Reservoir modeling method and device based on automatic calculation of Kriging variation function

Abstract

The invention relates to the technical field of oil field reservoir geological modeling, in particular to a reservoir modeling method and device based on automatic calculation of a Kriging variation function. The method comprises the following steps: acquiring work area logging data and corresponding hard data thereof, and seismic data and corresponding soft data thereof, and performing grid division on the hard data and the soft data; dividing the data into training data and verification data; inputting the training data into an established reservoir geologic model based on a Kriging variation function, and training the model; and if the error between the trained model and the well point corresponding part of the verification data is greater than a first predetermined percentage, optimizing the trained model by using an optimization algorithm. The problems that an existing reservoir geologic model establishing method seriously depends on prior cognition of geologic experts, and the subjectivity of an interpretation result is high are solved; the problems that the uncertainty of model feature distribution performance is too large and reliability evaluation is difficult due to the fact that a trained model lacks effective verification are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of oilfield reservoir geological modeling, and particularly to a reservoir modeling method and device based on automatic calculation of Kriging variogram. Background Art

[0002] In the process of oil and gas exploration and development, with the increase in the amount of drilling data, petroleum geological researchers mainly conduct geological modeling research manually through conventional geological modeling software. The geological model update speed is slow and the cycle is long, resulting in low collaborative efficiency of geological evaluation and engineering design. Conventional geological modeling relies on the empirical knowledge of geological experts and lacks effective correction of the model after quality control, resulting in low prediction accuracy of geological and engineering parameter space models.

[0003] Currently, there is a reservoir geological model established by machine learning methods, which can effectively improve the update efficiency and prediction accuracy of the space model. The model includes deterministic modeling and stochastic modeling. Deterministic modeling starts from known data and uses deterministic algorithms to establish a unique and deterministic reservoir prediction model for grid nodes in the unknown area between wells. Simulating the distribution of underground strongly heterogeneous reservoir properties with deterministic information will inevitably increase the risk of formation exploration. Stochastic modeling is based on stochastic theory and conducts stochastic simulation on logging data through the Kriging function to obtain multiple stochastic reservoir structure models, reflecting the complexity and uncertainty of the underground reservoir.

[0004] When establishing the current reservoir geological model, all prior data are used as the training basis for the model. The reservoir geological model constructed by this scheme is highly subjective and lacks effective verification of actual well data. Summary of the Invention

[0005] The present invention provides a reservoir modeling method and device based on automatic calculation of Kriging variogram to solve the problems that the existing reservoir geological model establishment method seriously relies on the prior knowledge of geological experts, the interpretation result is highly subjective; the trained model lacks effective verification, resulting in excessive uncertainty in the model feature distribution and difficult reliability evaluation.

[0006] According to one aspect of the present invention, there is provided a reservoir modeling method based on automatic calculation of Kriging variogram, including:

[0007] Obtain the logging data of the work area and its corresponding formation parameter interpretation results, as well as seismic data and its corresponding formation parameter prediction results. Among them, the formation parameter interpretation results are hard data, and the formation parameter prediction results are soft data, and grid division is performed on the hard data and soft data;

[0008] Divide the hard data, soft data, corresponding logging data, and seismic data after grid division into training data and validation data;

[0009] Input the training data into the established reservoir geological model based on the Kriging variogram and train the model;

[0010] Determine the error between the trained model and the well point corresponding part of the validation data;

[0011] Judge whether the error is greater than the first predetermined percentage. If so, use an optimization algorithm to optimize the trained model to obtain the final reservoir geological model.

[0012] Preferably, the method of dividing the hard data, soft data, corresponding logging data, and seismic data after grid division into training data and validation data includes:

[0013] Use all the soft data after grid division, their corresponding seismic data, and the hard data and their corresponding logging data of the second predetermined percentage as training data;

[0014] Use the other hard data and their corresponding logging data except the second predetermined percentage as validation data.

[0015] Preferably, the reservoir geological model is a model established by the Kriging two-point geostatistical algorithm.

[0016] Preferably, the method of determining the error between the trained model and the well point corresponding part of the validation data includes:

[0017] Use an error loss function to determine the error between the trained model and the well point corresponding part of the validation data;

[0018] Among them, the error loss function is:

[0019]

[0020]

[0021] In the formula: respectively represent the model data corresponding to the well location coordinates of the training data H 1 , validation data H 2 , and λ is the weight parameter of the validation loss.

