Depth learning-based variation function automatic modeling method, device and equipment

Abstract

The invention discloses a variation function automatic modeling method, device and equipment based on deep learning, relates to the technical field of geostatistics, and can solve the technical problems of inaccurate variation function modeling and low efficiency at present. Comprising the steps that a to-be-trained variation function curved surface data set and an initial theoretical variation function parameter prediction model are constructed, and the to-be-trained variation function curved surface data set comprises a to-be-trained variation function curved surface and corresponding to-be-trained variation function parameters; training the initial theoretical variation function parameter prediction model according to the to-be-trained variation function curved surface and the corresponding to-be-trained variation function parameters to obtain a trained theoretical variation function parameter prediction model; and obtaining an experimental variation function value, calculating a variation function curved surface corresponding to the experimental variation function value, and predicting a theoretical variation function parameter corresponding to the variation function curved surface by using the theoretical variation function parameter prediction model so as to obtain a theoretical variation function according to the theoretical variation function parameter.

Description

Technical Field

[0001] The present invention relates to the technical field of geological statistics, and in particular to a method, device and equipment for automatic modeling of variogram based on deep learning. Background Art

[0002] In geostatistics, the theoretical variogram is a tool used to measure the difference or degree of variation between data values ​​at different locations in the geological space. It describes the spatial relationship and variability between data points and helps understand the spatial variation pattern of data. It is used to reveal the spatial correlation, spatial trend and spatial structure of data, and is helpful for analysis such as spatial interpolation, spatial prediction and geological resource assessment, and is of great significance in engineering practice.

[0003] At present, the conventional method for modeling theoretical variogram is to calculate the experimental variogram in the specified direction and then fit it to obtain the theoretical variogram. On the one hand, when calculating the experimental variogram in the specified direction, the reliability of the calculation result is closely related to the spatial distribution of the experimental variogram value. In the case of strict regularity in the spatial distribution of the experimental variogram value, there is no need to consider parameters such as tolerance, and only the basic lag distance, search direction and longest search distance are considered. However, in general, the distribution of experimental variogram values ​​is not strictly regular. At this time, if only the basic lag distance is considered, many sample points in the experimental variogram value cannot be involved in the calculation, and the obtained experimental variogram cannot fully represent the spatial structure of the experimental variogram value. In this case, the parameters that need to be set often include azimuth, bandwidth, search radius, tolerance angle, lag distance, lag distance tolerance, number of lag distances, etc., which are difficult to calculate, and the setting of parameters also requires a certain industry foundation. Reasonable parameters need to be set according to the industry background, which is subjective. Inappropriate parameter settings may result in inaccurate experimental variograms, which ultimately affect the modeling effect of the theoretical variogram model. On the other hand, when the theoretical variogram is obtained by fitting the experimental variogram, the experimental variogram scatter plot is drawn based on the experimental variogram, and the parameters such as the range value, arch height and nugget value are estimated from the experimental variogram scatter plot. The theoretical variogram model is drawn by the estimated range value, arch height and nugget value, and compared with the experimental variogram scatter plot, and the range value, arch height and nugget value are modified. This process is repeated until the theoretical variogram and the experimental variogram scatter plot have the highest matching degree. The conventional method has low fitting efficiency and high time cost. Summary of the invention

[0004] In view of this, the present invention provides a method, device and equipment for automatic modeling of variogram based on deep learning, which can solve the current technical problems of inaccurate and low efficiency of variogram modeling.

[0005] According to a first aspect of the present invention, a method for automatic modeling of a variogram based on deep learning is provided, the method comprising:

[0006] Constructing a variogram surface data set to be trained and constructing an initial theoretical variogram parameter prediction model, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained;

[0007] Training the initial theoretical variogram parameter prediction model according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model;

[0008] Acquire experimental variogram values, calculate the variogram surface corresponding to the experimental variogram values, and use the theoretical variogram parameter prediction model to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain the theoretical variogram according to the theoretical variogram parameters.

[0009] Preferably, the step of constructing a variogram surface dataset to be trained includes:

[0010] Randomly generate multiple sets of variogram parameters to be trained;

[0011] Obtaining a preset coordinate range, within the preset coordinate range, using the variogram parameters to be trained of each group to calculate the variogram values ​​to be trained of all data points, and calculating the variogram surface to be trained corresponding to the variogram values ​​to be trained;

[0012] Each group of the to-be-trained variogram parameters and the corresponding to-be-trained variogram surface is taken as a group of to-be-trained variogram surface data, and multiple groups of to-be-trained variogram data are determined as a to-be-trained variogram surface data set.

[0013] Preferably, the variogram parameters to be trained include the variogram type to be trained, and the step of constructing an initial theoretical variogram parameter prediction model includes:

[0014] Constructing a regression sub-model, wherein one regression sub-model corresponds to one type of the variogram to be trained;

[0015] A classification sub-model is constructed, and the classification sub-model and all the regression sub-models are determined as an initial theoretical variogram parameter prediction model.

[0016] Preferably, the constructing of the regression sub-model includes:

[0017] Construct convolutional layer, pooling layer, fully connected layer and regression sub-model output layer to obtain the regression sub-model;

[0018] The nodes of the output layer of the regression sub-model are determined according to the variogram parameters to be trained.

