A method, device and related equipment for dynamic characterization of reservoir development

Abstract

This invention discloses a method, apparatus, and related equipment for dynamic characterization of oil reservoir development. The method first establishes a continuous oil saturation generation algorithm in the time dimension to address the problem of discontinuous time in actual oil saturation monitoring data. Then, it establishes spatial parameter field generation algorithms in each time dimension, which, combined with physical boundary condition constraints, generate planar physical fields for each parameter based on scattered data points. Next, it utilizes a deep neural network to unify the initial oil saturation of each well based on the characteristics of monitoring data from adjacent wells, solving the problem of missing initial oil saturation in some wells due to inconsistent logging times. Finally, it constructs a dynamic parameter intelligent prediction model capable of mining spatiotemporal data characteristics to predict the future distribution of the saturation physical field and / or pressure physical field.

Description

Technical Field

[0001] This invention relates to the field of oil and gas development technology, and in particular to a method, apparatus and related equipment for dynamic characterization of oil reservoir development. Background Technology

[0002] As oilfield development progresses, reservoir dynamic parameters such as pressure and remaining oil distribution directly influence the formulation of technical policies and potential tapping plans. These parameters are crucial for achieving the ultimate recovery rate and promoting economical oilfield development, and are a key focus for reservoir engineers in the mid-to-late stages of reservoir development. However, due to the extremely complex geological structure of reservoirs, existing technologies are insufficient for exploring remaining oil. Furthermore, the distribution characteristics of remaining oil in the mid-to-late stages of oilfield development are highly dispersed and uncertain, thus limiting the application of conventional techniques. For example, logging interpretation methods based on static data are costly, and traditional numerical simulation methods involve multi-parameter collaborative calculations, making it difficult to accurately describe complex oil and gas seepage processes, resulting in uncertainties in prediction accuracy. Summary of the Invention

[0003] In view of the above problems, the present invention is proposed to provide a method, apparatus and related equipment for dynamic characterization of reservoir development that overcomes or at least partially solves the above problems.

[0004] In a first aspect, embodiments of the present invention provide a method for dynamic characterization of reservoir development, which may include:

[0005] Based on well logging data and dynamic production data of the study block, a dataset of continuous oil saturation, water saturation and gas saturation of each well in the study block in the time dimension was determined.

[0006] Based on the well logging data of the study block, the pressure values ​​in the dynamically generated data, and the datasets of oil saturation, water saturation, and gas saturation of each well in the time dimension, the continuous datasets of each parameter in the two-dimensional planar physical field of the study block in each time dimension are determined.

[0007] The initial saturation of each well in the study block is supplemented based on data from neighboring wells to unify the initial saturation of each well.

[0008] Based on the supplemented initial saturation and the pressure values ​​of each well in the study block in the time dimension, as well as the static parameters in the logging data, a dynamic and static parameter data structure is constructed.

[0009] After dividing the dynamic and static parameter data structure into preset time windows, the dynamic and static parameter data structure with the set time windows is input into a pre-trained deep neural network model containing a convolutional neural network and a long short-term memory network. The two-dimensional planar parameter features output by the convolutional neural network are input into the long short-term memory network to obtain the saturation physical field distribution and / or pressure physical field distribution of the study block.

[0010] Optionally, the dataset of continuous oil saturation, water saturation, and gas saturation for each well within the study block, based on well logging data and dynamic production data of the study block, includes:

[0011] Acquire well logging data and dynamic production data for the study block; the well logging data includes permeability and porosity; the dynamic production data includes: oil production, water production, gas production, pressure, water injection volume, and gas injection volume;

[0012] The oil saturation, water saturation, and gas saturation of each well in the study block are determined based on the secondary interpretation of the logging data.

[0013] Based on the oil saturation, water saturation, and gas saturation of each well in the study block, and the dynamic production data, a corresponding relationship between the oil saturation, water saturation, and gas saturation of each well in the study block in the time dimension is constructed to determine the dataset of the continuous oil saturation, water saturation, and gas saturation of each well in the study block in the time dimension.

[0014] Optionally, the dataset based on well logging data of the study block, pressure values ​​in the dynamically generated data, and continuous oil saturation, water saturation, and gas saturation of each well in the time dimension, determines the continuous dataset of each parameter in the two-dimensional planar physical field of the study block in each time dimension, including:

[0015] Based on the permeability and porosity data from the well logging data of the study block, a two-dimensional permeability physical field and a two-dimensional porosity physical field were constructed in the time dimension, respectively.

[0016] A two-dimensional saturation physical field in the time dimension is constructed based on the dataset of continuous oil saturation, water saturation and gas saturation of each well in the time dimension; wherein, the two-dimensional saturation physical field includes: a two-dimensional oil saturation physical field, a two-dimensional water saturation physical field and a two-dimensional gas saturation physical field.

[0017] A two-dimensional pressure physical field in the time dimension is constructed based on the pressure value of each well in the time dimension;

[0018] Based on the two-dimensional permeability physical field, two-dimensional porosity physical field, two-dimensional saturation physical field, and two-dimensional pressure physical field, a continuous dataset of each parameter in the two-dimensional planar physical field of the study block is determined in each time dimension.

[0019] Optionally, supplementing the initial saturation of each well within the study block based on neighboring well data to unify the initial saturation of each well may include:

[0020] The neighboring well data corresponding to the missing time of the target well are input into a pre-trained BP neural network model to obtain the saturation data of the missing time of the target well, so as to unify the initial time saturation of each well.

[0021] Optionally, the BP neural network model can be pre-trained through the following steps:

[0022] Obtain a training sample set, wherein each sample in the training sample set includes target well saturation data and adjacent well saturation data for the same time interval;

[0023] The BP neural network model is trained using samples from the training sample set, wherein adjacent well saturation data is input into the BP neural network model, and target well saturation data is used as the output of the BP neural network model to estimate the weight parameters and threshold parameters of the BP neural network model.

[0024] Optionally, the deep neural network model can be pre-trained through the following steps:

[0025] Obtain a training sample set, wherein each sample in the training sample set includes a dynamic and static parameter data structure for a preset time window and an oil saturation distribution of a two-dimensional planar physical field;

[0026] The deep neural network model, which includes a convolutional neural network and a long short-term memory network, is trained using samples from the training sample set. The two-dimensional planar parameter features output by the convolutional upgrade network are input into the long short-term memory network to estimate the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block.

[0027] Optionally, after obtaining the saturation physical field distribution and / or pressure physical field distribution of the study block, the method may further include: evaluating the influence weights of each parameter field in the two-dimensional planar physical field on the deep neural network model based on noise measurement.

[0028] Secondly, embodiments of the present invention provide a method for training a machine learning model, which may include:

[0029] Obtain a training sample set, wherein each sample in the training sample set includes a dynamic and static parameter data structure for a preset time window and an oil saturation distribution of a two-dimensional planar physical field;

[0030] The deep neural network model, which includes a convolutional neural network and a long short-term memory network, is trained using samples from the training sample set. The two-dimensional planar parameter features output by the convolutional upgrade network are input into the long short-term memory network to estimate the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block.

[0031] Optionally, obtaining the training sample set specifically includes:

[0032] Based on well logging data and dynamic production data of the study block, a dataset of continuous oil saturation, water saturation and gas saturation of each well in the study block in the time dimension was determined.

