manual selection, which affects both the automation level and the prediction accuracy of the spatiotemporal local modeling method.

[0049] In order to solve the above technical problems, the present invention provides a geographic element prediction method based on a non-parametric weight function model.

[0050] 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.

[0051] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 A flowchart of a geographic element prediction method based on a non-parametric weight function model provided by an embodiment of the present invention, such as Figure 1 As shown, the method includes:

[0052] S101. Determine the type of feature quantity, geographic element prediction model and spatiotemporal neighborhood control parameters required for target geographic element prediction.

[0053] It should be noted that when the target geographical element is different, the required types of feature quantities, geographical element prediction models and spatiotemporal neighborhood control parameters may be different. Therefore, when the present invention executes step S101, it is necessary to first determine the types of feature quantities, geographical element prediction models and spatiotemporal neighborhood control parameters required for the prediction of the target geographical element. For example, when the target geographical element is PM2.5, the types of feature quantities required for the prediction of the target geographical element may include aerosol optical depth (AOD), altitude, temperature, precipitation, wind speed and other feature quantities.

[0054] In practical applications, the above-mentioned geographic element prediction models can be weighted linear regression, weighted nonlinear regression, weighted partial least squares regression, weighted conditional probability density function, weighted logistic regression, weighted lasso regression, weighted elastic network regression, weighted Poisson regression, weighted negative binomial regression, weighted beta regression, weighted neural network, weighted random forest, weighted XGBoost, weighted LightGBM, weighted support vector machine, multi-scale weighted linear regression, multi-scale weighted Poisson regression and other weighted or multi-scale weighted models.

[0055] When the geographic element prediction model required for the prediction of the target geographic element is a multi-scale weighted model, a corresponding non-parametric weight function model can be set for each characteristic quantity.

[0056] In implementation, the above-mentioned spatiotemporal neighborhood may be a spatiotemporal neighborhood of any dimension greater than or equal to 1. The above-mentioned spatiotemporal neighborhood may be a pure spatial neighborhood or a pure temporal neighborhood, etc.

[0057] The above-mentioned spatiotemporal neighborhood control parameters can be used to control the shape, size and position of control points within the spatiotemporal neighborhood.

[0058] In a specific implementation, the above-mentioned spatiotemporal neighborhood control parameters may specifically include multiple parameters, which control the shape, size and position of the control points in the spatiotemporal neighborhood. For example, the spatiotemporal neighborhood control parameters may be the maximum number of samples S of the spatiotemporal neighborhood, the edge length of the cubic spatiotemporal neighborhood, the radius of the spherical spatiotemporal neighborhood, the spacing of the gridded control points in the neighborhood, and the position coordinates of any control point relative to the spatiotemporal neighborhood.

[0059] It should be pointed out that the samples of the present invention all have corresponding unified reference time and space coordinates (ie, the time and space coordinates of the sampling points).

[0060] S102. Determine the non-parametric weight function model required for prediction of the target geographic element according to the spatiotemporal neighborhood control parameters; the input of the non-parametric weight function model is the distance from the sampling point or discrete grid point to a set of control points in its spatiotemporal neighborhood, and the output is the weight of the control point position in the spatiotemporal neighborhood.

[0061] It should be noted that the above non-parametric weight function model can be an image-to-image transformation type machine learning model or a sequence-to-sequence transformation type machine learning model. For example, when the spatiotemporal neighborhood is a square (or cube, or its high-dimensional analogue hypercube) spatiotemporal neighborhood and the control points in the spatiotemporal neighborhood are a set of uniformly distributed grid control points, the non-parametric weight function model can use a convolutional neural network; when the spatiotemporal neighborhood is a variable spatiotemporal neighborhood controlled by the maximum number of samples S of the spatiotemporal neighborhood, the non-parametric weight function model can use a sequence-to-sequence transformation type machine learning model; when the control points of the spatiotemporal neighborhood are irregularly distributed points set artificially, the non-parametric weight function model can use a sequence-to-sequence transformation type machine learning model.

