Reservoir property parameter processing method and device, storage medium and program product

Abstract

The application provides a reservoir physical property parameter processing method, device, storage medium and program product, and relates to the technical field of oil exploration and geological research. Patch data of each source logging data in multi-source logging data of a target well is obtained, so that each patch data contains data of the corresponding source logging data at a set depth, wherein the multi-source logging data is obtained by a plurality of measuring devices based on different physical principles and facing the target well; the patch data of each source logging data is input into a reservoir physical property parameter processing model, feature extraction is performed respectively and mapping to the same feature space is performed, feature dimension alignment and spatiotemporal identification information are added in the feature space, and a spatiotemporal feature vector of the corresponding patch data is obtained; and a multi-head attention mechanism is used to perform operation based on the spatiotemporal feature vector corresponding to each source logging data, so as to obtain the reservoir physical property parameter of the target well, thereby realizing high-precision and high-reliability reservoir physical property parameter processing.

Description

Technical Field

[0001] This application relates to the fields of petroleum exploration and geological research technology, and in particular to a method, equipment, storage medium and program product for processing reservoir physical property parameters. Background Technology

[0002] In the fields of petroleum exploration and geological research, well logging technology is a crucial subsurface resource assessment technique. It obtains subsurface structural information by recording the formation's response to physical quantities such as electrical, acoustic, and radioactive signals. Well logging data, as a direct representation of these responses, plays an irreplaceable role in identifying formation characteristics and assessing oil and gas reservoir properties. Among these, reservoir physical parameters, including porosity, permeability, and saturation, are the core content of well logging interpretation. These parameters directly affect oil and gas resource reserve assessment, development plan formulation, and production efficiency optimization. High-precision reservoir physical parameters can not only improve the success rate of exploration and development but also effectively reduce development costs and enhance the economic benefits of oil and gas fields.

[0003] Traditional methods for calculating reservoir physical parameters, such as porosity (commonly using the Wyllie time-averaged equation or density-neutron cross-section method), permeability (based on the Kozeny-Carman equation or Coates permeability model), and saturation (using the Archie formula or Simandoux equation), often rely on empirical formulas and simplifying assumptions. These methods are ill-suited to complex and variable reservoir environments, such as heterogeneous or fractured formations, resulting in poor calculation accuracy. Therefore, a high-precision, high-reliability reservoir physical parameter processing technique is urgently needed. Summary of the Invention

[0004] This application provides a method, device, storage medium, and program product for processing reservoir physical parameters, which can achieve high-precision and high-reliability processing of reservoir physical parameters.

[0005] Firstly, this application provides a method for processing reservoir physical property parameters, including:

[0006] Obtain patch data for each source logging data in the multi-source logging data for the target well. Each patch data contains the corresponding source logging data at a set depth. Multi-source logging data is a variety of logging data obtained by measuring the target well with multiple measuring devices based on different physical principles.

[0007] The patch data corresponding to each source logging data is input into the reservoir physical parameter processing model. In the reservoir physical parameter processing model, the patch data corresponding to each source logging data is feature extracted and mapped to the same feature space. Feature dimension alignment and spatiotemporal identification information are added in the feature space to obtain the spatiotemporal feature vector of the corresponding patch data. Then, a multi-head attention mechanism is used to calculate the reservoir physical parameters of the target well based on the spatiotemporal feature vectors corresponding to each source logging data.

[0008] In one possible implementation, the reservoir property parameter processing model includes an encoding module, a fusion module, and a decoding module, wherein:

[0009] The encoding module is used to extract local features from the patch data corresponding to each source logging data and map them to the same feature space. In the feature space, feature dimension alignment is performed and spatiotemporal identification information is added to obtain the spatiotemporal feature vector of the corresponding patch data.

[0010] The fusion module is used to concatenate the spatiotemporal feature vectors of the corresponding source logging data in depth order to obtain the spatiotemporal feature sequence of the corresponding source logging data, and input the spatiotemporal feature sequence of each source logging data into the decoding module;

[0011] The decoding module is used to perform calculations on the spatiotemporal feature sequences corresponding to the multi-source logging data based on the spatiotemporal dependencies of each source logging data using a multi-head attention mechanism, in order to obtain the reservoir physical parameters of the target well.

[0012] In one possible implementation, the encoding module employs a residual network or a multilayer perceptron, and the decoding module employs a Transformer decoder structure.

[0013] In one possible implementation, patch data for each source logging data in the multi-source logging data for the target well is obtained, including:

[0014] Acquire multi-source logging data for the target well;

[0015] Preprocessing is performed on the logging data from each source in the multi-source logging data to obtain the target data of the corresponding source logging data. The preprocessing is used to eliminate the influence of dimensions and amplitude.

[0016] The target data corresponding to each source logging data is patched at equal intervals to obtain patch data for the corresponding source logging data.

[0017] In one possible implementation, a joint loss function is used when training the reservoir property parameter processing model. The joint loss function includes a first part and a second part. The first part is the model loss, and the second part is used to constrain and correct the model.

[0018] In one possible implementation, the first part calculates the mean square error; the second part calculates the loss based on the physical mechanism of reservoir properties, using a loss constraint and correction model.

[0019] Secondly, this application provides a reservoir physical property parameter processing apparatus, comprising:

[0020] The acquisition module is used to acquire patch data of each source logging data in the multi-source logging data for the target well. Each patch data contains the data of the corresponding source logging data at a set depth. The multi-source logging data is a variety of logging data obtained by measuring the target well with multiple measuring devices based on different physical principles.