[0033] Based on the well logging data of the study block, the pressure values ​​in the dynamically generated data, and the datasets of oil saturation, water saturation, and gas saturation of each well in the time dimension, the continuous datasets of each parameter in the two-dimensional planar physical field of the study block in each time dimension are determined.

[0034] The initial saturation of each well in the study block is supplemented based on data from neighboring wells to unify the initial saturation of each well.

[0035] Based on the supplemented initial saturation and the pressure values ​​of each well in the study block in the time dimension, as well as the static parameters in the logging data, a dynamic and static parameter data structure is constructed.

[0036] After dividing the dynamic and static parameter data structure into preset time windows, the dynamic and static parameter data structure of the set time window and the saturation physical field distribution and / or pressure physical field distribution of the corresponding time window are used as training samples to obtain a training sample set composed of training samples.

[0037] Optionally, it may also include: adjusting the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block based on an error function of pressure value and saturation data.

[0038] Thirdly, embodiments of the present invention provide another method for training a machine learning model, which may include:

[0039] Obtain a training sample set, wherein each sample in the training sample set includes target well saturation data and adjacent well saturation data for the same time interval;

[0040] The BP neural network model is trained using samples from the training sample set, wherein adjacent well saturation data is input into the BP neural network model, and target well saturation data is used as the output of the BP neural network model to estimate the weight parameters and threshold parameters of the BP neural network model.

[0041] Fourthly, embodiments of the present invention provide a dynamic characterization device for reservoir development, which may include:

[0042] The first determination module is used to determine the dataset of oil saturation, water saturation and gas saturation of each well in the study block in the time dimension based on the logging data and dynamic production data of the study block.

[0043] The second determining module is used to determine the continuous dataset of each parameter in the two-dimensional planar physical field of the study block in each time dimension based on the well logging data of the study block, the pressure value in the dynamically generated data, and the dataset of continuous oil saturation, water saturation and gas saturation of each well in the time dimension.

[0044] The supplementary module is used to supplement the initial saturation of each well in the study block based on adjacent well data, so as to unify the initial saturation of each well.

[0045] A construction module is used to construct a dynamic and static parameter data structure based on the supplemented initial saturation, the pressure values ​​of each well in the study block in the time dimension, and the static parameters in the logging data.

[0046] The physical field distribution determination module is used to divide the dynamic and static parameter data structure into preset time windows, and then input the dynamic and static parameter data structure of the set time window into a pre-trained deep neural network model containing a convolutional neural network and a long short-term memory network. The two-dimensional planar parameter features output by the convolutional neural network are input into the long short-term memory network to obtain the saturation physical field distribution and / or pressure physical field distribution of the study block.

[0047] Fifthly, embodiments of the present invention provide a training apparatus for a machine learning model, which may include:

[0048] The first acquisition module is used to acquire a training sample set, wherein each sample in the training sample set includes a dynamic and static parameter data structure of a preset time window and an oil saturation distribution of a two-dimensional planar physical field.

[0049] The first training module is used to train a deep neural network model containing a convolutional neural network and a long short-term memory network using samples from the training sample set. The two-dimensional planar parameter features output by the convolutional upgrade network are input into the long short-term memory network to estimate the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block.

[0050] Sixthly, embodiments of the present invention provide another training apparatus for a machine learning model, which may include:

[0051] The second acquisition module is used to acquire a training sample set, wherein each sample in the training sample set includes target well saturation data and adjacent well saturation data in the same time interval;

[0052] The second training module is used to train the BP neural network model using samples from the training sample set, wherein adjacent well saturation data are input into the BP neural network model, and target well saturation data are used as the output of the BP neural network model to estimate the weight parameters and threshold parameters of the BP neural network model.

[0053] In a seventh aspect, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the reservoir development dynamic characterization method described in the first aspect, or the machine learning model training method described in the second aspect, or the machine learning model training method described in the third aspect.

[0054] Eighthly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the reservoir development dynamic characterization method described in the first aspect, or the machine learning model training method described in the second aspect, or the machine learning model training method described in the third aspect.

[0055] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:

[0056] This invention provides a method, apparatus, and related equipment for dynamic characterization of oil reservoir development. The method first establishes a continuous oil saturation generation algorithm in the time dimension to address the problem of discontinuous time in actual oil saturation monitoring data. Then, it establishes spatial parameter field generation algorithms in each time dimension, which, combined with physical boundary condition constraints, generate planar physical fields for each parameter based on scattered data points. Next, it utilizes a deep neural network to unify the initial oil saturation of each well based on the characteristics of monitoring data from adjacent wells, solving the problem of missing initial oil saturation in some wells due to inconsistent logging times. Finally, it constructs a dynamic parameter intelligent prediction model capable of mining spatiotemporal data characteristics to predict the future distribution of the saturation physical field and / or pressure physical field.

[0057] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0058] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0059] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0060] Figure 1 This is a flowchart of the training method for the machine learning model provided in Embodiment 1 of the present invention;

[0061] Figure 2 Here is a flowchart of step S11;

[0062] Figure 3 Here is a flowchart of step S111;

[0063] Figure 4 This is a schematic diagram of the two-dimensional oil saturation physical field constructed in each time dimension as provided in Embodiment 1 of the present invention;

[0064] Figure 5 This is a schematic diagram showing the initial oil saturation of each well as provided in Embodiment 1 of the present invention;

[0065] Figure 6 This is a schematic diagram of the deep neural network model provided in Embodiment 1 of the present invention;

[0066] Figure 7 This is a schematic diagram of the remaining oil distribution field and the new well location deployment decision distribution provided in Embodiment 1 of the present invention;

[0067] Figure 8 This is a schematic diagram of the structure of the training device for the machine learning model provided in Embodiment 1 of the present invention;

[0068] Figure 9 This is a flowchart of the reservoir development dynamic characterization method provided in Embodiment 2 of the present invention;

[0069] Figure 10 This is a schematic diagram of the reservoir development dynamic characterization device provided in Embodiment 2 of the present invention;

[0070] Figure 11 This is a flowchart of another training method for a machine learning model provided in Embodiment 3 of the present invention;

[0071] Figure 12 This is a schematic diagram of the structure of a training device for another machine learning model provided in Embodiment 3 of the present invention. Detailed Implementation

[0072] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0073] Example 1

[0074] Embodiment 1 of the present invention provides a method for training a machine learning model, referring to... Figure 1 As shown, the method may include the following steps:

[0075] Step S11: Obtain the training sample set. Each sample in the training sample set includes a dynamic and static parameter data structure for a preset time window and an oil saturation distribution of a two-dimensional planar physical field.

[0076] Step S12: Train a deep neural network model containing a convolutional neural network and a long short-term memory network using samples from the training sample set. The two-dimensional planar parameter features output by the convolutional upgrade network are input into the long short-term memory network to estimate the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block.

[0077] In this embodiment of the invention, when training the deep neural network model, a large amount of dynamic and static data from the oil and gas development process is used as the basis. The deep neural network model is trained to fully explore the implicit relationships between various types of data. At the same time, the deep neural network model is corrected based on pressure values ​​and other data in the dynamic data, so as to quickly and accurately depict the distribution of remaining oil in the reservoir and predict reservoir pressure, thereby providing a basis for oilfield development plan management decisions.