[0062] The input of the above non-parametric weight function model is the distance from the sampling point or discrete grid point to a set of control points in its spatiotemporal neighborhood. Optionally, the distance may include any one or any combination of the following:

[0063] Distance based on space, distance based on time, and distance based on geographical attributes.

[0064] In implementation, the calculated distances between sampling points or discrete grid points and control points may specifically include: distances based on space, such as Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, and distance in a certain coordinate direction; may also include: distances based on time, such as periodic time distances and non-periodic time distances between sampling points and control points; may also include: distances based on geographic attributes, such as the distances between sampling points or discrete grid points and control points calculated based on geographic attributes such as wind direction, etc.

[0065] For example, when the spatiotemporal neighborhood is a simple square spatial neighborhood, the control points within the square spatial neighborhood are a set of uniformly distributed grid control points, and the non-parametric weight function model adopts a convolutional neural network, the Euclidean distance from the sampling point or discrete grid point to a set of uniformly distributed grid control points in the square spatial neighborhood can be calculated, and then the matrix consisting of the Euclidean distances of all control points can be used as the input of the non-parametric weight function model.

[0066] For another example, when the spatiotemporal neighborhood is a variable spatiotemporal neighborhood controlled by the maximum number of samples S in the spatiotemporal neighborhood and the non-parametric weight function model adopts a sequence-to-sequence transformation type machine learning model, the Euclidean distance from the sampling point or discrete grid point to a set of control points in the variable spatiotemporal neighborhood can be calculated, and then the sequence consisting of the Euclidean distances of all control points can be used as the input of the non-parametric weight function model.

[0067] For another example, when the spatiotemporal neighborhood is a variable spatiotemporal neighborhood controlled by the maximum number of samples S in the spatiotemporal neighborhood and the non-parametric weight function model adopts a sequence-to-sequence transformation type machine learning model, the distance in the x-coordinate direction and the distance in the y-coordinate direction from the sampling point or discrete grid point to a set of control points in the variable spatiotemporal neighborhood can be calculated simultaneously, and the distance in the x-coordinate direction and the distance in the y-coordinate direction from the sampling point or discrete grid point to each control point can be formed into a two-dimensional distance vector, and then the two-dimensional distance vectors of all control points or the sequence composed of the transformed two-dimensional distance vectors can be used as the input of the non-parametric weight function model.

[0068] S103, obtaining the feature quantities and geographical elements of the sampling points required for training the non-parametric weight function model, constructing a first sample set, and using the first sample set and the geographical element prediction model to train the non-parametric weight function model.

[0069] In implementation, when executing step S103, the feature quantity A1 and the geographical elements of the sampling point B1 required for the non-parametric weight function model training can be obtained to construct the first sample set. After the first sample set is constructed, the constructed first sample set can be preprocessed and quality controlled and stored for later use.

[0070] The non-parametric weight function model determined in step S102 is trained using the stored first sample set and the geographic element prediction model determined in step S101.

[0071] S104, obtaining the feature quantities required for prediction of the target geographical elements and the geographical elements of the sampling points, constructing a second sample set, and interpolating part or all of the feature quantities to the required discrete grids.

[0072] In implementation, when executing step S104, the feature quantity A2 required for prediction of the target geographical element and the geographical element of the sampling point B2 can be obtained, a second sample set can be constructed, and part or all of the feature quantities in the feature quantity A2 can be interpolated to the required discrete grid as needed. After the second sample set is constructed, the constructed second sample set can be preprocessed and quality controlled, and stored for later use.

[0073] It should be added that the samples in the first sample set and the second sample set may be in the form of "sampling point target geographical element, feature quantity 1, feature quantity 2, ..., feature quantity n". In addition, the preprocessing and quality control of the first sample set and the second sample set may include but are not limited to removing redundant samples and abnormal samples, converting samples as needed, such as normalizing or standardizing geographical elements and feature quantities, and performing unique hot encoding on geographical elements and feature quantities.

[0074] S105 , using the first sample set or / and the second sample set, the geographic element prediction model and the trained non-parametric weight function model to predict the target geographic element of any discrete grid point.