[0022] Preferably, the optimization algorithm is: Hyperopt algorithm.

[0023] Preferably, the method of using an optimization algorithm to optimize the trained model to obtain the final reservoir geological model includes:

[0024] Given the hyperparameter space and the number of iterations of the reservoir geological model based on the Kriging variogram;

[0025] Model the optimization objective function through the kernel density estimation surrogate model of the Hyperopt algorithm;

[0026] Use the acquisition function of the Hyperopt algorithm to find the next optimal parameter evaluation point of the model;

[0027] Continuously update the surrogate function of the Hyperopt algorithm through the optimal parameter evaluation point to approximate the objective function, thereby determining the optimal hyperparameters of the model.

[0028] According to an aspect of the present invention, there is provided a reservoir modeling device for automatically obtaining based on the Kriging variogram, including:

[0029] An acquisition unit for acquiring well logging data in the work area and its corresponding formation parameter interpretation results, as well as seismic data and its corresponding formation parameter prediction results, wherein the formation parameter interpretation results are hard data, the formation parameter prediction results are soft data, and grid division is performed on the hard data and soft data;

[0030] A data division unit for dividing the hard data and soft data after grid division and the corresponding well logging data and seismic data into training data and validation data;

[0031] A model training unit for inputting the training data into the established reservoir geological model based on the Kriging variogram and training the model;

[0032] An error determination unit for determining the error between the trained model and the well point corresponding part of the validation data;

[0033] A model optimization unit for determining whether the error is greater than a first predetermined percentage. If so, use an optimization algorithm to optimize the trained model to obtain a final reservoir geological model.

[0034] The present invention has at least the following beneficial effects:

[0035] The present invention proposes a reservoir modeling method and device for automatically obtaining based on the Kriging variogram. By dividing well logging data, seismic data, hard data, and soft data into two parts, they are respectively used for training and validating the trained model. The model is updated through the Kriging variogram, and training is repeatedly optimized to minimize the error of the model, thereby effectively improving the accuracy of the reservoir geological model. Description of the Drawings

[0036] The accompanying drawings here are incorporated into the specification and form a part of this specification. These drawings show embodiments consistent with the present invention and, together with the specification, are used to illustrate the technical solutions of the present invention.

[0037] Figure 1 A flowchart showing a reservoir modeling method automatically obtained based on the Kriging variogram according to an embodiment of the present invention;

[0038] Figure 2 A schematic diagram showing sample construction according to an embodiment of the present invention;

[0039] Figure 3 A schematic diagram showing the principle of reservoir seismic modeling according to an embodiment of the present invention;

[0040] Figure 4 A reconstructed result diagram of a reservoir seismic image output from a model according to an embodiment of the present invention. Detailed implementation manners

[0041] Various exemplary embodiments, features, and aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0042] The term "exemplary" used herein means "serving as an example, embodiment, or illustration". Any embodiment described as "exemplary" here does not have to be construed as superior to or better than other embodiments.

[0043] The term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the term "at least one" in this article means any one of a plurality or any combination of at least two of a plurality. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set composed of A, B, and C.

[0044] In addition, to better illustrate the present invention, numerous specific details are given in the following detailed implementation manners. Those skilled in the art should understand that the present invention can also be implemented without some specific details. In some instances, methods, means, elements, and circuits well known to those skilled in the art are not described in detail so as to highlight the gist of the present invention.

[0045] Figure 1 A flowchart showing a reservoir modeling method automatically obtained based on the Kriging variogram according to an embodiment of the present invention; Figure 2 A schematic diagram showing the construction of training samples according to an embodiment of the present invention; Figure 3Schematic diagram showing the reservoir seismic modeling principle according to an embodiment of the present invention; Figure 4 Diagram showing the reconstructed result of the reservoir seismic image output from the model according to an embodiment of the present invention. As Figures 1-4 shown, a reservoir modeling method for automatically obtaining based on the Kriging variogram includes: Step S01: Obtain the well logging data of the work area and its corresponding formation parameter interpretation results, as well as the seismic data and its corresponding formation parameter prediction results. Among them, the formation parameter interpretation result is hard data, and the formation parameter prediction result is soft data. Perform grid division on the hard data and soft data; Step S02: Divide the grid-divided hard data and soft data and their corresponding well logging data and seismic data into training data and verification data; Step S03: Input the training data into the established reservoir geological model based on the Kriging variogram and train the model; Step S04: Determine the error between the trained model and the well point corresponding part of the verification data; Step S05: Judge whether the error is greater than the first predetermined percentage. If so, use an optimization algorithm to optimize the trained model to obtain the final reservoir geological model.