[0019] Preferably, the step of determining the nodes of the output layer of the regression sub-model according to the variogram parameters to be trained includes:

[0020] If the variogram surface data set to be trained is two-dimensional, the variogram parameters to be trained are the variogram type to be trained, the nugget value, the main direction, the main direction range value and the range ratio, and the nodes of the output layer of the regression sub-model are determined to be: the nugget value, the main direction, the main direction range value and the range ratio;

[0021] If the variogram surface data set to be trained is three-dimensional, then the variogram parameters to be trained are the variogram type to be trained, the main direction angle, the secondary direction angle, the short direction angle, the nugget value, the main direction range value, the secondary direction range value and the short direction range value, and the nodes of the output layer of the regression sub-model are determined to be: the main direction angle, the secondary direction angle, the short direction angle, the nugget value, the main direction range value, the secondary direction range value and the short direction range value.

[0022] Preferably, the step of training the initial theoretical variogram parameter prediction model according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model comprises:

[0023] Training the classification sub-model according to the variogram surface to be trained and the corresponding variogram type to be trained;

[0024] The regression sub-model corresponding to the variogram type to be trained is trained according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model.

[0025] Preferably, the step of predicting the theoretical variogram parameters corresponding to the variogram surface using the theoretical variogram parameter prediction model includes:

[0026] Mapping the variogram surface into the preset coordinate range to obtain a mapped variogram surface;

[0027] The mapped variogram surface is input into the theoretical variogram parameter prediction model to obtain the theoretical variogram parameters corresponding to the variogram surface.

[0028] According to a second aspect of the present invention, a device for automatic modeling of variogram based on deep learning is provided, the device comprising:

[0029] A construction module is used to construct a variogram surface data set to be trained and to construct an initial theoretical variogram parameter prediction model, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained;

[0030] A training module, used for training the initial theoretical variogram parameter prediction model according to the variogram surface to be trained and the corresponding variogram parameters to be trained, to obtain a trained theoretical variogram parameter prediction model;

[0031] A prediction module is used to obtain experimental variogram values, calculate the variogram surface corresponding to the experimental variogram values, and use the theoretical variogram parameter prediction model to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain a theoretical variogram based on the theoretical variogram parameters.

[0032] According to the third aspect of the present application, a storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned automatic modeling method of the variogram based on deep learning is implemented.

[0033] According to the fourth aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein when the processor executes the program, the above-mentioned automatic modeling method of variogram based on deep learning is implemented.

[0034] By means of the above technical scheme, the present invention provides a method, device and equipment for automatic modeling of variogram based on deep learning. First, a variogram surface data set to be trained and an initial theoretical variogram parameter prediction model are constructed, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained; then, the initial theoretical variogram parameter prediction model is trained according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model; finally, the experimental variogram value is obtained, the variogram surface corresponding to the experimental variogram value is calculated, and the theoretical variogram parameter prediction model is used to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain the theoretical variogram according to the theoretical variogram parameters. Through the technical scheme of the present invention, on the one hand, the theoretical variogram parameter prediction model obtained by training learns the relationship between the variogram surface to be trained and the variogram parameters to be trained, so that when the theoretical variogram parameter prediction model is applied, the theoretical variogram parameters can be directly obtained from the variogram surface corresponding to the experimental variogram value, without the need for repeated estimation through the experimental variogram scatter plot, thereby improving the efficiency of automatic variogram modeling. On the other hand, the prior art and the present invention are both based on the experimental variogram value, but the prior art calculates the experimental variogram in a specified direction based on the experimental variogram value, which requires some parameters to be set manually, while the calculation of the variogram surface of the present invention uses the experimental variogram value in the entire space, and does not limit the direction. Therefore, the calculation of the variogram surface does not require some parameters to be set manually, so it is not affected by subjective factors. The variogram surface corresponding to the experimental variogram value is used to replace the experimental variogram in the specified direction of the prior art, thereby improving the accuracy of automatic variogram modeling.

[0035] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation on the local application. In the drawings:

[0037] Figure 1 A schematic diagram of a process flow of a method for automatic modeling of a variogram based on deep learning provided by an embodiment of the present invention is shown;

[0038] Figure 2A schematic flow chart of another method for automatic modeling of a variogram based on deep learning provided by an embodiment of the present invention is shown;

[0039] Figure 3 A schematic diagram of the structure of a variogram automatic modeling device based on deep learning provided by an embodiment of the present invention is shown;

[0040] Figure 4 A schematic diagram of the structure of another automatic modeling device for variogram based on deep learning provided by an embodiment of the present invention is shown;

[0041] Figure 5 An example diagram of a two-dimensional variogram surface to be trained provided by an embodiment of the present invention is shown;

[0042] Figure 6 An example diagram of a three-dimensional variogram surface to be trained provided by an embodiment of the present invention is shown;

[0043] Figure 7 A two-dimensional experimental variation function value provided by an embodiment of the present invention is shown;

[0044] Figure 8 A three-dimensional experimental variation function value provided by an embodiment of the present invention is shown;

[0045] Fig. 9 A two-dimensional variogram surface provided by an embodiment of the present invention is shown;

[0046] Fig.10 A three-dimensional variogram surface provided by an embodiment of the present invention is shown;

[0047] Fig.11 An existing two-dimensional theoretical variogram modeling result provided by an embodiment of the present invention is shown;

[0048] Fig.12 The figure shows a two-dimensional theoretical variogram modeling result of an embodiment of the present application provided by the embodiment of the present invention;

[0049] Fig.13 An existing three-dimensional theoretical variogram modeling result provided by an embodiment of the present invention is shown;

[0050] Fig.14 The figure shows a three-dimensional theoretical variogram modeling result of an embodiment of the present application provided by the embodiment of the present invention;

[0051] Fig.15 A two-dimensional Kriging estimation error histogram provided by an embodiment of the present invention is shown;

[0052] Fig.16A three-dimensional Kriging estimation error histogram provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0053] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0054] This embodiment provides a method for automatic modeling of variogram based on deep learning, such as Figure 1 As shown, the method includes:

[0055] 101. Construct a variogram surface data set to be trained and construct an initial theoretical variogram parameter prediction model, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained.