[0078] The above steps regarding obtaining the training sample set refer to... Figure 2 As shown, the specific steps may include:

[0079] Step S111: Based on the logging data and dynamic production data of the study block, determine the dataset of continuous oil saturation, water saturation and gas saturation of each well in the study block in the time dimension.

[0080] This step involves determining the saturation data of the study block using well logging and dynamic production data, referring to... Figure 3 As shown, the specific steps are as follows:

[0081] Step S1111: Obtain well logging data and dynamic production data for the study block; well logging data includes permeability and porosity; dynamic production data includes: oil production, water production, gas production, pressure, water injection, and gas injection.

[0082] The well logging interpretation data and dynamic production data collected from the oilfield mainly include static geological characteristic data of the locations of 54 wells, S=[K x,y ,Φ x,y ] and dynamic dataset D = [So x,y,t Sw x,y,t ,Sg x,y,t Qo x,y,t ,Qw x,y,t ,Qg x,y,t P x,y,t ,Iw x,y,t Ig x,y,t ]; where x, y, and t represent longitude, latitude, and time, S is a static geological dataset, K and Φ represent permeability and porosity, respectively; So, Sw, Sg, Qo, Qw, Qg, P, Iw, and Ig represent oil saturation, water saturation, gas saturation, oil production, water production, gas production, pressure, water injection, and gas injection, respectively.

[0083] Step S1112: Determine the oil saturation, water saturation, and gas saturation of each well in the study block based on the secondary interpretation of well logging data.

[0084] Q, P, and I data are derived from real-time dynamic monitoring in the field, thus exhibiting high continuity over time. So, Sw, and Sg are primarily calculated through secondary interpretation of well logging data. Due to the high cost of well logging measures, a well can only be tested 2-3 times over time, resulting in only 2-3 real data points for So, Sw, and Sg over time. Therefore, it is necessary to establish a continuous generation algorithm over time to improve the data.

[0085] Step S1113: Based on the oil saturation, water saturation and gas saturation of each well in the study block, and dynamic production data, construct the corresponding relationship between the oil saturation, water saturation and gas saturation of each well in the study block in the time dimension.

[0086]

[0087]

[0088]

[0089] Sg + So + Sw = 1 (4)

[0090] Wherein, α, β, and γ are the influence weights of the corresponding variables, which can be calculated and defined using actual So, Sw, and Sg data points and expert experience. 'a' represents the number of injection wells around the target well. Continuous oil saturation is calculated using equation (1) based on oil production, pressure, and the injection volume of surrounding wells. If the injection well injects water, continuous water saturation changes are calculated using equation (2). If the injection well injects gas, continuous gas saturation changes are calculated using equation (3). Equation (4) is the constraint condition. Finally, the corresponding relationships of continuous oil saturation, water saturation, and gas saturation in the time dimension can be obtained.

[0091] Step S1114: Determine the dataset of continuous oil saturation, water saturation, and gas saturation of each well in the study block over time.

[0092] This step is based on the correspondence between the oil saturation, water saturation and gas saturation in the time dimension obtained in step S1113 above, to determine the dataset of oil saturation, water saturation and gas saturation in the time dimension for each well.

[0093] Step S112: Based on the well logging data of the study block, the pressure values ​​in the dynamically generated data, and the datasets of oil saturation, water saturation, and gas saturation of each well in the time dimension, determine the continuous datasets of each parameter in the two-dimensional planar physical field of the study block in each time dimension.

[0094] Specifically, this step may include:

[0095] Based on the permeability and porosity data from the well logging data of the study block, a two-dimensional permeability physical field and a two-dimensional porosity physical field were constructed in the time dimension, respectively.

[0096] A two-dimensional saturation physical field is constructed based on a dataset of continuous oil saturation, water saturation, and gas saturation for each well over time. This two-dimensional saturation physical field includes: a two-dimensional oil saturation physical field, a two-dimensional water saturation physical field, and a two-dimensional gas saturation physical field. (Refer to...) Figure 4 As shown, the two-dimensional oil saturation physical field is constructed under various time dimensions, where the horizontal axis represents time and the vertical axis represents oil saturation, as well as the oil saturation field diagrams generated at two time points.

[0097] A two-dimensional pressure physical field in the time dimension is constructed based on the pressure value of each well in the time dimension;

[0098] Based on the two-dimensional permeability physical field, two-dimensional porosity physical field, two-dimensional saturation physical field, and two-dimensional pressure physical field, a continuous dataset of each parameter in the two-dimensional planar physical field of the study block is determined in each time dimension.

[0099] This step mainly establishes the spatial parameter field generation algorithm under each time dimension. For static data, the parameter changes are not significant with time, so the static field can be considered consistent at each time point. For dynamic fields, the parameter changes are more obvious with time. Whether it is a static field or a dynamic field, the data of the entire field needs to be generated based on the actual values ​​of the parameters at known well locations.

[0100] Taking the permeability K as an example, equation (5):

[0101]

[0102] in, This represents the estimated permeability at coordinates (x0, y0), where n is the number of points with known permeability, λ is the weighting coefficient, and k is the actual permeability at the known coordinate location. For any coordinate point, the permeability k... (x，y) The assumptions are as shown in the system of equations (6):

[0103]

[0104] in, Given the expected value of the actual permeability, σ 2 R represents the variance of the actual penetration rate. (x，y) This represents random bias. The solution process for λ in the above equation and R... (x，y)The relevant calculations can be performed by referring to the calculation principle of Kriging interpolation.

[0105] For predicting dynamic fields, since the changes are significant over time, constraints need to be added at the boundary to make them more consistent with actual physical understanding. Taking oil saturation as an example, the constraint at the boundary is shown in equation (7):

[0106]

[0107] Where, x c y c Let represent the boundary values ​​in the x and y directions, and let So represent the oil saturation.

[0108] The above constraints and algorithms can be used to generate physical fields with various parameters based on known data points.

[0109] Step S113: Supplement the initial saturation of each well in the study block based on the data of neighboring wells, so as to unify the initial saturation of each well.

[0110] This step involves inputting the neighboring well data corresponding to the missing time of the target well into a pre-trained BP neural network model to obtain the saturation data of the missing time of the target well, so as to unify the initial saturation of each well.

[0111] The BP neural network model in this embodiment of the invention is pre-trained through the following steps:

[0112] Obtain a training sample set, where each sample includes target well saturation data and adjacent well saturation data for the same time interval;

[0113] The BP neural network model is trained using samples from the training sample set, wherein adjacent well saturation data is input into the BP neural network model, and target well saturation data is used as the output of the BP neural network model to estimate the weight parameters and threshold parameters of the BP neural network model.

[0114] In this embodiment of the invention, while obtaining continuous oil saturation in the time dimension, the initial time dimension is not uniform due to the different logging or initial production time of the well. For example, a deep neural network model (BP neural network model) containing two hidden layers and 20 neurons in each layer is established to establish the relationship between the oil saturation of adjacent wells and the oil saturation of the target well under the same time dimension. The trained model is used to predict the unknown oil saturation of the target well.