[0075] In implementation, after executing step S103 and step S104, step S105 can be executed to use the first sample set constructed in step S103 and / or the second sample set constructed in step S104, the geographic element prediction model determined in step S101, and the trained non-parametric weight function model to predict the target geographic element of any discrete grid point.

[0076] In the above-mentioned geographic element prediction method based on the non-parametric weight function model provided by the embodiment of the present invention, the type of feature quantity, geographic element prediction model and spatiotemporal neighborhood control parameters required for the prediction of the target geographic element are first determined; then the non-parametric weight function model required for the prediction of the target geographic element is determined according to the spatiotemporal neighborhood control parameters; the input of the model is the distance from the sampling point or discrete grid point to a group of control points in its spatiotemporal neighborhood, and the output is the weight of the control point position in the spatiotemporal neighborhood; then the feature quantity required for the training of the non-parametric weight function model and the geographic elements of the sampling point are obtained, a first sample set is constructed, and the non-parametric weight function model is trained using the first sample set and the geographic element prediction model; in addition, the feature quantity required for the prediction of the target geographic element and the geographic elements of the sampling point are obtained, a second sample set is constructed, and part or all of the feature quantities in the feature quantity are interpolated to the required discrete grid; finally, the first sample set or / and the second sample set, the geographic element prediction model and the trained non-parametric weight function model are used to predict the target geographic element of any discrete grid. The non-parametric weight function model used in the present invention has a stronger expression ability, requires less human intervention, and enables the entire prediction method to have a higher level of automation and prediction accuracy.

[0077] Furthermore, in specific implementation, the geographic element prediction method based on the non-parametric weight function model provided in the embodiment of the present invention may also include: when there are multiple continuous target geographic elements that need to be predicted, determining the error weight of each continuous target geographic element to simultaneously model and predict multiple continuous target geographic elements; when there are multiple classified target geographic elements that need to be predicted, combining the multiple classified target geographic elements into a new classified target geographic element to simultaneously model and predict multiple classified target geographic elements.

[0078] In implementation, the target geographic element can be a continuous target geographic element or a categorical target geographic element. When there are multiple continuous target geographic elements that need to be predicted, after determining the weights of their respective errors, multiple continuous target geographic elements can be modeled and predicted at the same time. When there are multiple categorical target geographic elements that need to be predicted, multiple categorical target geographic elements can be combined into a new categorical target geographic element, thereby achieving simultaneous modeling and prediction of multiple categorical target geographic elements.

[0079] Furthermore, in specific implementation, the above-mentioned geographic element prediction method based on the non-parametric weight function model provided in the embodiment of the present invention may also include: when the control point is a sampling point within the space-time neighborhood, the output of the non-parametric weight function model is the weight of each sample within the space-time neighborhood; when the control point is a group of uniformly or unevenly distributed points artificially set within the space-time neighborhood, the weight of the control point position is interpolated to the sampling point position within the space-time neighborhood by interpolation to obtain the weight of the sample corresponding to the sampling point position.

[0080] In implementation, the output of the nonparametric weight function model is the weight of the control point position in the spatiotemporal neighborhood. The control point can be a sampling point in the spatiotemporal neighborhood, or a set of evenly or unevenly distributed points artificially set in the spatiotemporal neighborhood. When the control point is a sampling point in the spatiotemporal neighborhood, the output of the nonparametric weight function model can be the weight of each sample in the spatiotemporal neighborhood; and when the control point is a set of evenly or unevenly distributed points artificially set in the spatiotemporal neighborhood, the weight of the control point position can be interpolated to the sampling point position in the spatiotemporal neighborhood by interpolation to obtain the weight of the sample corresponding to the sampling point position.

[0081] Furthermore, in a specific implementation, in the above-mentioned geographic element prediction method based on the non-parametric weight function model provided in an embodiment of the present invention, step S103 uses the first sample set and the geographic element prediction model to train the non-parametric weight function model, which may specifically include: for any sampling point in the first sample set, according to the spatiotemporal neighborhood control parameters, determining the samples in the spatiotemporal neighborhood corresponding to the sampling point and the control points in the spatiotemporal neighborhood; using the determined samples in the spatiotemporal neighborhood and the control points in the spatiotemporal neighborhood, the geographic element prediction model and the non-parametric weight function model to be trained, constructing the relationship between the target geographic element and the feature quantity; using the constructed relationship between the target geographic element and the feature quantity, as well as the feature quantity of the sample corresponding to the sampling point, obtaining the predicted value of the target geographic element corresponding to the sampling point.