[0056] 102. Train the initial theoretical variogram parameter prediction model according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model.

[0057] For step 101 of the embodiment, the variogram surface data set to be trained includes multiple groups of variogram surface data to be trained, each group of variogram surface data to be trained includes a variogram surface to be trained and a corresponding group of variogram parameters to be trained, a group of variogram parameters to be trained includes multiple variogram parameters to be trained, and each variogram surface to be trained is labeled with the corresponding multiple variogram parameters to be trained.

[0058] Taking the example that the variogram surface data set to be trained includes 3 groups of variogram surface data to be trained, and one group of variogram parameters to be trained includes 3 variogram parameters to be trained, the variogram surface data set to be trained includes: {[variogram surface to be trained 1, variogram parameter to be trained 11, variogram parameter to be trained 12, variogram parameter to be trained 13]; [variogram surface to be trained 2, variogram parameter to be trained 21, variogram parameter to be trained 22, variogram parameter to be trained 23]; [variogram surface to be trained 3, variogram parameter to be trained 31, variogram parameter to be trained 32, variogram parameter to be trained 33]}.

[0059] For the embodiment steps 101 and 102, the initial theoretical variogram parameter prediction model is untrained, and the variogram surface data set to be trained is divided into a training set, a test set and a validation set according to a preset ratio, and the initial theoretical variogram parameter prediction model is trained according to the training set, the test set and the validation set to obtain a trained theoretical variogram parameter prediction model, wherein the initial theoretical variogram parameter prediction model can be a convolutional neural network (CNN), which is not limited here. During the training process, as an implementation method, the epoch is set to 300, and Adam is selected as the optimizer. Adam uses momentum as the parameter update direction, can adaptively adjust the learning rate to improve the gradient descent, has efficient and stable optimization performance, can usually converge quickly and has high training efficiency, and the weight decay parameter in the Adam optimizer is used to perform L2 regularization on the weight parameters of the model. By applying a penalty term to the weight parameters, the model can be better generalized to new data and reduce the risk of overfitting. The weight decay parameter of the Adam optimizer is set to 0.0001. During the training process, StepLR is used as the learning rate adjuster (implementing the strategy of adjusting the learning rate by step size). The step size and decay factor of StepLR are set to 10 and 0.5, that is, after every 10 training cycles, the learning rate will be updated to half of the current value. The initial learning rate is set to 0.001, and the batch size is set to 32. The model is trained with the above parameter settings.

[0060] 103. Obtain experimental variogram values, calculate the variogram surface corresponding to the experimental variogram values, and use the theoretical variogram parameter prediction model to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain the theoretical variogram according to the theoretical variogram parameters.

[0061] For this embodiment, the specific process of calculating the variogram surface corresponding to the experimental variogram value is the same as the prior art, and will not be repeated here. When applying the theoretical variogram parameter prediction model, the input is the variogram surface corresponding to the experimental variogram value, and the output is the theoretical variogram parameter, which is the theoretical variogram parameter required for automatically modeling the experimental variogram value into the theoretical variogram.

[0062] The trained theoretical variogram parameter prediction model can effectively capture the trends and overall characteristics of the variogram surface, apply deep learning methods to the field of variogram modeling, automatically optimize the theoretical variogram parameters, and improve the accuracy and efficiency of automatic variogram modeling.

[0063] The present invention provides a method, device and equipment for automatic modeling of variogram based on deep learning. First, a variogram surface data set to be trained and an initial theoretical variogram parameter prediction model are constructed, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained; then, the initial theoretical variogram parameter prediction model is trained according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model; finally, experimental variogram values ​​are obtained, the variogram surface corresponding to the experimental variogram values ​​are calculated, and the theoretical variogram parameter prediction model is used to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain the theoretical variogram according to the theoretical variogram parameters. Through the technical scheme of the present invention, on the one hand, the theoretical variogram parameter prediction model obtained by training learns the relationship between the variogram surface to be trained and the variogram parameters to be trained, so that when the theoretical variogram parameter prediction model is applied, the theoretical variogram parameters can be directly obtained from the variogram surface corresponding to the experimental variogram value, without the need for repeated estimation through the experimental variogram scatter plot, thereby improving the efficiency of automatic variogram modeling. On the other hand, the prior art and the present invention are both based on the experimental variogram value, but the prior art calculates the experimental variogram in a specified direction based on the experimental variogram value, which requires some parameters to be set manually, while the calculation of the variogram surface of the present invention uses the experimental variogram value in the entire space, and does not limit the direction. Therefore, the calculation of the variogram surface does not require some parameters to be set manually, so it is not affected by subjective factors. The variogram surface corresponding to the experimental variogram value is used to replace the experimental variogram in the specified direction of the prior art, thereby improving the accuracy of automatic variogram modeling.

