Stratum carbon dioxide saturation evaluation model generation method and system

Abstract

The invention relates to the field of stratum monitoring, and discloses a stratum carbon dioxide saturation evaluation model generation method, which comprises the following steps of: performing mesh generation on a target stratum to obtain a plurality of reservoir intervals so as to establish a stratum model; geological parameters and carbon dioxide migration parameters of different reservoir intervals are obtained, the carbon dioxide saturation of the different reservoir intervals serves as a response variable, and parameter sensitivity analysis is conducted; high-sensitivity parameters corresponding to the carbon dioxide saturation of different reservoir intervals are screened out according to the parameter sensitivity analysis result; and sequentially training the carbon dioxide saturation evaluation models by using the training set by adopting a full convolutional neural network. According to the method, the carbon dioxide saturation evaluation model is obtained based on the high-sensitivity parameters, training of the carbon dioxide saturation evaluation model is carried out for different reservoir areas, and the influence of the influence degree change between different position parameters of the target stratum on the final prediction result is considered; and the accuracy of a prediction result is improved through partition prediction.

Description

Technical Field

[0001] The present invention relates to the field of formation monitoring in the process of carbon dioxide sequestration, and particularly relates to a method and system for generating an evaluation model of formation carbon dioxide saturation. Background Art

[0002] During the process of carbon dioxide sequestration in the formation, it is necessary to comprehensively monitor the carbon dioxide sequestration situation to avoid carbon dioxide leakage and ensure environmental safety. Carbon dioxide sequestration can adopt sequestration methods such as geological sequestration, surface sequestration, and ocean sequestration. Among them, geological carbon dioxide sequestration is to transport and inject high-purity carbon dioxide into sequestration sites such as depleted oil and gas reservoirs, deep saline aquifers, and coal seams through pipeline technology. Since carbon dioxide is sequestered in underground saline aquifers or oil layers several kilometers deep, it is impossible to directly observe the stability and safety of carbon dioxide sequestration. It is necessary to monitor the migration of carbon dioxide plumes in the formation by means of acoustic methods, nuclear measurement methods, electrical methods, etc.

[0003] The prior art patent with the publication number CN118861545A discloses a method and related device for generating an evaluation model of formation carbon dioxide saturation, which adopts a multi-parameter and multi-model fusion method to realize the characterization of carbon dioxide saturation, making the evaluation result of carbon dioxide saturation more accurate and improving the reliability of carbon dioxide monitoring results.

[0004] Although the data characterization method of multi-parameter fusion takes into account the influence of multiple parameters on carbon dioxide saturation, it ignores the influence degree of different parameters in different regions of the formation on carbon dioxide saturation and the mutual influence between carbon dioxide saturations in different regions, resulting in uneven carbon dioxide monitoring results in different regions and low overall accuracy. Summary of the Invention

[0005] Therefore, the present invention provides a method and system for generating an evaluation model of formation carbon dioxide saturation, effectively solving the technical problem that in the prior art, the influence degree of different parameters in different regions of the formation on carbon dioxide saturation and the mutual influence between carbon dioxide saturations in different regions are ignored, resulting in uneven carbon dioxide monitoring results in different regions and low overall accuracy.

[0006] To solve the above technical problems, the present invention specifically provides the following technical solutions: A method for generating an evaluation model of formation carbon dioxide saturation includes the following steps:

[0007] Select a target formation, and perform grid meshing on the target formation in the transverse and longitudinal directions to obtain a number of reservoir intervals to establish a formation model;

[0008] Obtain the geological parameters and carbon dioxide migration parameters of different reservoir intervals, use the carbon dioxide saturation of different reservoir intervals as the response variable, and conduct parameter sensitivity analysis based on the geological parameters and carbon dioxide migration parameters;

[0009] Screen out the highly sensitive parameters corresponding to the carbon dioxide saturation of different reservoir intervals according to the results of the parameter sensitivity analysis;

[0010] Construct a carbon dioxide saturation evaluation model for each reservoir interval, obtain the data of the highly sensitive parameters and carbon dioxide saturation and use them as the training set;

[0011] Use a fully convolutional neural network to train the carbon dioxide saturation evaluation model in turn using the training set to optimize the carbon dioxide saturation evaluation model.