[0082] In implementation, the present invention can determine the samples and control points in the spatiotemporal neighborhood of any selected sampling point in the first sample set according to the spatiotemporal neighborhood control parameters, and then use the samples in the spatiotemporal neighborhood, the geographic element prediction model determined in step S101, and the non-parametric weight function model to be trained determined in step S102 to construct the relationship between the target geographic element and the characteristic quantity. The predicted value of the target geographic element corresponding to the sampling point is obtained by using the constructed relationship between the target geographic element and the characteristic quantity and the characteristic quantity of the sample corresponding to the sampling point. In the process of training the non-parametric weight function model, samples from multiple sampling points can be used to calculate the loss function.

[0083] Furthermore, in specific implementation, the above steps utilize samples within a determined space-time neighborhood and control points within the space-time neighborhood, a geographic element prediction model, and a non-parametric weight function model to be trained to construct a relationship between the target geographic element and the characteristic quantity. Specifically, this may include: calculating the distance from the sampling point to each control point within the determined space-time neighborhood; using the non-parametric weight function model to be trained to calculate the weight of each control point within the determined space-time neighborhood; obtaining the weight of each sample within the determined space-time neighborhood based on the calculated weight of each control point; and using samples within the determined space-time neighborhood and the corresponding weights, as well as a geographic element prediction model, to construct a relationship between the target geographic element and the characteristic quantity.

[0084] In implementation, when constructing the relationship between the target geographic element and the characteristic quantity, the present invention can first calculate the distance from the sampling point to each control point; then use the non-parametric weight function model to be trained determined in step S102 to calculate the weight of each control point; then, based on the weight of each control point, obtain the weight of each sample in the spatiotemporal neighborhood; finally, use the samples in the spatiotemporal neighborhood and the corresponding weights and the geographic element prediction model determined in step S101 to construct the relationship between the target geographic element and the characteristic quantity.

[0085] Furthermore, in specific implementation, in the above-mentioned geographic element prediction method based on the non-parametric weight function model provided in the embodiment of the present invention, step S105 uses the first sample set or / and the second sample set, the geographic element prediction model and the trained non-parametric weight function model to predict the target geographic element of any discrete grid point, which may specifically include: for any discrete grid point, according to the spatiotemporal neighborhood control parameters, determining the samples in the spatiotemporal neighborhood and the control points in the spatiotemporal neighborhood corresponding to the discrete grid point; using the determined samples in the spatiotemporal neighborhood and the control points in the spatiotemporal neighborhood, the geographic element prediction model and the trained non-parametric weight function model, constructing the relationship between the target geographic element and the feature quantity; using the constructed relationship between the target geographic element and the feature quantity, and the feature quantity corresponding to the discrete grid point, obtaining the predicted value of the target geographic element corresponding to the discrete grid point.

[0086] In implementation, the present invention can determine the samples and control points in the spatiotemporal neighborhood for any discrete grid point according to the spatiotemporal neighborhood control parameters, and use the samples in the spatiotemporal neighborhood, the geographic element prediction model and the trained non-parametric weight function model to construct the relationship between the target geographic element and the characteristic quantity. Then, the predicted value of the target geographic element corresponding to the discrete grid point is obtained by using the constructed relationship between the target geographic element and the characteristic quantity and the characteristic quantity corresponding to the discrete grid point.

[0087] Furthermore, in specific implementation, the above steps utilize samples and control points within a determined space-time neighborhood, a geographic element prediction model, and a trained non-parametric weight function model to construct a relationship between the target geographic element and the characteristic quantity. Specifically, this may include: calculating the distance from a discrete grid point to each control point within a determined space-time neighborhood; using a trained non-parametric weight function model to calculate the weight of each control point within the determined space-time neighborhood; obtaining the weight of each sample within the determined space-time neighborhood based on the calculated weight of each control point; and using samples within a determined space-time neighborhood and corresponding weights, as well as a geographic element prediction model, to construct a relationship between the target geographic element and the characteristic quantity.