[0064] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, another automatic modeling method of the variogram based on deep learning is provided, such as Figure 2 As shown, the method includes:

[0065] 201. Construct a variogram surface dataset to be trained, wherein the variogram surface dataset to be trained includes a variogram surface to be trained and corresponding variogram parameters to be trained, and the variogram parameters to be trained include a variogram type to be trained.

[0066] For this embodiment, as an implementation method, the construction of the to-be-trained variogram surface data set includes: randomly generating multiple groups of to-be-trained variogram parameters; obtaining a preset coordinate range, within the preset coordinate range, using each group of the to-be-trained variogram parameters to calculate the to-be-trained variogram values ​​of all data points, and calculating the to-be-trained variogram surface corresponding to the to-be-trained variogram values; taking each group of the to-be-trained variogram parameters and the corresponding to-be-trained variogram surface as a group of to-be-trained variogram surface data, and determining multiple groups of the to-be-trained variogram data as a to-be-trained variogram surface data set.

[0067] Among them, the variogram surface data set to be trained is divided into two-dimensional and three-dimensional. Regardless of whether the variogram surface data set to be trained is two-dimensional or three-dimensional, the construction method is the same. Each group of variogram parameters to be trained includes multiple variogram parameters to be trained, one of which is a variogram type to be trained. The variogram types to be trained include: spherical model, exponential model and Gaussian model.

[0068] If the variogram surface data set to be trained is two-dimensional, then the variogram parameters to be trained are the variogram type to be trained, the nugget value, the main direction, the main direction range value and the range ratio; if the variogram surface data set to be trained is three-dimensional, then the variogram parameters to be trained are the variogram type to be trained, the main direction angle, the secondary direction angle, the short direction angle, the nugget value, the main direction range value, the secondary direction range value and the short direction range value.

[0069] Among them, Figure 5 As shown in the figure, there is a set of two-dimensional variogram surface examples to be trained, and the corresponding set of two-dimensional variogram parameters to be trained are: model type, exponential model; nugget value, 0.1540, main direction, 57.99°, main direction range value, 651.77, range ratio, 0.61. Figure 6 As shown, a set of three-dimensional variogram surface examples to be trained are displayed through the X-axis section, Y-axis section, Z-axis section and XY diagonal section, and the corresponding set of three-dimensional variogram parameters to be trained are: model type, exponential model; main direction angle, 161.47°; secondary direction angle, 16.32°; short direction angle, -82.21°; nugget value, 0.3952; main direction range value, 98.14; secondary direction range value, 65.75; short direction range value, 46.12.

[0070] The specific construction method is as follows: if the variogram surface data set to be trained is two-dimensional, a preset coordinate range is obtained (this preset coordinate range is two-dimensional), within the preset coordinate range, the variogram values ​​to be trained of all data points (the data points are two-dimensional data points) are calculated using the variogram parameters to be trained (two-dimensional) of each group, and the variogram surface to be trained corresponding to the variogram values ​​to be trained is calculated. If the variogram surface data set to be trained is three-dimensional, a preset coordinate range is obtained (this preset coordinate range is three-dimensional), within the preset coordinate range, the variogram values ​​to be trained of all data points (the data points are three-dimensional data points) are calculated using the variogram parameters to be trained (three-dimensional) of each group, and the variogram surface to be trained corresponding to the variogram values ​​to be trained is calculated.

[0071] The specific calculation process for calculating the training variogram surface corresponding to the training variogram value is referred to the prior art and will not be described in detail here. Preferably, the constructed training variogram surface data set can be preprocessed, including data standardization, size adjustment and other operations.

[0072] 202. Construct an initial theoretical variogram parameter prediction model.

[0073] For this embodiment, as an implementation method, the construction of the initial theoretical variogram parameter prediction model includes: constructing a regression sub-model, wherein one regression sub-model corresponds to one type of the variogram to be trained; constructing a classification sub-model, and determining one classification sub-model and all the regression sub-models as the initial theoretical variogram parameter prediction model.

[0074] Among them, the initial theoretical variogram parameter prediction model includes a two-dimensional initial theoretical variogram parameter prediction model and a three-dimensional initial theoretical variogram parameter prediction model. Regardless of the two-dimensional initial theoretical variogram parameter prediction model or the three-dimensional one, when the variogram types to be trained are spherical model, exponential model and Gaussian model, the initial theoretical variogram parameter prediction model is a classification sub-model and three regression sub-models. The classification sub-model is used to learn the variogram types to be trained. When there are three types of variograms to be trained, there are three results output by the output layer of the classification sub-model. The three regression sub-models are the regression sub-model corresponding to the spherical model, the regression sub-model corresponding to the exponential model, and the regression sub-model corresponding to the Gaussian model.