[0012] Furthermore, conduct parameter sensitivity analysis based on the geological parameters and carbon dioxide migration parameters, including the following steps:

[0013] Establish a sensitivity analysis model;

[0014] Decompose the total variance of the sensitivity analysis model into the variance of a single parameter acting alone and the variance of multiple parameters acting simultaneously according to the following variance calculation formula:

[0015]

[0016] In the formula, D is the total variance of the sensitivity analysis model, D i is the variance of the parameter x i acting alone, D i，j is the variance of the parameter x i and x j acting simultaneously, D 1,2，…，n is the variance of n parameters acting simultaneously;

[0017] Normalize the variance calculation formula using the following formula to calculate the sensitivity of a single parameter or the interaction between parameters in the sensitivity analysis model:

[0018]

[0019] In the formula, S i is the first-order sensitivity of a single parameter x i , S i，j is the second-order sensitivity of the interaction between the parameters x i and x j , S 1，...，n is the nth-order sensitivity of the interaction between n parameters;

[0020] Adjust the variance calculation formula to:

[0021]

[0022] Calculate the parameter x using the following formula i for sensitivity:

[0023] S zi = ∑S (i) ;

[0024] wherein, S (i) is the first-order, ……, n-order sensitivity of all parameters containing the parameter x i .

[0025] Further, the geological parameters include porosity, liquid-phase saturation, compressibility, gas entry pressure;

[0026] The carbon dioxide migration parameters include permeability, carbon dioxide saturation.

[0027] Further, calculate the sensitivity of the parameter x i and sort it according to the numerical value;

[0028] Select the first m parameters according to the order of sensitivity and define them as high-sensitivity parameters.

[0029] Further, determine whether the parameter can be tested and predicted, and define the parameter that cannot be obtained through testing and cannot be predicted as an unknown parameter;

[0030] Before constructing the training set, remove the high-sensitivity parameters belonging to the unknown parameters, re-screen the high-sensitivity parameters, and use the data of the newly screened high-sensitivity parameters as the training set and input it into the corresponding carbon dioxide saturation evaluation model.

[0031] Further, start from the boundary position of the target formation and train the carbon dioxide saturation evaluation model for different reservoir intervals in turn.

[0032] Further, number the different reservoir intervals according to the training order;

[0033] Before determining whether the parameter can be predicted, determine whether the high-sensitivity parameter is the same type of parameter as the output result of the carbon dioxide saturation evaluation model;

[0034] If so, determine whether the numbering order of the current reservoir interval is prior to the numbering order of the reservoir interval corresponding to the high-sensitivity parameter;

[0035] If the numbering order of the current reservoir interval is prior to the numbering order of the reservoir interval corresponding to the high-sensitivity parameter, it is determined as non-predictable. If the numbering order of the current reservoir interval is subsequent to the numbering order of the reservoir interval corresponding to the high-sensitivity parameter, it is determined as predictable.

[0036] Further, a fully convolutional neural network is used to train the carbon dioxide saturation evaluation model using a training set, including the following steps:

[0037] Use a convolutional layer to convolve and add the input data, and obtain the output feature map through an activation function;

[0038] Use a pooling layer to screen and filter the feature map output by the convolutional layer to remove redundant features in the feature map;

[0039] Use a deconvolutional layer to enlarge the pooled feature map;

[0040] Use a skip structure to combine the feature map before enlargement with the enlarged feature map to obtain the final prediction result.