[0088] In implementation, the present invention can first calculate the distance from the discrete grid point to each control point; then use the trained non-parametric weight function model to calculate the weight of each control point; then, according to the weight of each control point, obtain the weight of each sample in the space-time neighborhood; finally, use the samples in the space-time neighborhood and the corresponding weights and geographic element prediction model to construct the relationship between the target geographic element and the characteristic quantity.

[0089] Furthermore, in specific implementation, the above-mentioned geographic element prediction method based on the non-parametric weight function model provided in the embodiment of the present invention may also include: when the target geographic element is a continuous geographic element, generating a normalized or standardized value of the geographic element, and denormalizing or destandardizing the generated normalized or standardized value of the geographic element to obtain a final prediction value; when the target geographic element is a categorized geographic element, generating probability values ​​for each category of the geographic element, and obtaining the category to which the geographic element belongs as the final prediction value based on the generated probability values ​​for each category of the geographic element.

[0090] In implementation, when the target geographic element is a continuous geographic element, the predicted value is usually a normalized or standardized value; when the output geographic element is a categorized geographic element, the predicted value is usually a probability value of each category. In other words, after the geographic element is normalized or standardized, what is generated in this step is usually a normalized or standardized value of the geographic element. Therefore, in order to facilitate user use, after executing step S105, the present invention can further denormalize or destandardize the generated normalized or standardized value of the geographic element, so as to generate a geographic element value that is easy for the user to understand; if the probability value of each category of the geographic element is generated, the category to which the discrete grid point geographic element belongs can be further output based on the above probability value.

[0091] In the above embodiments, the geographic element prediction method based on the non-parametric weight function model is described in detail, and the present invention also provides a geographic element prediction device based on the non-parametric weight function model and a corresponding embodiment of the geographic element prediction device based on the non-parametric weight function model. It should be noted that the present invention describes the embodiments of the device part from two perspectives, one is based on the functional module perspective, and the other is based on the hardware perspective.

[0092] The embodiment of the present invention provides a geographic element prediction device based on a non-parametric weight function model from the perspective of functional modules, and the device includes:

[0093] The first determination module is used to determine the type of feature quantity, geographic element prediction model and spatiotemporal neighborhood control parameters required for the prediction of the target geographic element;

[0094] The second determination module is used to determine the non-parametric weight function model required for the prediction of the target geographic element according to the spatiotemporal neighborhood control parameters; the input of the non-parametric weight function model is the distance from the sampling point or discrete grid point to a group of control points in its spatiotemporal neighborhood, and the output is the weight of the control point position in the spatiotemporal neighborhood;

[0095] A model training module is used to obtain the feature quantities and geographical elements of the sampling points required for the training of the non-parametric weight function model, construct a first sample set, and train the non-parametric weight function model using the first sample set and the geographical element prediction model;

[0096] A grid interpolation module is used to obtain the feature quantities required for prediction of target geographic elements and the geographic elements of sampling points, construct a second sample set, and interpolate part or all of the feature quantities to the required discrete grid;

[0097] The model prediction module is used to predict the target geographical element of any discrete grid point by using the first sample set or / and the second sample set, the geographical element prediction model and the trained non-parametric weight function model.

[0098] In the geographic element prediction device based on the non-parametric weight function model provided in the embodiment of the present invention, the target geographic elements can be accurately acquired through the interaction of the above five modules, thereby improving the automation level and prediction accuracy.

[0099] Since the embodiments of the device part correspond to the embodiments of the method part, the embodiments of the device part refer to the description of the embodiments of the method part, which will not be described here. And it has the same beneficial effects as the above-mentioned geographic element prediction method based on the non-parametric weight function model.

[0100] Figure 2 Schematic diagram of the structure of the geographical element prediction device based on the non-parametric weight function model provided by the embodiment of the present invention. This embodiment is based on the hardware perspective, such as Figure 2 As shown, the geographic element prediction device based on the non-parametric weight function model includes:

[0101] A memory 20, used for storing computer programs;

[0102] The processor 21 is used to implement the steps of the geographic element prediction method based on the non-parametric weight function model mentioned in the above embodiment when executing the computer program.