[0075] Specifically, for the regression sub-model corresponding to the spherical model, if it is a two-dimensional initial theoretical variogram parameter prediction model, it is used to learn the nugget value, main direction, main direction range value and range ratio corresponding to the spherical model; if it is a three-dimensional initial theoretical variogram parameter prediction model, it is used to learn the main direction angle, secondary direction angle, short direction angle, nugget value, main direction range value, secondary direction range value and short direction range value corresponding to the spherical model. For the regression sub-model corresponding to the exponential model, if it is a two-dimensional initial theoretical variogram parameter prediction model, it is used to learn the nugget value, main direction, main direction range value and range ratio corresponding to the exponential model; if it is a three-dimensional initial theoretical variogram parameter prediction model, it is used to learn the main direction angle, secondary direction angle, short direction angle, nugget value, main direction range value, secondary direction range value and short direction range value corresponding to the exponential model. For the regression sub-model corresponding to the Gaussian model, if it is a two-dimensional initial theoretical variogram parameter prediction model, it is used to learn the nugget value, main direction, main direction range value and range ratio corresponding to the Gaussian model; if it is a three-dimensional initial theoretical variogram parameter prediction model, it is used to learn the main direction angle, secondary direction angle, short direction angle, nugget value, main direction range value, secondary direction range value and short direction range value corresponding to the Gaussian model.

[0076] As an implementation method, for two-dimensional, the structure of the classification sub-model (in order) is: classification sub-model input layer (Input), the first convolutional layer (Conv1), batch normalization layer (ie BN layer, Batch Normalization), activation function layer (Relu), maximum pooling layer (MaxPool), second convolutional layer (Conv2), third convolutional layer (Conv3), fourth convolutional layer (Conv4), fifth convolutional layer (Conv5), global average pooling layer (Global Average Pooling), fully connected layer (Fully connected layer, i.e. FC), classification sub-model output layer (Output).

[0077] For three-dimensional, the structure of the classification sub-model is shown in Table 1:

[0078]

[0079] The basic idea of ​​Dropout in Table 1 is to randomly dropout (discard) the activation of some neurons during training to prevent the model from overfitting, which can make the model more robust.

[0080] Wherein, the construction of the regression sub-model includes: constructing a convolution layer, a pooling layer, a fully connected layer and a regression sub-model output layer to obtain a regression sub-model; determining the nodes of the regression sub-model output layer according to the parameters of the variogram to be trained. As an implementation method, for two-dimensional, the structure of the regression sub-model (in order) is: regression sub-model input layer (Input), the first convolution layer (Conv1), the first pooling layer (Pool1), the second convolution layer (Conv2), the second pooling layer (Pool2), the third convolution layer (Conv3), the third pooling layer (Pool3), the fourth convolution layer (Conv4), the fourth pooling layer (Pool4), the fifth convolution layer (Conv5), the fifth pooling layer (Pool5), the global average pooling layer (Global AveragePooling, FC), the fully connected layer (Fully connected layer), the regression sub-model output layer (Output).

[0081] For three dimensions, the structure of the regression sub-model is shown in Table 2:

[0082] Table 2:

[0083]

[0084] The flatten in Table 2 refers to the tensor operation.

[0085] For two-dimensional and three-dimensional, in addition to the different structures of the regression sub-models, the nodes of the output layer of the regression sub-models are also different. Specifically: the determining the nodes of the output layer of the regression sub-model according to the variogram parameters to be trained includes: if the variogram surface data set to be trained is two-dimensional, the variogram parameters to be trained are the variogram type to be trained, the nugget value, the main direction, the main direction range value and the range ratio, and the nodes of the output layer of the regression sub-model are determined to be: the nugget value, the main direction, the main direction range value and the range ratio; if the variogram surface data set to be trained is three-dimensional, the variogram parameters to be trained are the variogram type to be trained, the main direction angle, the secondary direction angle, the short direction angle, the nugget value, the main direction range value, the secondary direction range value and the short direction range value, and the nodes of the output layer of the regression sub-model are determined to be: the main direction angle, the secondary direction angle, the short direction angle, the nugget value, the main direction range value, the secondary direction range value and the short direction range value.

[0086] 203. Train the classification sub-model according to the variogram surface to be trained and the corresponding variogram type to be trained.

[0087] 204. Train the regression sub-model corresponding to the variogram type to be trained according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model.

[0088] For the embodiment steps 203 and 204, it is divided into using the two-dimensional variogram surface data set to be trained to train a two-dimensional initial theoretical variogram parameter prediction model (a two-dimensional classification sub-model and a two-dimensional regression sub-model), and using the three-dimensional variogram surface data set to be trained to train a three-dimensional initial theoretical variogram parameter prediction model (a three-dimensional classification sub-model and a three-dimensional regression sub-model).

[0089] Wherein, both the two-dimensional classification sub-model and the three-dimensional classification sub-model are used to learn the type of the variogram to be trained. For training the two-dimensional initial theoretical variogram parameter prediction model (the two-dimensional classification sub-model and the two-dimensional regression sub-model) using the two-dimensional variogram surface data set to be trained, specifically, if the variogram type to be trained is a spherical model, then the regression sub-model corresponding to the two-dimensional spherical model is trained, if the variogram type to be trained is an exponential model, then the regression sub-model corresponding to the two-dimensional exponential model is trained, if the variogram type to be trained is a Gaussian model, then the regression sub-model corresponding to the two-dimensional Gaussian model is trained. For training the three-dimensional initial theoretical variogram parameter prediction model (the three-dimensional classification sub-model and the three-dimensional regression sub-model) using the three-dimensional variogram surface data set to be trained, specifically, if the variogram type to be trained is a spherical model, then the regression sub-model corresponding to the three-dimensional spherical model is trained, if the variogram type to be trained is an exponential model, then the regression sub-model corresponding to the three-dimensional exponential model is trained, if the variogram type to be trained is a Gaussian model, then the regression sub-model corresponding to the three-dimensional Gaussian model is trained.