[0041] To solve the above technical problems, the present invention further provides the following technical solution: A formation carbon dioxide saturation evaluation model generation system, including:

[0042] A parameter acquisition module, which acquires geological parameters and carbon dioxide migration parameters of different reservoir intervals;

[0043] A sensitivity analysis module, which performs parameter sensitivity analysis based on geological parameters and carbon dioxide migration parameters, and screens out highly sensitive parameters corresponding to the carbon dioxide saturation of different reservoir intervals according to the results of parameter sensitivity analysis;

[0044] A training module, which constructs a carbon dioxide saturation evaluation model corresponding to each reservoir interval, acquires data of highly sensitive parameters and carbon dioxide saturation and uses them as a training set, and uses a fully convolutional neural network to train the carbon dioxide saturation evaluation model in turn using the training set to generate the final carbon dioxide saturation evaluation model.

[0045] Further, acquire partial data of highly sensitive parameters and carbon dioxide saturation as a test set, and use the test set to substitute into the final carbon dioxide saturation evaluation model to test the accuracy of the carbon dioxide saturation evaluation model.

[0046] The present invention has the following beneficial effects compared with the prior art:

[0047] In the present invention, the target formation is first meshed into different reservoir intervals, parameter sensitivity analysis is performed on geological parameters and carbon dioxide migration parameters between different reservoir intervals, highly sensitive parameters are screened out, a carbon dioxide saturation evaluation model is obtained based on the highly sensitive parameters, and the carbon dioxide saturation evaluation model is trained for different reservoir areas respectively. Considering the influence of the change in the influence degree between parameters at different positions of the target formation on the final prediction result, the accuracy of the prediction result is improved by zonal prediction;

[0048] Furthermore, through the sensitivity analysis of parameters between different reservoir intervals, the influencing factors between different reservoir intervals and the same type of parameters are considered, and the accuracy of the prediction structure is improved by associating the data relationships of the same type of parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and those of ordinary skill in the art can also obtain other implementation drawings according to the provided drawings without creative efforts.

[0050] Figure 1 It is a flowchart of a method for generating a formation carbon dioxide saturation evaluation model provided by an embodiment of the present invention;

[0051] Figure 2 It is a structural block diagram of a system for generating a formation carbon dioxide saturation evaluation model provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0053] As Figure 1 shown, the present invention provides a method for generating a formation carbon dioxide saturation evaluation model, including the following steps:

[0054] Select a target formation, and perform grid meshing on the target formation in the horizontal and vertical directions to obtain a number of reservoir intervals, so as to establish a formation model;

[0055] Obtain the geological parameters and carbon dioxide migration parameters of different reservoir intervals, use the carbon dioxide saturation of different reservoir intervals as the response variable, and perform parameter sensitivity analysis based on the geological parameters and carbon dioxide migration parameters;

[0056] Screen out the highly sensitive parameters corresponding to the carbon dioxide saturation of different reservoir intervals according to the results of the parameter sensitivity analysis;

[0057] Construct a carbon dioxide saturation evaluation model corresponding to each reservoir interval, obtain the data of the highly sensitive parameters and carbon dioxide saturation, and use them as the training set;

[0058] The full convolutional neural network is used to train the carbon dioxide saturation evaluation model in sequence with the training set to optimize the carbon dioxide saturation evaluation model.

[0059] The reservoir interval obtained through grid meshing is a three-dimensional interval, generally in the shape of a cube.

[0060] Multiple adjacent reservoir intervals can also be aggregated into a reservoir area. If the carbon dioxide saturation evaluation models finally obtained for the reservoir intervals that can form a reservoir area are similar, the reservoir area can also be regarded as an independent individual, and the carbon dioxide saturation evaluation model corresponding to the reservoir area can be obtained through steps such as data acquisition, data screening and analysis, and training.