[0046] The reservoir modeling method for automatically obtaining based on the Kriging variogram provided by the embodiment of the present invention specifically includes the following steps:

[0047] Step S01: Obtain the well logging data of the work area and its corresponding formation parameter interpretation results, as well as the seismic data and its corresponding formation parameter prediction results. Among them, the formation parameter interpretation result is hard data, and the formation parameter prediction result is soft data. Perform grid division on the hard data and soft data.

[0048] In the embodiment of the present invention, the well logging data of the work area is interpreted to obtain the formation parameter interpretation result, and this formation parameter interpretation result is hard data; according to the seismic attributes in the seismic data, the formation parameters are predicted to obtain the formation parameter prediction result, and this formation parameter prediction result is soft data.

[0049] Among them, the hard data H obtained through well logging interpretation is: Through conventional well logging curves such as natural gamma (GR), caliper (CAL), spontaneous potential (SP), acoustic time difference (AC), density (DEN), compensated neutron (CNL), resistivity logging (R s 、R d 、R xo ), etc., well logging data interpretation can obtain the required parameters such as the porosity, permeability, oil saturation and shale content of the formation, and use the interpretation result as the hard data on the well for model training.

[0050] Among them, the soft data S on the plane obtained through seismic attribute interpretation is as follows: Seismic attribute technology can be divided into four categories: time attributes, amplitude attributes, frequency attributes, and absorption attenuation attributes. Seismic attributes can be used to predict parameters such as reservoir porosity, permeability, saturation, and shale content, and the prediction results are used as the soft data for model training.

[0051] Grid the hard data H and soft data S according to the well location coordinates in the work area, and establish a spatial layer data model.

[0052] Step S02: Divide the gridded hard data, soft data, and their corresponding logging data and seismic data into training data and validation data.

[0053] In the present invention, the method of dividing the gridded hard data, soft data, and their corresponding logging data and seismic data into training data and validation data includes: using all the gridded soft data, their corresponding seismic data, and a second predetermined percentage of the hard data and their corresponding logging data as training data; using the other hard data and their corresponding logging data except for the second predetermined percentage as validation data.

[0054] In an embodiment of the present invention, during the model training stage, the logging data (logging data and hard data) of the area to be reconstructed is randomly divided into K parts, and K - 1 parts (the second predetermined percentage), for example, the second predetermined percentage is 80%, and all seismic data and soft data are used as the training data set of the automatic reservoir geological model building algorithm based on the Kriging variogram, denoted as H 1 , and the remaining 1 part of the data, that is, 20%, is used as the validation data set, denoted as H 2 .

[0055] Usually, the training of model data only includes logging data as the main training data, and the accuracy of the trained model is relatively low. In the present invention, by using seismic data and their corresponding soft data as input, the content of model training data can be further enriched, making the accuracy of the trained model higher and the prediction results more accurate. The constructed sample data, that is, training data and validation data, is as Figure 2 shown, where R in Figure 2 is the data ratio of the divided training data set, that is, the second predetermined percentage.

[0056] Step S03: Input the training data into the established reservoir geological model based on the Kriging variogram and train the model.

[0057] In the present invention, the reservoir geological model based on the Kriging variogram is a model established by the Kriging two-point geostatistical algorithm.

[0058] In the embodiments of the present invention, Kriging, also known as the Kriging variogram, is the basis of statistical geological modeling. The two-point geostatistical method based on the variogram solves the spatial linear interpolation problem of geological variables by determining the correlation between two points in space. The two-point geostatistical method includes deterministic models based on methods such as ordinary Kriging, universal Kriging, co-Kriging, and indicator Kriging, and stochastic models based on methods such as Gaussian simulation, truncated Gaussian simulation, co-Gaussian simulation, Markov random field simulation, and fractal random field simulation. Stochastic modeling is based on known data, with random functions as the theory. By applying stochastic simulation algorithms to grid nodes in the unknown area between wells, multiple possible prediction results are given, an optional and equiprobable reservoir prediction model is established, and through uncertainty analysis, an objective evaluation of the risk of oilfield development decisions is carried out.