[0115] Reference Figure 5 As shown, taking the adjacent well_1 and the target well_2 as examples, the oil saturation data of well_1 is So_1=[So1,So2,So3,…,So…].t The oil saturation data for well_2 is So_2 = [So z So z+1 So z+2 So t ], where t is time. We can see that the initial time of well_2 is z, and data from time 1 to z-1 is missing. Taking well_1 [So z So z+1 So z+2 So t The data is used as input to the deep neural network, with well_2[So] as the input. z So z+1 So z+2 So t As output, the implicit relationship between the oil saturation of wells 1 and 2 is constructed. The weights and thresholds in this model are calculated using the backpropagation algorithm to obtain the optimal model B. Then, the [So1,So2,So3,…,So] values ​​of well 1 are used to calculate the oil saturation. z-1 As input to the model, the trained model B can directly predict [So1, So2, So3, ..., So] of well_2. z-1 The oil saturation data were used to unify the initial oil saturation of each well.

[0116] Step S114: Based on the supplemented initial saturation and the pressure values ​​of each well in the study block in the time dimension, as well as the static parameters in the logging data, construct a dynamic and static parameter data structure.

[0117] Using the above-described generation algorithm based on time and spatial dimension parameter fields, a static geological dataset SS = [K, Φ] can be obtained, where K and Φ represent the permeability field and porosity field distribution, respectively (the data dimension of the parameter fields in this experiment is 64 x 64), and the dynamic parameter field is DD. t =[P t So t Sw t Sg t P, So, Sw, and Sg represent the pressure field, oil saturation field, water saturation field, and gas saturation field, respectively (all data dimensions are 64x64), and t represents time. To simultaneously extract dynamic and static features, a feature field overlay method is used to construct a new data structure A. t =[P t So t Sw t ;Sg t ;K;Φ],A tThe data structure representing time t is constructed by stacking pressure field, oil saturation field, water saturation field, gas saturation field, permeability, and porosity field (i.e., data dimension is 64 x 64 x 6); the resulting dataset is A = [A1, A2, A3, ..., A...]. t The dataset is divided into a training set and a test set according to a certain ratio. The training set is A_train = [A1, A2, A3, ..., A...]. m The test set is A_test = [A m+1 A m+2 A m+3 ,…,A t (For example, if the ratio is 9:1, then m:t = 9 / 10); the input in each sample is determined by a time window. Taking the training set as an example, the time window is set to h, X d =[A d+1 A d+2 A d+3 ,…,A d+h ], X d Let X_train be the input to the d-th sample in the training set. Then the input to the training set is X_train = [X1, X2, X3, ..., X...]. m-h+1 The output of the training set is Y_train = [So] h+1 So h+2 So h+3 ,…,So m Similarly, the input to the test set, X_test, can be obtained as [X...]. m+1 X m+2 X m+3 ,…,X t-h+1 ] and output Y_test = [So m+h+1 So m+h+2 So m+h+3 ,…,So t ].

[0118] Step S115: After dividing the dynamic and static parameter data structure into preset time windows, the dynamic and static parameter data structure of the set time window and the saturation physical field distribution and / or pressure physical field distribution of the corresponding time window are used as a training sample to obtain a training sample set composed of training samples.

[0119] In another optional embodiment, step S12 above uses the training sample set obtained in step S115 for training to construct a dynamic parameter intelligent prediction model for mining spatiotemporal data features. The deep neural network model described in this embodiment mainly consists of two parts: one part uses a convolutional neural network to extract features from spatial physical field data, and the other part uses a Long Short-Term Memory (LSTM) network to learn the changing characteristics in the time dimension based on the extracted spatial physical field features, thereby predicting the future distribution of saturation physical fields and / or pressure physical fields. Figure 6 As shown, the deep neural network model mainly consists of an input layer, convolutional layer, transposed layer, LSTM layer, deconvolutional layer, and output layer. The convolutional kernel size in the convolutional layer is 4x4x6, which can extract each multi-layer data structure A within a window length h in one operation. d+1 ~A d+h Spatial data features are utilized by transposing the multi-layer data structure A after convolution. d+1 ~A d+h Centralized extraction is performed on the same parameter layer, such as A. d+1 ~A d+h The first layer consists entirely of pressure fields, concentrating the pressure fields from all the data structures to facilitate subsequent extraction of pressure field variations over time using LSTM. The LSTM layer is constructed from two-layer Convlstm units, each with 60 neurons, used to extract parametric features from each physical field. Finally, a deconvolution layer is used to input the oil saturation field, with a deconvolution kernel size of 4x4x6.

[0120] In this embodiment of the invention, a predictive model for mining spatiotemporal data features is finally established. On the one hand, it can accurately predict the future pressure field, and on the other hand, it can characterize the oil saturation distribution features. Based on existing static fields such as permeability field and porosity field, as well as dynamic fields such as pressure field and saturation field, it can predict the future pressure and remaining oil distribution.

[0121] In another alternative embodiment, refer to Figure 1 As shown, the method may further include: step S13, adjusting the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block based on the error function of pressure value and saturation data.

[0122] First, initialize the weights and thresholds of each neuron in the model, and then input the training set X_train = [X1, X2, X3, ..., X...]. m-h+1 In the imported model, the input parameters are weighted and calculated using the weight and threshold matrices of the neurons through forward propagation. The output model's predicted value Y_train_pre = [So h+1 ',So h+2 ',So h+3',…,So m '], where So h+1 'This is the predicted value for the first sample; the predicted value and the actual value Y_train are calculated using the residual function.' h+1 So h+2 So h+3 ,…,So m The error is calculated, and the weights and thresholds in the network are updated using the backpropagation algorithm based on the error, thereby training the intelligent model. Since the change in residual oil saturation and pressure have a significant nonlinear relationship during actual field development, the training of the intelligent model can be guided by pressure field data characteristics. Therefore, a pressure field-guided error function f is considered. error As shown in equation (8):

[0123]

[0124] Where N is the number of samples (e.g., in the training set, N = m - h + 1), So is the actual saturation value, So' is the saturation value predicted by the model, P is the actual pressure value, P' is the pressure value calculated internally by the model using the chain rule, and μ1 and μ2 are the influence weights of oil saturation and pressure, respectively, both 0.5 in this experiment. After training the model on the training set, the predictive performance of the model is evaluated using the test set to obtain the optimal trained model. Figure 7 The figure shows the physical field characterization of oil saturation, which can be used to determine the distribution of remaining oil and then to make decisions on the deployment of new well locations.

[0125] Based on the same inventive concept, this embodiment of the invention also provides a training device for a machine learning model, referring to... Figure 8 As shown, the device may include a first acquisition module 81 and a first training module 82, and its working principle is as follows:

[0126] The first acquisition module 81 is used to acquire a training sample set. Each sample in the training sample set includes a dynamic and static parameter data structure of a preset time window and an oil saturation distribution of a two-dimensional planar physical field.

[0127] The first training module 82 is used to train a deep neural network model containing a convolutional neural network and a long short-term memory network using samples from the training sample set. The two-dimensional planar parameter features output by the convolutional upgrade network are input into the long short-term memory network to estimate the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block.

[0128] In an optional embodiment, the first acquisition module 81 is specifically used to: determine a dataset of oil saturation, water saturation and gas saturation of each well in the study block in a continuous time dimension based on the logging data and dynamic production data of the study block.

[0129] Based on the well logging data of the study block, the pressure values ​​in the dynamically generated data, and the datasets of oil saturation, water saturation, and gas saturation of each well in the time dimension, the continuous datasets of each parameter in the two-dimensional planar physical field of the study block in each time dimension are determined.

[0130] The initial saturation of each well in the study block is supplemented based on data from neighboring wells to unify the initial saturation of each well.