[0061] In the present invention, the target formation is subjected to grid meshing to obtain reservoir intervals, the parameter sensitivity analysis is carried out on the geological parameters and carbon dioxide migration parameters between different reservoir intervals, the highly sensitive parameters are screened out, the carbon dioxide saturation evaluation model is obtained based on the highly sensitive parameters, and the carbon dioxide saturation evaluation model is trained for different reservoir areas respectively. Considering the influence of the change of the influence degree between the parameters at different positions of the target formation on the final prediction result, the accuracy of the prediction result is improved by regional prediction.

[0062] In addition, through the parameter sensitivity analysis between different reservoir intervals, the influence factors between different reservoir intervals and the same type of parameters are considered, the data relationship association of the same type of parameters is realized, and the accuracy of the prediction structure is improved.

[0063] In the present invention, first, the geological parameters and carbon dioxide migration parameters of different reservoir intervals need to be obtained. Among them, the geological parameters include but are not limited to porosity, liquid phase saturation, compressibility, gas entry pressure, and the carbon dioxide migration parameters include but are not limited to permeability, carbon dioxide saturation.

[0064] In practical applications, the geological parameters may also include all parameters that may affect the carbon dioxide migration process, such as conductivity, polarizability, etc., and the carbon dioxide migration parameters may also include all parameters that may affect the carbon dioxide migration process, such as migration velocity, etc.

[0065] The obtained geological parameters and carbon dioxide migration parameters generally correspond to each reservoir interval. After obtaining the geological parameters and carbon dioxide migration parameters of each reservoir interval, the sensitivity analysis needs to be carried out on the geological parameters and carbon dioxide migration parameters of each reservoir interval.

[0066] Generally, the sensitivity analysis is carried out on different reservoir intervals in sequence. For example, in the first reservoir interval, the carbon dioxide saturation of different reservoir intervals is used as the response variable, and the sensitivity analysis is carried out on the geological parameters and carbon dioxide migration parameters of this reservoir interval and the adjacent reservoir intervals.

[0067] Among them, a sensitivity analysis of the carbon dioxide saturation in the nearby reservoir interval is included. That is to say, during the sensitivity analysis of this reservoir interval, the relationship between the carbon dioxide saturation in this reservoir interval and the carbon dioxide saturation in other reservoir intervals needs to be analyzed.

[0068] Generally, only the geological parameters and carbon dioxide migration parameters in the nearby reservoir area are sensitively correlated with the carbon dioxide saturation in this reservoir area. Therefore, during the actual sensitivity analysis process, only several of the nearest reservoir intervals can be selected, and their sensitivity analysis with the carbon dioxide saturation in the current reservoir area can be carried out.

[0069] Among them, if during the sensitivity analysis of the carbon dioxide saturation in a reservoir area, a certain parameter in the nearby reservoir area is determined to be a highly sensitive parameter, it is defined as: the geological parameters and carbon dioxide migration parameters in the nearby reservoir area are sensitively correlated with the carbon dioxide saturation in this reservoir area.

[0070] Specifically, based on the geological parameters and carbon dioxide migration parameters, the parameter sensitivity analysis includes the following steps:

[0071] Establish a sensitivity analysis model;

[0072] Decompose the total variance of the sensitivity analysis model into the variance of a single parameter acting alone and the variance of multiple parameters acting simultaneously according to the following variance calculation formula:

[0073]

[0074] In the formula, D is the total variance of the sensitivity analysis model, D i is the variance of the parameter x i acting alone, D i，j is the variance of the parameters x i and x j acting simultaneously, D 1,2，…，n is the variance of n parameters acting simultaneously;

[0075] Normalize the variance calculation formula using the following formula to calculate the sensitivity of a single parameter or the interaction between parameters in the sensitivity analysis model:

[0076]

[0077] In the formula, S i is the first-order sensitivity of the single parameter x i , S i，j is the second-order sensitivity of the interaction between the parameters x i and x j , S 1，...，nis the nth-order sensitivity of the interaction between n parameters;

[0078] Adjust the variance calculation formula to:

[0079]

[0080] Calculate the parameter x using the following formula i The sensitivity S zi :

[0081] S zi =∑S (i) ;

[0082] In the formula, S (i) For all the parameters x i The 1st, ..., nth order sensitivity of .