[0090] Finally, a two-dimensional theoretical variogram parameter prediction model and a three-dimensional theoretical variogram parameter prediction model are obtained.

[0091] 205. Obtain experimental variogram values, calculate the variogram surface corresponding to the experimental variogram values, and use the theoretical variogram parameter prediction model to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain the theoretical variogram according to the theoretical variogram parameters.

[0092] For this embodiment, the use of the theoretical variogram parameter prediction model to predict the theoretical variogram parameters corresponding to the variogram surface includes: mapping the variogram surface into the preset coordinate range to obtain the mapped variogram surface; inputting the mapped variogram surface into the theoretical variogram parameter prediction model to obtain the theoretical variogram parameters corresponding to the variogram surface.

[0093] Among them, the experimental variogram value includes the two-dimensional experimental variogram value (such as Figure 7 ) and the three-dimensional experimental variogram value (such as Figure 8 ), the method of calculating the variogram surface corresponding to the experimental variogram value refers to the prior art and will not be described here. Since the theoretical variogram parameter prediction model is trained within a preset coordinate range, in order to ensure the prediction accuracy, it is necessary to map the variogram surface to the preset coordinate range. Specifically, if it is a two-dimensional experimental variogram value, the two-dimensional variogram surface (such as Fig. 9 ), map the two-dimensional variogram surface to a two-dimensional preset coordinate range to obtain a two-dimensional mapped variogram surface, input the two-dimensional mapped variogram surface into a two-dimensional theoretical variogram parameter prediction model, and obtain the two-dimensional theoretical variogram parameters corresponding to the two-dimensional variogram surface. Once the two-dimensional theoretical variogram parameters are known, the automatic modeling of the two-dimensional theoretical variogram is completed (such as Fig.12 is the automatic modeling result of the two-dimensional theoretical variation function of the embodiment of the present application, and Fig.11 The existing two-dimensional modeling results can be compared). The automatic modeling of the three-dimensional theoretical variogram is similar and will not be described here (such as Fig.10 The three-dimensional variogram surface is calculated from the three-dimensional experimental variogram value, such as Fig.14 The automatic modeling result of the three-dimensional theoretical variation function of the embodiment of the present application is Fig.13 The existing three-dimensional modeling results can be compared).

[0094] After obtaining the theoretical variogram, the theoretical variogram can be used to analyze the characteristics of spatial data. It is mainly used in Kriging interpolation to estimate the value of unknown locations based on the spatial correlation between sample points at known locations. It can be widely used in geological resource assessment, pollutant diffusion prediction, surface water or atmospheric data modeling, etc. Fig.15 and Fig.16 As shown, whether it is two-dimensional or three-dimensional, the experimental variogram fitting model in the figure is the prior art, and the CNN fitting model is the automatic modeling result of the embodiment of the present application. The two are used to perform Kriging valuation respectively, and the obtained Kriging valuation results are as follows: Fig.15 As shown, the three-dimensional Fig.16 As shown, by comparison, from the two aspects of RMSE (Root Mean Square Error) and MAE (Mean Absolute Error) of error comparison, the Kriging estimation error in the actual application of the CNN fitting model is smaller than that of the experimental variogram fitting model. Therefore, it can be confirmed that the automatic modeling results of the embodiment of the present application are more accurate than those of the prior art.

[0095] The present invention provides a method, device and equipment for automatic modeling of variogram based on deep learning. First, a variogram surface data set to be trained and an initial theoretical variogram parameter prediction model are constructed, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained; then, the initial theoretical variogram parameter prediction model is trained according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model; finally, experimental variogram values ​​are obtained, the variogram surface corresponding to the experimental variogram values ​​are calculated, and the theoretical variogram parameter prediction model is used to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain the theoretical variogram according to the theoretical variogram parameters. Through the technical scheme of the present invention, on the one hand, the theoretical variogram parameter prediction model obtained by training learns the relationship between the variogram surface to be trained and the variogram parameters to be trained, so that when the theoretical variogram parameter prediction model is applied, the theoretical variogram parameters can be directly obtained from the variogram surface corresponding to the experimental variogram value, without the need for repeated estimation through the experimental variogram scatter plot, thereby improving the efficiency of automatic variogram modeling. On the other hand, the prior art and the present invention are both based on the experimental variogram value, but the prior art calculates the experimental variogram in a specified direction based on the experimental variogram value, which requires some parameters to be set manually, while the calculation of the variogram surface of the present invention uses the experimental variogram value in the entire space, and does not limit the direction. Therefore, the calculation of the variogram surface does not require some parameters to be set manually, so it is not affected by subjective factors. The variogram surface corresponding to the experimental variogram value is used to replace the experimental variogram in the specified direction of the prior art, thereby improving the accuracy of automatic variogram modeling.

[0096] Further, as Figure 1 and Figure 2 The specific implementation of the method shown in the figure, the embodiment of the present invention provides a device for automatically modeling a variogram based on deep learning, such as Figure 3 As shown, the device includes: a construction module 31, a training module 32, and a prediction module 33;

[0097] A construction module 31 is used to construct a variogram surface data set to be trained and to construct an initial theoretical variogram parameter prediction model, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained;

[0098] A training module 32, used for training the initial theoretical variogram parameter prediction model according to the variogram surface to be trained and the corresponding variogram parameters to be trained, to obtain a trained theoretical variogram parameter prediction model;

[0099] The prediction module 33 is used to obtain the experimental variogram value, calculate the variogram surface corresponding to the experimental variogram value, and use the theoretical variogram parameter prediction model to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain the theoretical variogram according to the theoretical variogram parameters.