[0059] According to the selection of the reservoir geological modeling algorithm, that is, the two-point geostatistical algorithm (Kriging), a mapping relationship from the input data to the model output is established:

[0060] Model output =F(ω,H 1 ,S) (1);

[0061] Where: F is the objective function of the three-dimensional reservoir geological model; ω is the model parameter (such as: direction angle, bandwidth, etc.); H 1 is the hard data; S is the soft data.

[0062] The training data is input into the established reservoir geological model for training. During the model training process, a mapping relationship is established between the logging data parameters and seismic data parameters in the training data and reservoir information such as formation porosity, permeability, lithology, oil saturation, and shale content. Among them, the model of reservoir geological modeling (three-dimensional reservoir geological model) is selected as continuous modeling or discrete modeling. Continuous modeling includes ordinary Kriging and co-Gaussian simulation, and discrete modeling includes indicator Kriging and co-indicator simulation.

[0063] Step S04: Determine the error between the corresponding part of the well points of the trained model and the validation data.

[0064] In the present invention, the method for determining the error between the corresponding part of the well points of the trained model and the validation data includes: using an error loss function to determine the error between the corresponding part of the well points of the trained model and the validation data;

[0065] Among them, the error loss function is:

[0066]

[0067]

[0068] Wherein: respectively represent the reservoir geological model data corresponding to the well location coordinates of the well training data H 1 and the verification data H 2 , and λ is the weight parameter of the verification loss.

[0069] In the embodiment of the present invention, error verification is performed on the trained model, that is: the posterior well data, that is, the well logging data and the corresponding predicted formation parameter data included in the verification data are compared. For the soft data in the verification data, that is, the part corresponding to the well points of the formation parameters originally and the formation parameters obtained by running the constructed reservoir geological model is calculated for error, and error loss functions are established as shown in formulas (2) and (3).

[0070] The error of the model is determined through the established loss function. If the error meets the condition, that is, the error is greater than the first predetermined percentage, such as: 3%, it indicates that the model has a poor interpretation effect on the well logging data (well logging data), and then the parameters of the trained model are optimized and updated by using an optimization algorithm.

[0071] Step S05: Determine whether the error is greater than the predetermined percentage. If so, use an optimization algorithm to optimize the trained model to obtain the final reservoir geological model.

[0072] In the present invention, the optimization algorithm is: Hyperopt algorithm.

[0073] In the present invention, the method of using the optimization algorithm to optimize the trained model to obtain the final reservoir geological model includes: giving the hyperparameter space and the number of iterations of the reservoir geological model based on the Kriging variogram; modeling the optimization objective function through the kernel density estimation surrogate model of the Hyperopt algorithm; using the acquisition function of the Hyperopt algorithm to find the next optimal parameter evaluation point of the model; continuously updating the surrogate function of the Hyperopt algorithm through the optimal parameter evaluation point to make it approach the target function, so as to determine the optimal hyperparameters of the model.

[0074] In the embodiment of the present invention, the Hyperopt algorithm is Distributed Asynchronous Hyperparameter Optimization. The Hyperopt algorithm is used to establish an optimization model for the reservoir geological modeling algorithm, automatically search for the parameters with the best two-point correlation between well network data, so that the model after optimizing the variogram fully characterizes the spatial well logging data.

[0075] The specific steps are: (1) Give the initial hyperparameters ω of the reservoir geological modeling model in step S031 and the number of iterations n. (2) Through the kernel density estimation surrogate function of the Hyperopt algorithm, a model is constructed for the optimization objective function (Error loss ), that is, a surrogate model. (3) Use the acquisition function of the Hyperopt algorithm to find the optimal parameter ω of the next reservoir geological modeling model 2 corresponding evaluation point. (4) Continuously update the surrogate function of the Hyperopt algorithm through the parameter evaluation point, so as to approximate the objective function (Error loss ), that is, the target value, to find the optimal hyperparameter ω * .