[0103] Among them, the processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one hardware form of a digital signal processor (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU; the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 21 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0104] The memory 20 may include one or more computer-readable storage media, which may be non-transitory. The memory 20 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 20 is at least used to store the following computer program 201, wherein, after the computer program is loaded and executed by the processor 21, it can implement the relevant steps of the geographic element prediction method based on the non-parametric weight function model disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 20 may also include an operating system 202 and data 203, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 202 may include Windows, Unix, Linux, etc. The data 203 may include but is not limited to the data involved in the above-mentioned geographic element prediction method based on the non-parametric weight function model.

[0105] In some embodiments, the geographic element prediction device based on the non-parametric weight function model may also include a display screen 22, an input and output interface 23, a communication interface 24, a power supply 25 and a communication bus 26. Those skilled in the art will appreciate that Figure 2The structure shown in does not constitute a limitation on the geographic element prediction device based on the non-parametric weight function model, and may include more or fewer components than shown in the figure. The geographic element prediction device based on the non-parametric weight function model provided by the embodiment of the present invention includes a memory and a processor. When the processor executes the program stored in the memory, it can implement the geographic element prediction method based on the non-parametric weight function model mentioned above, and the effect is the same as above.

[0106] Finally, the present invention also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps recorded in the above method embodiment are implemented.

[0107] It is understandable that if the method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and executes 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), random access memory (RAM), disk or optical disk and other media that can store program code. The computer-readable storage medium provided by the present invention can realize the above-mentioned geographic element prediction method based on the non-parametric weight function model, and the effect is the same as above.

[0108] Finally, the present invention also provides an embodiment corresponding to a computer program product. The computer program product includes a computer program / instruction, and when the computer program / instruction is executed by a processor, the steps described in the embodiment of the geographic element prediction method based on the non-parametric weight function model are implemented. The computer program product provided by the present invention can implement the above-mentioned geographic element prediction method based on the non-parametric weight function model, and the effect is the same as above.

[0109] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0110] The above is a detailed introduction to the geographic element prediction method, device and medium based on the non-parametric weight function model provided by the present invention. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referenced to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the present invention.

Claims

1. A geographical element prediction method based on a non-parametric weight function model, characterized in that: The method comprises: Determine the types of characteristic quantities, geographic element prediction models and spatiotemporal neighborhood control parameters required for target geographic element prediction; Determine the non-parametric weight function model required for the prediction of the target geographic element according to the spatiotemporal neighborhood control parameters; the input of the non-parametric weight function model is the distance from the sampling point or discrete grid point to a group of control points in its spatiotemporal neighborhood, and the output is the weight of the control point position in the spatiotemporal neighborhood; Acquire the feature quantities and geographical elements of the sampling points required for training the non-parametric weight function model, construct a first sample set, and train the non-parametric weight function model using the first sample set and the geographical element prediction model; Acquire the feature quantities required for prediction of the target geographic elements and the geographic elements of the sampling points, construct a second sample set, and interpolate part or all of the feature quantities to the required discrete grids; The first sample set or / and the second sample set, the geographic element prediction model and the trained non-parametric weight function model are used to predict the target geographic element of any discrete grid point.

2. The geographic element prediction method based on the non-parametric weight function model according to claim 1 is characterized in that: Also includes: When there are multiple continuous target geographic elements that need to be predicted, the error weight of each continuous target geographic element is determined to simultaneously model and predict the multiple continuous target geographic elements; When there are multiple categorized target geographic elements that need to be predicted, the multiple categorized target geographic elements are combined into new categorized target geographic elements to simultaneously model and predict the multiple categorized target geographic elements.

3. The geographic element prediction method based on the non-parametric weight function model according to claim 1 is characterized in that: Also includes: When the control point is a sampling point in the spatiotemporal neighborhood, the output of the non-parametric weight function model is the weight of each sample in the spatiotemporal neighborhood; When the control point is a set of evenly or unevenly distributed points artificially set in the spatiotemporal neighborhood, the weight of the control point position is interpolated to the sampling point position in the spatiotemporal neighborhood by interpolation to obtain the weight of the sample corresponding to the sampling point position.