[0131] Based on the supplemented initial saturation and the pressure values ​​of each well in the study block in the time dimension, as well as the static parameters in the logging data, a dynamic and static parameter data structure is constructed.

[0132] After dividing the dynamic and static parameter data structure into preset time windows, the dynamic and static parameter data structure of the set time window and the saturation physical field distribution and / or pressure physical field distribution of the corresponding time window are used as training samples to obtain a training sample set composed of training samples.

[0133] Based on the same inventive concept, this embodiment of the invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is the training method of the aforementioned machine learning model.

[0134] Based on the same inventive concept, this embodiment of the invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the training method of the above-mentioned machine learning model.

[0135] The principles by which the above-mentioned devices, media, and related equipment in the embodiments of the present invention solve the problem are similar to those of the aforementioned methods. Therefore, their implementation can refer to the implementation of the aforementioned methods, and repeated details will not be repeated.

[0136] Example 2

[0137] Embodiment 2 of this invention provides a method for dynamic characterization of reservoir development. The reservoir dynamics in this embodiment mainly involve the dynamic changes in the pressure physical field and the saturation physical field. (Refer to...) Figure 9 As shown, the method may include the following steps:

[0138] Step S91: Based on the logging data and dynamic production data of the study block, determine the dataset of continuous oil saturation, water saturation and gas saturation of each well in the study block in the time dimension.

[0139] The specific execution of this step is as follows: Obtain well logging data and dynamic production data for the study block; well logging data includes permeability and porosity; dynamic production data includes: oil production, water production, gas production, pressure, water injection volume, and gas injection volume; determine the oil saturation, water saturation, and gas saturation of each well in the study block based on the secondary interpretation of the well logging data; construct the corresponding relationship between the oil saturation, water saturation, and gas saturation of each well in the study block and the dynamic production data in the time dimension, to determine the dataset of continuous oil saturation, water saturation, and gas saturation of each well in the study block in the time dimension.

[0140] In a specific example, well logging interpretation data and dynamic production data provided by the oilfield were collected. In this embodiment, the study block mainly includes static geological feature data S = [K] of the locations of 54 wells. x,y ,Φ x,y ] and dynamic dataset D = [So x,y,t Sw x,y,t ,Sg x,y,t Qo x,y,t ,Qw x,y,t ,Qg x,y,t P x,y,t ,Iw x,y,t Ig x,y,t [; where x, y, and t represent longitude, latitude, and time, S is a static geological dataset, and K and Φ represent permeability and porosity, respectively; So, Sw, Sg, Qo, Qw, Qg, P, Iw, and Ig represent oil saturation, water saturation, gas saturation, oil production, water production, gas production, pressure, water injection, and gas injection, respectively. Q, P, and I data are derived from real-time dynamic monitoring in the field, thus exhibiting high continuity over time. So, Sw, and Sg are primarily calculated through secondary interpretation of well logging data. Due to the high cost of well logging, a single well can only be tested 2-3 times over time, resulting in only 2-3 real data points for So, Sw, and Sg over time. Therefore, a continuous generation algorithm over time is needed to improve the data.]

[0141]

[0142]

[0143]

[0144] Sg + So + Sw = 1 (4)

[0145] Wherein, α, β, and γ are the influence weights of the corresponding variables, which can be calculated and defined using actual So, Sw, and Sg data points and expert experience. 'a' represents the number of injection wells around the target well. Continuous oil saturation is calculated using equation (1) based on oil production, pressure, and the injection volume of surrounding wells. If the injection well injects water, continuous water saturation changes are calculated using equation (2). If the injection well injects gas, continuous gas saturation changes are calculated using equation (3). Equation (4) is the constraint condition. Finally, the corresponding relationships of continuous oil saturation, water saturation, and gas saturation in the time dimension can be obtained. Finally, based on the obtained corresponding relationships of continuous oil saturation, water saturation, and gas saturation in the time dimension, the dataset of continuous oil saturation, water saturation, and gas saturation for each well in the time dimension is determined.

[0146] Step S92: Based on the well logging data of the study block, the pressure values ​​in the dynamically generated data, and the datasets of continuous oil saturation, water saturation, and gas saturation of each well in the time dimension, determine the continuous datasets of each parameter in the two-dimensional planar physical field of the study block in each time dimension.

[0147] Specifically, this step is performed as follows: Based on the permeability and porosity data from the well logging data of the study block, construct a two-dimensional permeability physical field and a two-dimensional porosity physical field in the time dimension, respectively; based on the continuous oil saturation, water saturation, and gas saturation datasets of each well in the time dimension, construct a two-dimensional saturation physical field in the time dimension; wherein, the two-dimensional saturation physical field includes: a two-dimensional oil saturation physical field, a two-dimensional water saturation physical field, and a two-dimensional gas saturation physical field; based on the pressure values ​​of each well in the time dimension, construct a two-dimensional pressure physical field in the time dimension; based on the two-dimensional permeability physical field, the two-dimensional porosity physical field, the two-dimensional saturation physical field, and the two-dimensional pressure physical field, determine the continuous datasets of each parameter in the two-dimensional planar physical field of the study block in each time dimension.

[0148] This step mainly establishes the spatial parameter field generation algorithm under each time dimension. For static data, the parameter changes are not significant with time, so the static field can be considered consistent at each time point. For dynamic fields, the parameter changes are more obvious with time. Whether it is a static field or a dynamic field, the data of the entire field needs to be generated based on the actual values ​​of the parameters at known well locations.

[0149] Taking the permeability K as an example, equation (5):

[0150]

[0151] in, This represents the estimated permeability at coordinates (x0, y0), where n is the number of points with known permeability, λ is the weighting coefficient, and k is the actual permeability at the known coordinate location. For any coordinate point, the permeability k... (x，y) The assumptions are as shown in the system of equations (6):

[0152]

[0153] in, Given the expected value of the actual permeability, σ 2 R represents the variance of the actual penetration rate. (x，y) This represents random bias. The solution process for λ in the above equation and R... (x，y) The relevant calculations can be performed by referring to the calculation principle of Kriging interpolation.

[0154] For predicting dynamic fields, since the changes are significant over time, constraints need to be added at the boundary to make them more consistent with actual physical understanding. Taking oil saturation as an example, the constraint at the boundary is shown in equation (7):

[0155]

[0156] Where, x c y c Let represent the boundary values ​​in the x and y directions, and let So represent the oil saturation.

[0157] The above constraints and algorithms can be used to generate physical fields with various parameters based on known data points.

[0158] Step S93: Supplement the initial saturation of each well in the study block based on neighboring well data to unify the initial saturation of each well. Specifically, input the neighboring well data corresponding to the missing time of the target well into a pre-trained BP neural network model to obtain the saturation data of the missing time of the target well to unify the initial saturation of each well.

[0159] This step involves inputting the neighboring well data corresponding to the missing time of the target well into a pre-trained BP neural network model to obtain the saturation data of the missing time of the target well, so as to unify the initial saturation of each well.

[0160] The BP neural network model in this embodiment of the invention is pre-trained through the following steps:

[0161] Obtain a training sample set, where each sample includes target well saturation data and adjacent well saturation data for the same time interval;

[0162] The BP neural network model is trained using samples from the training sample set, wherein adjacent well saturation data is input into the BP neural network model, and target well saturation data is used as the output of the BP neural network model to estimate the weight parameters and threshold parameters of the BP neural network model.