[0100] Accordingly, in order to construct a variogram surface data set to be trained, the construction module 31 can be specifically used to randomly generate multiple groups of variogram parameters to be trained; obtain a preset coordinate range, and within the preset coordinate range, use the variogram parameters to be trained of each group to calculate the variogram values ​​to be trained of all data points, and calculate the variogram surface to be trained corresponding to the variogram values ​​to be trained; take each group of the variogram parameters to be trained and the corresponding variogram surface to be trained as a group of variogram surface data to be trained, and determine the multiple groups of variogram data to be trained as a variogram surface data set to be trained.

[0101] Correspondingly, the variogram parameter to be trained includes the variogram type to be trained. In order to construct an initial theoretical variogram parameter prediction model, the construction module 31 may specifically include: a first construction unit 311 and a second construction unit 312;

[0102] A first construction unit 311 may be specifically used to construct a regression sub-model, wherein one regression sub-model corresponds to one type of the variogram to be trained;

[0103] The second construction unit 312 can be specifically used to construct a classification sub-model, and determine the classification sub-model and all the regression sub-models as an initial theoretical variogram parameter prediction model.

[0104] Correspondingly, in order to construct the regression sub-model, the first construction unit 311 can be specifically used to construct a convolution layer, a pooling layer, a fully connected layer and an output layer to obtain the regression sub-model; and determine the nodes of the output layer according to the variogram parameters to be trained.

[0105] Correspondingly, in order to determine the nodes of the output layer according to the variogram parameters to be trained, the first construction unit 311 can be specifically used to determine the nodes of the output layer as follows: if the variogram surface data set to be trained is two-dimensional, then the variogram parameters to be trained are the variogram type to be trained, the nugget value, the main direction, the main direction range value and the range ratio, and the nodes of the output layer are determined to be: the nugget value, the main direction, the main direction range value and the range ratio; if the variogram surface data set to be trained is three-dimensional, then the variogram parameters to be trained are the variogram type to be trained, the main direction angle, the secondary direction angle, the short direction angle, the nugget value, the main direction range value, the secondary direction range value and the short direction range value, and the nodes of the output layer are determined to be: the main direction angle, the secondary direction angle, the short direction angle, the nugget value, the main direction range value, the secondary direction range value and the short direction range value.

[0106] Correspondingly, in order to train the initial theoretical variogram parameter prediction model according to the variogram surface to be trained and the corresponding variogram parameters to be trained, and obtain the trained theoretical variogram parameter prediction model, the training module 32 can be specifically used to train the classification sub-model according to the variogram surface to be trained and the corresponding variogram type to be trained; train the regression sub-model corresponding to the variogram type to be trained according to the variogram surface to be trained and the corresponding variogram parameters to be trained, and obtain the trained theoretical variogram parameter prediction model.

[0107] Correspondingly, in order to predict the theoretical variogram parameters corresponding to the variogram surface using the theoretical variogram parameter prediction model, the prediction module 33 can be specifically used to map the variogram surface to the preset coordinate range to obtain the mapped variogram surface; input the mapped variogram surface into the theoretical variogram parameter prediction model to obtain the theoretical variogram parameters corresponding to the variogram surface.

[0108] It should be noted that for other corresponding descriptions of the functional units involved in the automatic modeling device for variogram based on deep learning provided in this embodiment, please refer to Figure 1 to Figure 2 The corresponding description will not be repeated here.

[0109] Based on the above Figure 1 to Figure 2 The method shown in the embodiment, accordingly, also provides a storage medium, which can be volatile or non-volatile, and stores a computer program, which is executed by a processor to implement the above-mentioned Figure 1 to Figure 2 The deep learning-based automatic modeling method of variogram is shown.

[0110] Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of various implementation scenarios of the present invention.

[0111] Based on the above Figure 1 to Figure 2 The method shown and Figure 3 , Figure 4 In order to achieve the above-mentioned purpose, the present embodiment further provides a computer device, which includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1 to Figure 2 The deep learning-based automatic modeling method of variogram is shown.

[0112] Optionally, the computer device may further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0113] Those skilled in the art will appreciate that the computer device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0114] The storage medium may also include an operating network communication module. The operating is a program that manages the hardware and software resources of the above-mentioned computer device, supporting the operation of the information processing program and other software and / or programs. The network communication module is used to communicate between the components inside the storage medium, and to communicate with other hardware and software in the information processing entity device.

[0115] Through the description of the above implementation modes, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform, or by hardware.