4. The geographic element prediction method based on the non-parametric weight function model according to claim 1 is characterized in that: The non-parametric weight function model is trained using the first sample set and the geographic element prediction model, including: For any sampling point in the first sample set, determine, according to the spatiotemporal neighborhood control parameter, a sample in the spatiotemporal neighborhood and a control point in the spatiotemporal neighborhood corresponding to the sampling point; Using the samples and control points in the determined spatiotemporal neighborhood, the geographic element prediction model and the non-parametric weight function model to be trained, the relationship between the target geographic element and the characteristic quantity is constructed; By using the relationship between the constructed target geographic elements and the characteristic quantities, as well as the characteristic quantities of the samples corresponding to the sampling points, the predicted values ​​of the target geographic elements corresponding to the sampling points are obtained.

5. The geographic element prediction method based on the non-parametric weight function model according to claim 4 is characterized in that: Using the samples and control points in the determined spatiotemporal neighborhood, the geographic element prediction model and the non-parametric weight function model to be trained, a relationship between the target geographic element and the characteristic quantity is constructed, including: Calculate the distance from the sampling point to each control point in the determined spatiotemporal neighborhood; Calculating the weight of each control point in the determined spatiotemporal neighborhood using the non-parametric weight function model to be trained; According to the calculated weight of each control point, the weight of each sample in the determined spatiotemporal neighborhood is obtained; The determined samples and corresponding weights in the spatiotemporal neighborhood and the geographic element prediction model are used to construct a relationship between the target geographic element and the characteristic quantity.

6. The geographic element prediction method based on the non-parametric weight function model according to claim 1 is characterized in that: Using the first sample set or / and the second sample set, the geographic element prediction model and the trained non-parametric weight function model to predict the target geographic element of any discrete grid point includes: For any discrete grid point, according to the spatiotemporal neighborhood control parameter, the samples and control points in the spatiotemporal neighborhood corresponding to the discrete grid point are determined; Using the samples and control points in the determined spatiotemporal neighborhood, the geographic element prediction model and the trained non-parametric weight function model, the relationship between the target geographic element and the characteristic quantity is constructed; By using the relationship between the constructed target geographic elements and the characteristic quantities, as well as the characteristic quantities corresponding to the discrete grid points, the predicted values ​​of the target geographic elements corresponding to the discrete grid points are obtained.

7. The geographic element prediction method based on the non-parametric weight function model according to claim 6 is characterized in that: Using the samples and control points in the determined spatiotemporal neighborhood, the geographic element prediction model and the trained non-parametric weight function model, the relationship between the target geographic element and the characteristic quantity is constructed, including: Calculate the distance from the discrete grid point to each control point in the determined space-time neighborhood; The trained non-parametric weight function model is used to calculate the weight of each control point in the determined spatiotemporal neighborhood; According to the calculated weight of each control point, the weight of each sample in the determined spatiotemporal neighborhood is obtained; The relationship between the target geographic element and the characteristic quantity is constructed by using the determined samples and corresponding weights in the spatiotemporal neighborhood and the geographic element prediction model.

8. The geographic element prediction method based on the non-parametric weight function model according to claim 1 is characterized in that: Also includes: When the target geographic element is a continuous geographic element, a normalized or standardized value of the geographic element is generated, and the generated normalized or standardized value of the geographic element is denormalized or destandardized to obtain a final predicted value; When the target geographic element is a categorized geographic element, probability values ​​for each category of the geographic element are generated, and based on the generated probability values ​​for each category of the geographic element, the category to which the geographic element belongs is obtained as the final prediction value.

9. A geographical element prediction device based on a non-parametric weight function model, characterized in that: The device comprises: Memory for storing computer programs; A processor is used to implement the steps of the geographic element prediction method based on the non-parametric weight function model as described in any one of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the geographic element prediction method based on the non-parametric weight function model as described in any one of claims 1 to 8 are implemented.

Images