[0163] In this embodiment of the invention, while obtaining continuous oil saturation in the time dimension, the initial time dimension is not uniform due to the different logging or initial production time of the well. For example, a deep neural network model (BP neural network model) containing two hidden layers and 20 neurons in each layer is established to establish the relationship between the oil saturation of adjacent wells and the oil saturation of the target well under the same time dimension. The trained model is used to predict the unknown oil saturation of the target well.

[0164] Reference Figure 5 As shown, taking the adjacent well_1 and the target well_2 as examples, the oil saturation data of well_1 is So_1=[So1,So2,So3,…,So…]. t The oil saturation data for well_2 is So_2 = [So z So z+1 So z+2 So t ], where t is time. We can see that the initial time of well_2 is z, and data from time 1 to z-1 is missing. Taking well_1 [So z So z+1 So z+2 So t The data is used as input to the deep neural network, with well_2[So] as the input. z So z+1 So z+2 So t As output, the implicit relationship between the oil saturation of wells 1 and 2 is constructed. The weights and thresholds in this model are calculated using the backpropagation algorithm to obtain the optimal model B. Then, the [So1,So2,So3,…,So] values ​​of well 1 are used to calculate the oil saturation. z-1 As input to the model, the trained model B can directly predict [So1, So2, So3, ..., So] of well_2. z-1 The oil saturation data were used to unify the initial oil saturation of each well.

[0165] Step S94: Based on the supplemented initial saturation and the pressure values ​​of each well in the study block in the time dimension, as well as the static parameters in the logging data, construct a dynamic and static parameter data structure.

[0166] Using the above-described generation algorithm based on time and spatial dimension parameter fields, a static geological dataset SS = [K, Φ] can be obtained, where K and Φ represent the permeability field and porosity field distribution, respectively (the data dimension of the parameter fields in this experiment is 64 x 64), and the dynamic parameter field is DD. t =[P t So t Sw t Sg t P, So, Sw, and Sg represent the pressure field, oil saturation field, water saturation field, and gas saturation field, respectively (all data dimensions are 64x64), and t represents time. To simultaneously extract dynamic and static features, a feature field overlay method is used to construct a new data structure A. t =[P t So t Sw t ;Sg t ;K;Φ],A t The data structure representing time t is constructed by stacking pressure field, oil saturation field, water saturation field, gas saturation field, permeability and porosity field (i.e., the data dimension is 64×64×6).

[0167] Step S95: After dividing the dynamic and static parameter data structure into preset time windows, input the dynamic and static parameter data structure with the set time windows into a pre-trained deep neural network model containing a convolutional neural network and a long short-term memory network. The two-dimensional planar parameter features output by the convolutional neural network are input into the long short-term memory network to obtain the saturation physical field distribution and / or pressure physical field distribution of the study block.

[0168] It should be noted that the input in each sample is determined by a time window, which is set to h, X d =[A d+1 A d+2 A d+3 ,…,A d+h ], X d This is the input for the d-th sample in the training set. It should also be noted that the pre-trained deep neural network model containing convolutional neural networks and long short-term memory networks described in this embodiment can be pre-trained using the methods described in Embodiment 1 above, or it can be pre-trained using other methods. This embodiment does not specifically limit this.

[0169] The method described in this embodiment of the invention first establishes a continuous oil saturation generation algorithm in the time dimension to solve the problem of discontinuous time in actual oil saturation monitoring data; then, it establishes a spatial parameter field generation algorithm in each time dimension, which can combine physical boundary condition constraints to generate planar physical fields of various parameters based on scattered data points; next, it uses a deep neural network to unify the initial oil saturation of each well based on the monitoring data characteristics of adjacent wells, solving the problem of missing initial oil saturation of some wells due to inconsistent logging times; finally, it constructs a dynamic parameter intelligent prediction model that can mine spatiotemporal data characteristics, and characterizes the distribution of remaining oil by predicting future oil saturation.

[0170] In another alternative embodiment, reference is also made to Figure 9 As shown, the method may further include: step S96, evaluating the influence weights of each parameter field in the two-dimensional plane physical field on the deep neural network model based on the noise measurement method.

[0171] In the dynamic and static parameter data structure, noise of the same degree is added to each physical field, and the degree of decrease in oil saturation accuracy under different conditions is calculated. The higher the degree of decrease in accuracy, the greater the influence of the physical field on the dynamic parameter intelligent model. Taking the pressure field as an example, noise is added to the pressure field using formula (9):

[0172]

[0173] in, This is the estimated value after adding noise, where P is the actual value at that point. Let be the expected value of the pressure field at time t, where T is a random fluctuation function with a value range of [-0.4, 0.4]. With other physical fields remaining constant, a new dataset, denoted A_P, is created by adding noise to the pressure field and other physical fields to train a dynamic parameter intelligent model. The error is calculated using equation (10-11).

[0174]

[0175] Loss(A_P)=E(A_P)-E(A) (11)

[0176] Where E is the error value, N is the number of samples, So is the actual saturation value, and So' is the saturation value predicted by the model. E(A) represents the oil saturation error calculated from dataset A without noise, E(A_P) represents the oil saturation error calculated from the dataset after adding noise in the pressure field, and Loss(A_P) is the decrease in error calculated from the original dataset after adding noise in the pressure field. Similarly, the decrease in model error after adding noise in other physical fields can be calculated. Finally, the proportion of the decrease in error calculated from each physical field to the total decrease in error is calculated to determine the influence weight of each physical field on the model.

[0177] In the examples provided in this embodiment, oil saturation has the greatest impact, accounting for 47.93% of the weight, followed by pressure (13.25%), gas saturation (14.98%), water saturation (10.17%), permeability (6.24%), and porosity (7.43%).

[0178] Based on the same inventive concept, this invention also provides a reservoir development dynamic characterization device, referring to... Figure 10 As shown, the device may include: a first determining module 101, a second determining module 102, a supplementing module 103, a constructing module 104, and a physical field distribution determining module 105, and its working principle is as follows:

[0179] The first determining module 101 is used to determine the dataset of oil saturation, water saturation and gas saturation of each well in the study block in the time dimension based on the logging data and dynamic production data of the study block.

[0180] The second determining module 102 is used to determine the continuous dataset of each parameter in the two-dimensional planar physical field of the study block in each time dimension based on the well logging data of the study block, the pressure value in the dynamically generated data, and the dataset of oil saturation, water saturation and gas saturation of each well in the time dimension.

[0181] The supplementary module 103 is used to supplement the initial saturation of each well in the study block based on the data of adjacent wells, so as to unify the initial saturation of each well.

[0182] Module 104 is used to construct a dynamic and static parameter data structure based on the supplemented initial saturation, the pressure values ​​of each well in the study block in the time dimension, and the static parameters in the logging data.

[0183] The physical field distribution determination module 105 is used to divide the dynamic and static parameter data structure into preset time windows, and then input the dynamic and static parameter data structure of the set time window into a pre-trained deep neural network model containing a convolutional neural network and a long short-term memory network. The two-dimensional planar parameter features output by the convolutional neural network are input into the long short-term memory network to obtain the saturation physical field distribution and / or pressure physical field distribution of the study block.