[0076] Automatically adjust the parameters of the trained reservoir geological model according to the optimal hyperparameters obtained by the optimization algorithm, retrain, and verify until the error meets the conditions, so as to obtain the optimal reservoir geological model, that is, the final reservoir geological model, and use this final reservoir geological model to conduct geological research on the work area. The model training, verification and optimization process is as Figure 3 shown.

[0077] In the embodiment of the present invention, the data of a certain well is selected as an example for reservoir geological modeling. According to step S01, the conditional information is made from the multi-well core calibration conventional logging data and seismic data, and the data is grid-divided, that is, the well position coordinates in the x, y, and z spaces and the corresponding porosity, permeability, oil saturation or shale content values are spatially fused, that is, the well data is assigned to the corresponding grid points; then the well data is processed according to step S02, that is, divided into a training data set H 1 and a validation data set H 2 ; a reservoir geological model based on the Kriging variogram is constructed through step S03 and the model is trained by the training set; according to step S04, the constructed reservoir geological model is error-verified using the validation data; finally, according to step S05, the optimization algorithm is used to automatically optimize and update the parameters of the variogram reservoir seismic model to obtain the final reservoir geological model.

[0078] Finally, input the actual well data to be modeled into the optimized final reservoir geological model, and the reservoir geological image result can be obtained, as Figure 4 shown. Figure (a) is the constructed sample data, and Figure (b) is the reservoir geological image finally obtained according to the model. From Figure 4 it can be seen that the method of the present invention can better generate the reservoir geological model for the conditional well data, and can establish the reservoir geological model more accurately.

[0079] It can be understood that the above-mentioned various method embodiments mentioned in the present invention can be combined with each other to form a combined embodiment without violating the principle logic. Due to space limitations, the present invention will not be elaborated further.

[0080] The execution subject of the reservoir modeling method automatically obtained based on the Kriging variogram can be a reservoir modeling device automatically obtained based on the Kriging variogram. For example, the reservoir modeling method automatically obtained based on the Kriging variogram can be executed by a terminal device, a server, or other processing devices. Among them, the terminal device can be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementation manners, the reservoir modeling method automatically obtained based on the Kriging variogram can be implemented by a processor calling computer-readable instructions stored in a memory.

[0081] Those skilled in the art can understand that in the above method of the specific implementation manner, the writing order of each step does not mean a strict execution order and constitutes any limitation to the implementation process. The specific execution order of each step should be determined according to its function and possible internal logic.

[0082] The present invention also provides a reservoir modeling device automatically obtained based on the Kriging variogram, including: an acquisition unit, configured to acquire logging data of a work area and its corresponding formation parameter interpretation result, and seismic data and its corresponding formation parameter prediction result, where the formation parameter interpretation result is hard data, the formation parameter prediction result is soft data, and grid division is performed on the hard data and the soft data; a data division unit, configured to divide the hard data and soft data and the corresponding logging data and seismic data after grid division into training data and verification data; a model training unit, configured to input the training data into a established reservoir geological model based on the Kriging variogram and train the model; an error determination unit, configured to determine an error between a well point corresponding part of the trained model and the verification data; a model optimization unit, configured to determine whether the error is greater than a first predetermined percentage. If so, the trained model is optimized by using an optimization algorithm to obtain a final reservoir geological model.

[0083] In some embodiments, the functions or modules included in the device provided by the embodiments of the present invention can be used to execute the methods described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0084] At present, the following problems exist in the establishment of reservoir geological models based on geostatistical methods: (1) Due to the complex and diverse distribution characteristics of geological forms, actual modeling heavily relies on the prior knowledge of geological experts, resulting in strong subjectivity in well interpretation; (2) Regarding the conventional training mechanism for reservoir geological modeling, there is a lack of effective verification of the model by posterior well data, leading to excessive uncertainty in the distribution of model characteristics and difficulty in reliability assessment. Therefore, how to effectively establish a more accurate reservoir geological model based on prior logging information is the key to solving the problem of excessive randomness in reservoir modeling.