[0116] The present invention provides a method, device and equipment for automatic modeling of variogram based on deep learning. First, a variogram surface data set to be trained and an initial theoretical variogram parameter prediction model are constructed, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained; then, the initial theoretical variogram parameter prediction model is trained according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model; finally, experimental variogram values ​​are obtained, the variogram surface corresponding to the experimental variogram values ​​are calculated, and the theoretical variogram parameter prediction model is used to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain the theoretical variogram according to the theoretical variogram parameters. Through the technical scheme of the present invention, on the one hand, the theoretical variogram parameter prediction model obtained by training learns the relationship between the variogram surface to be trained and the variogram parameters to be trained, so that when the theoretical variogram parameter prediction model is applied, the theoretical variogram parameters can be directly obtained from the variogram surface corresponding to the experimental variogram value, without the need for repeated estimation through the experimental variogram scatter plot, thereby improving the efficiency of automatic variogram modeling. On the other hand, the prior art and the present invention are both based on the experimental variogram value, but the prior art calculates the experimental variogram in a specified direction based on the experimental variogram value, which requires some parameters to be set manually, while the calculation of the variogram surface of the present invention uses the experimental variogram value in the entire space, and does not limit the direction. Therefore, the calculation of the variogram surface does not require some parameters to be set manually, so it is not affected by subjective factors. The variogram surface corresponding to the experimental variogram value is used to replace the experimental variogram in the specified direction of the prior art, thereby improving the accuracy of automatic variogram modeling.

[0117] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present invention. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0118] The above serial numbers of the present invention are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of the present invention, but the present invention is not limited thereto, and any changes that can be thought of by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for automatic modeling of variogram based on deep learning, characterized in that: The method comprises: Constructing a variogram surface data set to be trained and constructing an initial theoretical variogram parameter prediction model, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained; Training the initial theoretical variogram parameter prediction model according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model; Acquire experimental variogram values, calculate the variogram surface corresponding to the experimental variogram values, and use the theoretical variogram parameter prediction model to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain the theoretical variogram according to the theoretical variogram parameters.

2. The method according to claim 1, characterized in that The step of constructing a variogram surface data set to be trained includes: Randomly generate multiple sets of variogram parameters to be trained; Obtaining a preset coordinate range, within the preset coordinate range, using the variogram parameters to be trained of each group to calculate the variogram values ​​to be trained of all data points, and calculating the variogram surface to be trained corresponding to the variogram values ​​to be trained; Each group of the to-be-trained variogram parameters and the corresponding to-be-trained variogram surface is taken as a group of to-be-trained variogram surface data, and multiple groups of to-be-trained variogram data are determined as a to-be-trained variogram surface data set.

3. The method according to claim 1, characterized in that The variogram parameters to be trained include the variogram type to be trained, and the initial theoretical variogram parameter prediction model is constructed, including: Constructing a regression sub-model, wherein one regression sub-model corresponds to one type of the variogram to be trained; A classification sub-model is constructed, and the classification sub-model and all the regression sub-models are determined as an initial theoretical variogram parameter prediction model.

4. The method according to claim 3, characterized in that The constructing of the regression sub-model comprises: Construct convolutional layer, pooling layer, fully connected layer and regression sub-model output layer to obtain the regression sub-model; The nodes of the output layer of the regression sub-model are determined according to the variogram parameters to be trained.

5. The method according to claim 4, characterized in that The step of determining the nodes of the output layer of the regression sub-model according to the variogram parameter to be trained comprises: If the variogram surface data set to be trained is two-dimensional, the variogram parameters to be trained are the variogram type to be trained, the nugget value, the main direction, the main direction range value and the range ratio, and the nodes of the output layer of the regression sub-model are determined to be: the nugget value, the main direction, the main direction range value and the range ratio; If the variogram surface data set to be trained is three-dimensional, then the variogram parameters to be trained are the variogram type to be trained, the main direction angle, the secondary direction angle, the short direction angle, the nugget value, the main direction range value, the secondary direction range value and the short direction range value, and the nodes of the output layer of the regression sub-model are determined to be: the main direction angle, the secondary direction angle, the short direction angle, the nugget value, the main direction range value, the secondary direction range value and the short direction range value.

6. The method according to claim 3, characterized in that The step of training the initial theoretical variogram parameter prediction model according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model includes: Training the classification sub-model according to the variogram surface to be trained and the corresponding variogram type to be trained; The regression sub-model corresponding to the variogram type to be trained is trained according to the variogram surface to be trained and the corresponding variogram parameters to be trained to obtain a trained theoretical variogram parameter prediction model.

7. The method according to claim 2, characterized in that: Predicting the theoretical variogram parameters corresponding to the variogram surface using the theoretical variogram parameter prediction model includes: Mapping the variogram surface into the preset coordinate range to obtain a mapped variogram surface; The mapped variogram surface is input into the theoretical variogram parameter prediction model to obtain the theoretical variogram parameters corresponding to the variogram surface.

8. A variogram automatic modeling device based on deep learning, characterized in that: The device comprises: A construction module is used to construct a variogram surface data set to be trained and to construct an initial theoretical variogram parameter prediction model, wherein the variogram surface data set to be trained includes the variogram surface to be trained and the corresponding variogram parameters to be trained; A training module, used for training the initial theoretical variogram parameter prediction model according to the variogram surface to be trained and the corresponding variogram parameters to be trained, to obtain a trained theoretical variogram parameter prediction model; A prediction module is used to obtain experimental variogram values, calculate the variogram surface corresponding to the experimental variogram values, and use the theoretical variogram parameter prediction model to predict the theoretical variogram parameters corresponding to the variogram surface, so as to obtain a theoretical variogram based on the theoretical variogram parameters.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for automatic modeling of variogram based on deep learning as described in any one of claims 1 to 7 is implemented.

10. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method for automatic modeling of variogram based on deep learning as described in any one of claims 1 to 7 is implemented.

Images