[0184] In an optional embodiment, the first determining module 101 is specifically used for:

[0185] Acquire well logging data and dynamic production data for the study block; the well logging data includes permeability and porosity; the dynamic production data includes: oil production, water production, gas production, pressure, water injection volume, and gas injection volume;

[0186] The oil saturation, water saturation, and gas saturation of each well in the study block are determined based on the secondary interpretation of the logging data.

[0187] Based on the oil saturation, water saturation, and gas saturation of each well in the study block, and the dynamic production data, a corresponding relationship between the oil saturation, water saturation, and gas saturation of each well in the study block in the time dimension is constructed to determine the dataset of the continuous oil saturation, water saturation, and gas saturation of each well in the study block in the time dimension.

[0188] In another optional embodiment, the second determining module 102 is specifically used for:

[0189] Based on the permeability and porosity data from the well logging data of the study block, a two-dimensional permeability physical field and a two-dimensional porosity physical field were constructed in the time dimension, respectively.

[0190] A two-dimensional saturation physical field in the time dimension is constructed based on the dataset of continuous oil saturation, water saturation and gas saturation of each well in the time dimension; wherein, the two-dimensional saturation physical field includes: a two-dimensional oil saturation physical field, a two-dimensional water saturation physical field and a two-dimensional gas saturation physical field.

[0191] A two-dimensional pressure physical field in the time dimension is constructed based on the pressure value of each well in the time dimension;

[0192] Based on the two-dimensional permeability physical field, two-dimensional porosity physical field, two-dimensional saturation physical field, and two-dimensional pressure physical field, a continuous dataset of each parameter in the two-dimensional planar physical field of the study block is determined in each time dimension.

[0193] In another optional embodiment, the supplementary module 103 is specifically used to: input the neighboring well data corresponding to the missing time of the target well into a pre-trained BP neural network model to obtain the saturation data of the missing time of the target well in order to unify the initial time saturation of each well.

[0194] The BP neural network model is pre-trained through the following steps:

[0195] Obtain a training sample set, wherein each sample in the training sample set includes target well saturation data and adjacent well saturation data for the same time interval;

[0196] The BP neural network model is trained using samples from the training sample set, wherein adjacent well saturation data is input into the BP neural network model, and target well saturation data is used as the output of the BP neural network model to estimate the weight parameters and threshold parameters of the BP neural network model.

[0197] In another alternative embodiment, refer to Figure 10 As shown, the device may further include: an evaluation module 106, which evaluates the influence weights of each parameter field in the two-dimensional planar physical field on the deep neural network model based on a noise measurement method.

[0198] Based on the same inventive concept, this embodiment of the invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described dynamic characterization method for reservoir development.

[0199] Based on the same inventive concept, this embodiment of the invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned dynamic characterization method for reservoir development.

[0200] The principles by which the above-mentioned devices, media, and related equipment in the embodiments of the present invention solve the problem are similar to those of the aforementioned methods. Therefore, their implementation can refer to the implementation of the aforementioned methods, and repeated details will not be repeated.

[0201] Example 3

[0202] Embodiment 3 of the present invention also provides another machine learning training method, referring to... Figure 11 As shown, the method may include the following steps:

[0203] Step S1101: Obtain the training sample set. Each sample in the training sample set includes the target well saturation data and the adjacent well saturation data in the same time interval.

[0204] Step S1102: Train the BP neural network model using samples from the training sample set. The saturation data of adjacent wells are input into the BP neural network model, and the saturation data of the target well is used as the output of the BP neural network model. Estimate the weight parameters and threshold parameters of the BP neural network model.

[0205] The specific execution in this embodiment of the invention can refer to the training process of the pre-trained BP neural network model in step S113 of embodiment 1 above, and... Figure 5 The relevant examples in the embodiments of the present invention will not be repeated here.

[0206] Based on the same inventive concept, this embodiment of the invention also provides a training device for a machine learning model, referring to... Figure 12 As shown, the device may include a second acquisition module 121 and a second training module 122, and its working principle is as follows:

[0207] The second acquisition module 121 is used to acquire a training sample set, wherein each sample in the training sample set includes target well saturation data and adjacent well saturation data in the same time interval;

[0208] The second training module 122 is used to train the BP neural network model using samples from the training sample set, wherein adjacent well saturation data is input into the BP neural network model, target well saturation data is used as the output of the BP neural network model, and parameter estimation of the weight parameters and threshold parameters of the BP neural network model is performed.

[0209] Based on the same inventive concept, this embodiment of the invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the training method of the above-mentioned machine learning model.

[0210] Based on the same inventive concept, this embodiment of the invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the training method of the above-mentioned machine learning model.

[0211] The principles by which the above-mentioned devices, media, and related equipment in the embodiments of the present invention solve the problem are similar to those of the aforementioned methods. Therefore, their implementation can refer to the implementation of the aforementioned methods, and repeated details will not be repeated.

[0212] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0213] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0214] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0215] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0216] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method of dynamic reservoir characterization, characterized in that, include: The oil saturation, water saturation and gas saturation of each well in the study block are determined based on the secondary interpretation of the well logging data in the study block. Based on the oil saturation, water saturation and gas saturation of each well in the study block, and the dynamic production data of the study block, a corresponding relationship between the oil saturation, water saturation and gas saturation of each well in the study block in the time dimension is constructed to determine the dataset of the oil saturation, water saturation and gas saturation of each well in the study block in the time dimension. Based on the well logging data of the study block, the pressure values ​​in the dynamic production data, and the continuous data sets of oil saturation, water saturation, and gas saturation of each well in the time dimension, the continuous data sets of each parameter in the two-dimensional planar physical field of the study block in each time dimension are determined. The neighboring well data corresponding to the missing time of the target well within the study block are input into a pre-trained BP neural network model to obtain the saturation data of the target well at the missing time, thereby supplementing the initial saturation of each well within the study block and unifying the initial saturation of each well. The BP neural network model is pre-trained through the following steps: obtaining a training sample set, where each sample includes target well saturation data and neighboring well saturation data for the same time interval; training the BP neural network model using the samples in the training sample set, wherein neighboring well saturation data is input into the BP neural network model, and target well saturation data is used as the output of the BP neural network model, to estimate the weight parameters and threshold parameters of the BP neural network model. Based on the supplemented initial saturation and the pressure values ​​of each well in the study block in the time dimension, as well as the static parameters in the logging data, a dynamic and static parameter data structure is constructed. After dividing the dynamic and static parameter data structure into preset time windows, the dynamic and static parameter data structure with the set time windows is input into a pre-trained deep neural network model containing a convolutional neural network and a long short-term memory network. The two-dimensional planar parameter features output by the convolutional neural network are input into the long short-term memory network to obtain the saturation physical field distribution and / or pressure physical field distribution of the study block.

2. The method according to claim 1, characterized in that, Before determining the datasets of continuous oil saturation, water saturation, and gas saturation for each well within the study block over time, the following steps are also included: Acquire well logging data and dynamic production data for the study block; the well logging data includes permeability and porosity; the dynamic production data includes: oil production, water production, gas production, pressure, water injection volume, and gas injection volume.