[0085] The Kriging variogram automatic modeling technology based on the blind well posterior mechanism can effectively alleviate this problem. The model trained from actual logging data can be tested against the reservoir geological model through posterior well information, that is, verification data. If the reservoir model parameters deviate significantly from the posterior well data, the modeling algorithm parameters can be automatically updated through an optimization algorithm. Therefore, the present invention introduces the Kriging automatic modeling technology based on the blind well posterior mechanism to reasonably and efficiently restore the spatial distribution information of logging data. By repeatedly optimizing and training the model, the posterior error of the model is minimized, effectively improving the accuracy of the reservoir geological model.

[0086] The present invention establishes a reservoir geological model by coupling the spatial structural characteristics of multi-well data, thereby obtaining information such as reservoir physical property parameters and seepage physical characteristics. The reservoir geological model is an important data source for effectively understanding the structural information of formation pores, lithology, etc., and can provide a reliable geological basis for the efficient exploration and development of oil and gas reservoirs.

[0087] The embodiments of the present invention have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other ordinary skill in the art in the technical field to understand the disclosed embodiments.

Claims

1. A reservoir modeling method based on automatic determination of Kriging variogram, characterized in that: include: Acquire the well logging data and the corresponding formation parameter interpretation results of the work area, as well as the seismic data and the corresponding formation parameter prediction results, wherein the formation parameter interpretation results are hard data and the formation parameter prediction results are soft data, and perform grid division on the hard data and the soft data; Dividing the hard data and soft data after the grid division and the corresponding well logging data and seismic data into training data and verification data; Inputting the training data into the established reservoir geological model based on the Kriging variogram, and training the model; Determining the error between the trained model and the corresponding portion of the well points of the validation data; It is determined whether the error is greater than a first predetermined percentage. If so, an optimization algorithm is used to optimize the trained model to obtain a final reservoir geological model.

2. The reservoir modeling method based on automatic determination of Kriging variogram according to claim 1, characterized in that: The method of dividing the grid-divided hard data and soft data and their corresponding well logging data and seismic data into training data and verification data includes: Using all the soft data after the gridding and the corresponding seismic data, and the second predetermined percentage of hard data and the corresponding well logging data as training data; The other hard data except the second predetermined percentage and the corresponding logging data are used as verification data.

3. The reservoir modeling method based on automatic determination of Kriging variogram according to claim 1, characterized in that: The reservoir geological model is a model established by a Kriging two-point geostatistical algorithm.

4. The reservoir modeling method based on automatic determination of Kriging variogram according to claim 1, characterized in that: The method for determining the error between the trained model and the well point corresponding part of the verification data comprises: Determine the error between the trained model and the corresponding part of the well points of the validation data using an error loss function; Wherein, the error loss function is: Where: Model output-H1 、Model output-H2 They represent the model data corresponding to the well location coordinates of the training data H1 and the validation data H2 respectively, and λ is the weight parameter of the validation loss.

5. The reservoir modeling method based on automatic determination of Kriging variogram according to any one of claims 1 to 4, characterized in that: The optimization algorithm is: Hyperopt algorithm.

6. The reservoir modeling method based on automatic determination of Kriging variogram according to claim 5, characterized in that: The method of optimizing the trained model by using an optimization algorithm to obtain a final reservoir geological model comprises: The hyperparameter space and the number of iterations of the reservoir geological model based on the Kriging variogram are given; The optimization objective function is modeled through the kernel density estimation surrogate model of the Hyperopt algorithm; Use the acquisition function of the Hyperopt algorithm to find the next optimal parameter evaluation point of the model; The proxy function of the Hyperopt algorithm is continuously updated through the optimal parameter evaluation point to make it close to the objective function, thereby determining the optimal hyperparameters of the model.

7. A reservoir modeling device based on automatic determination of Kriging variogram, characterized in that: include: An acquisition unit is used to acquire well logging data and corresponding formation parameter interpretation results of the work area, as well as seismic data and corresponding formation parameter prediction results, wherein the formation parameter interpretation results are hard data, and the formation parameter prediction results are soft data, and the hard data and the soft data are gridded; A data division unit, used for dividing the hard data and soft data after the grid division and the corresponding well logging data and seismic data into training data and verification data; A model training unit, used for inputting the training data into the established reservoir geological model based on the Kriging variogram, and training the model; an error determination unit, used to determine the error between the trained model and the well point corresponding part of the verification data; The model optimization unit is used to determine whether the error is greater than a first predetermined percentage. If so, the trained model is optimized using an optimization algorithm to obtain a final reservoir geological model.

Images