[0083] Through the above steps, the sensitivity of a parameter is calculated to calculate the parameter x i The sensitivities are sorted according to the numerical values, and then the first m parameters are selected in the order of sensitivity and defined as high-sensitivity parameters.

[0084] In practical applications, the value of m can be adjusted according to actual conditions to obtain the most accurate prediction results of the carbon dioxide saturation evaluation model.

[0085] The above provides a method for screening out highly sensitive parameters. In addition, a certain sensitivity threshold may be set to define parameters above the sensitivity threshold as highly sensitive parameters. The sensitivity threshold may also be adjusted according to actual conditions.

[0086] After obtaining the highly sensitive parameters, it is necessary to determine whether the highly sensitive parameters can be used in the carbon dioxide saturation evaluation model. Specifically, it is necessary to determine whether the parameters can be tested and predicted. The parameters that cannot be obtained through testing and cannot be predicted are defined as unknown parameters.

[0087] Before constructing the training set, the highly sensitive parameters that are unknown parameters are eliminated, and the highly sensitive parameters are re-screened, and the data of the newly screened highly sensitive parameters are input into the corresponding carbon dioxide saturation evaluation model as the training set.

[0088] In the above embodiment, after obtaining the highly sensitive parameters, it is first determined whether the highly sensitive parameters can be tested by traditional testing equipment and testing methods, and then it is determined whether the highly sensitive parameters can be predicted. The parameters that cannot be obtained by testing and cannot be predicted are defined as unknown parameters. If there are unknown parameters in the highly sensitive parameters, it is necessary to re-screen the highly sensitive parameters after eliminating the unknown parameters.

[0089] Since there may be sensitive correlations between the parameters of different reservoir intervals, the parameters of other reservoir intervals can also be used as training data for the carbon dioxide saturation evaluation model of the current reservoir area. To facilitate the correlation of the carbon dioxide saturation evaluation models between different reservoir intervals, the carbon dioxide saturation evaluation models are trained for different reservoir intervals sequentially starting from the boundary position of the target formation.

[0090] Taking the training process of the carbon dioxide saturation evaluation models for the 1st and 2nd reservoir intervals as an example, after the carbon dioxide saturation evaluation model for the 1st reservoir interval is trained, the carbon dioxide saturation corresponding to the first reservoir interval can be predicted using the carbon dioxide saturation evaluation model for the 1st reservoir interval. Before training the carbon dioxide saturation evaluation model for the 2nd reservoir interval, the carbon dioxide saturation of the 1st reservoir interval has been used as a predictable parameter. If it is also a highly sensitive parameter, it is used as a training set and input into the carbon dioxide saturation evaluation model for the 2nd reservoir interval for training, and so on.

[0091] Number the different reservoir intervals according to the training order;

[0092] Before determining whether a parameter can be predicted, determine whether the highly sensitive parameter is of the same type as the output result of the carbon dioxide saturation evaluation model;

[0093] If so, determine whether the numbering order of the current reservoir interval is prior to the numbering order of the reservoir interval corresponding to the highly sensitive parameter;

[0094] If the numbering order of the current reservoir interval is prior to the numbering order of the reservoir interval corresponding to the highly sensitive parameter, it is determined as non-predictable. If the numbering order of the current reservoir interval is posterior to the numbering order of the reservoir interval corresponding to the highly sensitive parameter, it is determined as predictable.