3. The method according to claim 1, characterized in that, The datasets based on well logging data of the study block, pressure values ​​from the dynamic production data, and continuous oil saturation, water saturation, and gas saturation of each well over time are used to determine the continuous datasets of various parameters in the two-dimensional planar physical field of the study block over various time dimensions, including: Based on the permeability and porosity data from the well logging data of the study block, a two-dimensional permeability physical field and a two-dimensional porosity physical field were constructed in the time dimension, respectively. A two-dimensional saturation physical field in the time dimension is constructed based on the dataset of continuous oil saturation, water saturation and gas saturation of each well in the time dimension; wherein, the two-dimensional saturation physical field includes: a two-dimensional oil saturation physical field, a two-dimensional water saturation physical field and a two-dimensional gas saturation physical field. A two-dimensional pressure physical field in the time dimension is constructed based on the pressure value of each well in the time dimension; Based on the two-dimensional permeability physical field, two-dimensional porosity physical field, two-dimensional saturation physical field, and two-dimensional pressure physical field, a continuous dataset of each parameter in the two-dimensional planar physical field of the study block is determined in each time dimension.

4. The method according to any one of claims 1 to 3, characterized in that, The deep neural network model is pre-trained through the following steps: Obtain a training sample set, wherein each sample in the training sample set includes a dynamic and static parameter data structure for a preset time window and an oil saturation distribution of a two-dimensional planar physical field; The deep neural network model, which includes a convolutional neural network and a long short-term memory network, is trained using samples from the training sample set. The two-dimensional planar parameter features output by the convolutional neural network are input into the long short-term memory network to estimate the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block.

5. The method according to any one of claims 1 to 3, characterized in that, After obtaining the saturation physical field distribution and / or pressure physical field distribution of the study area, the method further includes: evaluating the influence weights of each parameter field in the two-dimensional planar physical field on the deep neural network model based on the noise measurement method.

6. A method for training a machine learning model, characterized in that, include: Based on well logging data and dynamic production data of the study block, a dataset of continuous oil saturation, water saturation and gas saturation of each well in the study block in the time dimension was determined. The oil saturation, water saturation and gas saturation of each well in the study block are determined based on the secondary interpretation of the well logging data in the study block. Based on the oil saturation, water saturation and gas saturation of each well in the study block, and the dynamic production data of the study block, a corresponding relationship between the oil saturation, water saturation and gas saturation of each well in the study block in the time dimension is constructed to determine the dataset of the oil saturation, water saturation and gas saturation of each well in the study block in the time dimension. Based on the well logging data of the study block, the pressure values ​​in the dynamic production data, and the continuous data sets of oil saturation, water saturation, and gas saturation of each well in the time dimension, the continuous data sets of each parameter in the two-dimensional planar physical field of the study block in each time dimension are determined. The initial saturation of each well in the study block is supplemented based on data from neighboring wells to unify the initial saturation of each well. Based on the supplemented initial saturation and the pressure values ​​of each well in the study block in the time dimension, as well as the static parameters in the logging data, a dynamic and static parameter data structure is constructed. After dividing the dynamic and static parameter data structure into preset time windows, the dynamic and static parameter data structure of the set time window and the saturation physical field distribution and / or pressure physical field distribution of the corresponding time window are used as a training sample to obtain a training sample set composed of training samples. The deep neural network model, which includes a convolutional neural network and a long short-term memory network, is trained using samples from the training sample set. The two-dimensional planar parameter features output by the convolutional neural network are input into the long short-term memory network to estimate the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block.

7. The method according to claim 6, characterized in that, Also includes: Based on the error function of pressure value and saturation data, the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block are adjusted.

8. A dynamic characterization device for oil reservoir development, characterized in that, include: The first determining module is used to determine the oil saturation, water saturation and gas saturation of each well in the study block based on the secondary interpretation of the well logging data of the study block; Based on the oil saturation, water saturation and gas saturation of each well in the study block, as well as the dynamic production data of the study block, a corresponding relationship between the oil saturation, water saturation and gas saturation of each well in the study block in the time dimension is constructed to determine the dataset of the oil saturation, water saturation and gas saturation of each well in the study block in the time dimension. The second determining module is used to determine the continuous dataset of each parameter in the two-dimensional planar physical field of the study block in each time dimension based on the well logging data of the study block, the pressure value in the dynamic production data, and the dataset of continuous oil saturation, water saturation and gas saturation of each well in the time dimension. The supplementary module is used to input the neighboring well data corresponding to the missing time of the target well within the study block into a pre-trained BP neural network model to obtain the saturation data of the target well at the missing time, so as to supplement the initial saturation of each well within the study block and unify the initial saturation of each well; wherein, the BP neural network model is pre-trained through the following steps: obtaining a training sample set, wherein each sample in the training sample set includes the target well saturation data and neighboring well saturation data in the same time interval; training the BP neural network model with the samples in the training sample set, wherein the neighboring well saturation data is input into the BP neural network model, and the target well saturation data is used as the output of the BP neural network model, and parameter estimation of the weight parameters and threshold parameters of the BP neural network model is performed; A construction module is used to construct a dynamic and static parameter data structure based on the supplemented initial saturation, the pressure values ​​of each well in the study block in the time dimension, and the static parameters in the logging data. The physical field distribution determination module is used to divide the dynamic and static parameter data structure into preset time windows, and then input the dynamic and static parameter data structure of the set time window into a pre-trained deep neural network model containing a convolutional neural network and a long short-term memory network. The two-dimensional planar parameter features output by the convolutional neural network are input into the long short-term memory network to obtain the saturation physical field distribution and / or pressure physical field distribution of the study block.

9. A training device for a machine learning model, characterized in that, include: The first acquisition module is used to acquire a training sample set, including: determining a dataset of continuous oil saturation, water saturation, and gas saturation for each well in the study block over a time dimension based on well logging data and dynamic production data of the study block; determining the oil saturation, water saturation, and gas saturation for each well in the study block based on secondary interpretation of well logging data of the study block; constructing a correspondence formula for continuous oil saturation, water saturation, and gas saturation for each well in the study block over a time dimension based on the oil saturation, water saturation, and gas saturation of each well in the study block, and the dynamic production data of the study block, to determine a dataset of continuous oil saturation, water saturation, and gas saturation for each well in the study block over a time dimension; and based on the well logging data of the study block, the... The dynamic production data includes pressure values ​​and continuous oil saturation, water saturation, and gas saturation data for each well over time. This data is used to determine continuous datasets of various parameters in the two-dimensional planar physical field of the study block across different time dimensions. Initial saturation data for each well within the study block is supplemented based on neighboring well data to unify the initial saturation of each well. Based on the supplemented initial saturation, pressure values ​​of each well within the study block over time, and static parameters from well logging data, a dynamic and static parameter data structure is constructed. This structure is then divided into preset time windows. The dynamic and static parameter data structure within each time window, along with the corresponding saturation physical field distribution and / or pressure physical field distribution, is used as a training sample to obtain a training sample set. Each sample in the training sample set includes a dynamic and static parameter data structure for a preset time window and an oil saturation distribution of a two-dimensional planar physical field. The first training module is used to train a deep neural network model containing a convolutional neural network and a long short-term memory network using samples from the training sample set. The two-dimensional planar parameter features output by the convolutional neural network are input into the long short-term memory network to estimate the parameters of the saturation physical field distribution and / or pressure physical field distribution of the study block.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the reservoir development dynamic characterization method as described in any one of claims 1 to 5, or the training method of the machine learning model as described in any one of claims 6 to 7.

11. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the reservoir development dynamic characterization method as described in any one of claims 1 to 5, or the training method for the machine learning model as described in any one of claims 6 to 7.

Images