[0095] Specifically, assuming that the reservoir interval numbers are 1, 2, 3, ……, after obtaining the highly sensitive parameter, first determine whether the highly sensitive parameter can be measured by a test device. If it can, then determine whether the highly sensitive parameter is of the same type as the output result of the carbon dioxide saturation evaluation model, that is: determine whether the highly sensitive parameter is carbon dioxide saturation;

[0096] If it is, assume that the numbering order of the current reservoir interval is 6 and the numbering order of the reservoir interval corresponding to the highly sensitive parameter is 8. That is to say, there is a sensitive correlation between the carbon dioxide saturation of the 6th reservoir interval and the carbon dioxide saturation of the 8th. However, during the actual process of predicting the carbon dioxide saturation of the 6th reservoir interval, the carbon dioxide saturation of the 8th reservoir interval cannot be predicted in advance. Therefore, it is determined as non-predictable. For this situation, the highly sensitive parameter is defined as an unknown parameter;

[0097] Assume that the numbering order of the current reservoir interval is 8, and the numbering order of the reservoir interval corresponding to the highly sensitive parameter is also 8. That is to say, there is a sensitive correlation between the carbon dioxide saturation of the 8th reservoir interval and the 6th carbon dioxide saturation. During the actual prediction of the carbon dioxide saturation in the 8th reservoir interval, the carbon dioxide saturation of the 6th reservoir interval has been predicted in advance. The carbon dioxide saturation of the 6th reservoir interval can be used as the training set and input into the carbon dioxide saturation evaluation model of the 8th reservoir interval. Therefore, it is determined that it can be predicted.

[0098] In the specific implementation process, the parameter sensitivity analysis is carried out on different reservoir intervals in sequence to obtain the highly sensitive parameters. After obtaining the highly sensitive parameters, first judge whether the highly sensitive parameter can be measured by the test equipment. If it can, then judge whether the highly sensitive parameter is predictable. If it cannot be predicted, it is defined as an unknown parameter and removed from the highly sensitive parameters. Rescreen the highly sensitive parameters, and use the data of the highly sensitive parameters as the training set and input it into the carbon dioxide saturation evaluation model corresponding to different reservoir intervals, so as to obtain the carbon dioxide saturation evaluation model corresponding to the final reservoir area.

[0099] In the present invention, a convolutional neural network is used to train the carbon dioxide saturation evaluation model with the training set, including the following steps:

[0100] Use the convolutional layer to convolve and add the input data, and obtain the output feature map through the activation function;

[0101] Use the pooling layer to screen and filter the feature map output by the convolutional layer to remove the redundant features in the feature map;

[0102] Use the deconvolutional layer to enlarge the pooled feature map;

[0103] Use the skip structure to combine the feature map before enlargement and the enlarged feature map to obtain the final prediction result.

[0104] Specifically, the convolutional layer in the fully convolutional neural network is composed of several convolutional kernels. The parameters of the convolutional kernel include the convolutional kernel size, stride, and padding, which are calculated according to the backpropagation of the network. The convolutional kernel size can be set to any value smaller than the size of the input data image. The convolutional kernel convolves with the feature matrix input by the previous layer to obtain a feature map with a smaller size.

[0105] The pooling layer can be set to multiple. The pooling layer screens and filters the feature map output by the convolutional layer to remove the redundant features in the feature map and avoid the overfitting phenomenon of the network. The pooling layer is controlled by three parameters: pooling size, stride, and padding. The maximum pooling method can be used to extract the local maximum value of the features.

[0106] In the deconvolution layer, the convolution kernel parameters are exactly the same as those in the convolution layer. The pooled feature map is enlarged to complete the interpolation process. The specific steps are as follows: multiply the input feature map by the deconvolution kernel, combine the intermediate matrix with the step sizes along the horizontal and vertical axes, add the values in the pixel overlapping regions to extract the extended matrix of the input features, add biases to supplement the detailed information, and crop a new pixel matrix according to the target image size.

[0107] The deconvolution layer can map the pooled feature map into an image with the same size as the original feature map.

[0108] A skip structure is introduced in the fully convolutional neural network, that is, the prediction results of the last layer are combined with those of some previous layers to complete the final prediction of the network.

[0109] Assume that after each pooling operation in the network, the image is reduced to 1 / 2 of its original size. The network has 3 pooling layers, and after image pooling, it becomes 1 / 8 of its original size. If upsampling is directly performed at this time, a lot of detailed information will be lost. If the results of the previous several layers are also upsampled and then superimposed on the finally output feature map, the prediction accuracy of the network can be improved.

[0110] As Figure 2 shown, the present invention also provides a formation carbon dioxide saturation evaluation model generation system, including:

[0111] A parameter acquisition module, which acquires geological parameters and carbon dioxide migration parameters for different reservoir intervals;

[0112] A sensitivity analysis module, which performs parameter sensitivity analysis based on the geological parameters and carbon dioxide migration parameters, and screens out the highly sensitive parameters corresponding to the carbon dioxide saturation in different reservoir intervals according to the results of the parameter sensitivity analysis;

[0113] A training module, which constructs a carbon dioxide saturation evaluation model corresponding to each reservoir interval, acquires the data of the highly sensitive parameters and carbon dioxide saturation and uses them as a training set, and uses a fully convolutional neural network to train the carbon dioxide saturation evaluation model in turn with the training set to generate the final carbon dioxide saturation evaluation model.

[0114] The data of the acquired highly sensitive parameters and carbon dioxide saturation are all sample data. A part of the sample data is used as the training set, and another part is used as the test set. The training set is used to train the carbon dioxide saturation evaluation model, and the test set is used to substitute into the final carbon dioxide saturation evaluation model to test the accuracy of the carbon dioxide saturation evaluation model.

[0115] Therefore, in the present invention, multiple parameters are essentially considered comprehensively, and sensitivity analysis is performed on the parameters in actual evaluation. High-sensitivity parameters are preferentially considered for modeling, and then low-sensitivity parameters are considered. Based on this method, on the one hand, the amount of calculation in the early stage can be reduced, and on the other hand, the collaborative calculation of low-sensitivity parameters can be carried out based on high-sensitivity parameters, ultimately improving the accuracy of the evaluation model.

[0116] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the present application within the essence and protection scope of the present application, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present application.

Claims

1. A method for generating a formation carbon dioxide saturation evaluation model, characterized in that, The following steps are involved: Select a target stratum, and divide the target stratum into grids in the horizontal and vertical directions to obtain a number of reservoir intervals to establish a stratum model; Obtain geological parameters and CO2 migration parameters of different reservoir intervals, use CO2 saturation of different reservoir intervals as response variables, and perform parameter sensitivity analysis based on geological parameters and CO2 migration parameters; According to the results of parameter sensitivity analysis, highly sensitive parameters corresponding to carbon dioxide saturation in different reservoir intervals are screened out; Construct a CO2 saturation evaluation model corresponding to each reservoir interval, obtain data on highly sensitive parameters and CO2 saturation and use them as training sets; The fully convolutional neural network is used to train the carbon dioxide saturation evaluation model in sequence using the training set to optimize the carbon dioxide saturation evaluation model.

2. The method for generating the formation carbon dioxide saturation evaluation model according to claim 1, wherein Parameter sensitivity analysis based on geological parameters and carbon dioxide migration parameters includes the following steps: Establish sensitivity analysis model; The total variance of the sensitivity analysis model is decomposed into the variance of a single parameter and the variance of multiple parameters acting simultaneously according to the following variance calculation formula: where D is the total variance of the sensitivity analysis model, D i is the variance of the parameter x i acting alone, D i，j is the parameter x i and x j acting simultaneously, D 1,2，…，n is the variance of n parameters acting simultaneously; The variance calculation formula is normalized using the following formula to calculate the sensitivity of a single parameter or the interaction between parameters in the sensitivity analysis model: Where, S i is the first-order sensitivity of a single parameter x i , S i，j is the second-order sensitivity of the interaction between parameter x i and x j , and S 1，...，n is the nth-order sensitivity of the interaction among n parameters; Adjust the variance calculation formula to: Calculate the parameter x using the following formula i for the sensitivity S zi as follows S zi = ∑S (i) ; where S (i) is the first-order, ……, nth-order sensitivities of all those containing the parameter x i .

3. The method for generating a formation carbon dioxide saturation evaluation model according to claim 1, wherein, The geological parameters include porosity, liquid saturation, compressibility, and gas entry pressure; The carbon dioxide migration parameters include permeability and carbon dioxide saturation.

4. The method for generating a formation carbon dioxide saturation evaluation model according to claim 2, wherein, Calculate the parameter x i for sensitivity and sort them according to the numerical values; The first m parameters are selected in order of sensitivity and defined as highly sensitive parameters.

5. The method for generating the formation carbon dioxide saturation evaluation model according to claim 4, wherein Determine whether the parameters can be tested and predicted, and define the parameters that cannot be tested and predicted as unknown parameters; Before constructing the training set, the highly sensitive parameters that are unknown parameters are eliminated, and the highly sensitive parameters are re-screened, and the data of the newly screened highly sensitive parameters are input into the corresponding carbon dioxide saturation evaluation model as the training set.

6. The method for generating the formation carbon dioxide saturation evaluation model according to claim 5, wherein, Starting from the boundary position of the target formation, the carbon dioxide saturation evaluation model is trained for different reservoir intervals in turn.

7. The method for generating a formation carbon dioxide saturation evaluation model according to claim 6, wherein Number the different reservoir intervals according to the training sequence; Before determining whether a parameter can be predicted, determine whether the highly sensitive parameter is the same type of parameter as the output result of the carbon dioxide saturation assessment model; If so, determine whether the numbering sequence of the current reservoir interval precedes the numbering sequence of the reservoir interval corresponding to the high-sensitivity parameter; If the numbering sequence of the current reservoir interval is earlier than the numbering sequence of the reservoir interval corresponding to the high-sensitivity parameter, it is judged to be unpredictable. If the numbering sequence of the current reservoir interval is later than the numbering sequence of the reservoir interval corresponding to the high-sensitivity parameter, it is judged to be predictable.

8. The method for generating a formation carbon dioxide saturation evaluation model according to claim 1, wherein The carbon dioxide saturation evaluation model is trained using a fully convolutional neural network using a training set, including the following steps: The input data is convolved and added using the convolution layer, and the output feature map is obtained through the activation function; The pooling layer is used to screen and filter the feature map output by the convolution layer to remove redundant features in the feature map; Use the deconvolution layer to enlarge the pooled feature map; The jump structure is used to combine the feature map before and after amplification to obtain the final prediction result.

9. A system using the method for generating a formation carbon dioxide saturation evaluation model according to any one of claims 1 to 8, characterized in that, include: A parameter acquisition module that acquires geological parameters and carbon dioxide migration parameters for different reservoir intervals; A sensitivity analysis module that conducts parameter sensitivity analysis based on the geological parameters and carbon dioxide migration parameters, and screens out highly sensitive parameters corresponding to the carbon dioxide saturation of different reservoir intervals according to the results of the parameter sensitivity analysis; A training module that constructs a carbon dioxide saturation evaluation model corresponding to each reservoir interval, obtains the data of the highly sensitive parameters and carbon dioxide saturation and uses them as a training set, and uses a fully convolutional neural network to train the carbon dioxide saturation evaluation model in turn using the training set to generate the final carbon dioxide saturation evaluation model.

10. The formation carbon dioxide saturation evaluation model generation system according to claim 9, characterized in that Obtain partial data of the highly sensitive parameters and carbon dioxide saturation as a test set, and use the test set to substitute into the final carbon dioxide saturation evaluation model to test the accuracy of the carbon dioxide saturation evaluation model.

Images