[0021] The processing module is used to input the patch data corresponding to each source logging data into the reservoir physical parameter processing model. In the reservoir physical parameter processing model, the patch data corresponding to each source logging data is subjected to feature extraction and mapped to the same feature space. Feature dimension alignment and spatiotemporal identification information are added in the feature space to obtain the spatiotemporal feature vector of the corresponding patch data. Furthermore, a multi-head attention mechanism is used to perform calculations based on the spatiotemporal feature vectors corresponding to each source logging data to obtain the reservoir physical parameters of the target well.

[0022] In one possible implementation, the reservoir property parameter processing model includes an encoding module, a fusion module, and a decoding module, wherein:

[0023] The encoding module is used to extract local features from the patch data corresponding to each source logging data and map them to the same feature space. In the feature space, feature dimension alignment is performed and spatiotemporal identification information is added to obtain the spatiotemporal feature vector of the corresponding patch data.

[0024] The fusion module is used to concatenate the spatiotemporal feature vectors of the corresponding source logging data in depth order to obtain the spatiotemporal feature sequence of the corresponding source logging data, and input the spatiotemporal feature sequence of each source logging data into the decoding module;

[0025] The decoding module is used to perform calculations on the spatiotemporal feature sequences corresponding to the multi-source logging data based on the spatiotemporal dependencies of each source logging data using a multi-head attention mechanism, in order to obtain the reservoir physical parameters of the target well.

[0026] In one possible implementation, the encoding module employs a residual network or a multilayer perceptron, and the decoding module employs a Transformer decoder structure.

[0027] In one possible implementation, the acquisition module is specifically used for:

[0028] Acquire multi-source logging data for the target well;

[0029] Preprocessing is performed on the logging data from each source in the multi-source logging data to obtain the target data of the corresponding source logging data. The preprocessing is used to eliminate the influence of dimensions and amplitude.

[0030] The target data corresponding to each source logging data is patched at equal intervals to obtain patch data for the corresponding source logging data.

[0031] In one possible implementation, a joint loss function is used when training the reservoir property parameter processing model. The joint loss function includes a first part and a second part. The first part is the model loss, and the second part is used to constrain and correct the model.

[0032] In one possible implementation, the first part calculates the mean square error; the second part calculates the loss based on the physical mechanism of reservoir properties, using a loss constraint and correction model.

[0033] Thirdly, this application provides a reservoir physical property parameter processing device, including: a memory and a processor;

[0034] The memory stores the instructions that the computer executes;

[0035] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0036] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions that, when executed, are used to implement the first aspect and / or various possible embodiments of the first aspect.

[0037] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0038] The reservoir property parameter processing method, equipment, storage medium, and program product provided in this application acquire patch data of each source logging data in multi-source logging data for a target well. Each patch data contains data of the corresponding source logging data at a set depth. The multi-source logging data consists of various types of logging data obtained from multiple measurement devices targeting the target well, each based on different physical principles. The acquired patch data of each source logging data in the multi-source logging data for the target well is a local regional sequence, enabling the reservoir property parameter processing model to handle the temporal correlation between the multi-source logging data.

[0039] Furthermore, patch data corresponding to each source logging data are input into the reservoir property parameter processing model. Within this model, features are extracted from the patch data corresponding to each source logging data and mapped to the same feature space. Feature dimension alignment and spatiotemporal identification information are then added to the feature space to obtain spatiotemporal feature vectors containing both temporal and spatial information for each patch data. A multi-head attention mechanism is then employed to calculate the reservoir property parameters of the target well based on the spatiotemporal feature vectors corresponding to each source logging data. By fusing the attention mechanism within the reservoir property parameter processing model, the complex interaction relationships within each patch data corresponding to each source logging data and between it and other patch data are captured. Long-distance dependency features in the depth-time dimension are extracted, achieving accurate characterization of formation dynamic evolution. Based on this accurate characterization, reservoir property parameters can be predicted, improving the computational accuracy and reliability of the reservoir property parameter processing method. Additionally, predicting reservoir property parameters based on the fusion of multi-source logging data can eliminate the limitations of manual feature engineering. Attached Figure Description

[0040] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0041] Figure 1 A schematic diagram of a scenario for the reservoir property parameter processing method provided in the embodiments of this application;

[0042] Figure 2 A flowchart illustrating a reservoir property parameter processing method provided in an embodiment of this application;

[0043] Figure 3 This is a schematic diagram of the multi-source logging data feature alignment process provided in the embodiments of this application;

[0044] Figure 4 This is a schematic diagram of the residual network structure provided in an embodiment of this application;

[0045] Figure 5 This is a schematic diagram of the structure of 2D tokenization of well logging data provided in the embodiments of this application;

[0046] Figure 6 This is a schematic diagram of the structure for modeling the association between tokens provided in an embodiment of this application;

[0047] Figure 7 This is a schematic diagram of the Transformer decoder provided in the embodiments of this application;

[0048] Figure 8 A schematic diagram illustrating the prediction of reservoir physical property parameters provided in the embodiments of this application;

[0049] Figure 9a A schematic diagram showing the comparison between porosity test results and standard values ​​provided in an embodiment of this application;

[0050] Figure 9b A schematic diagram showing the comparison between the permeability test results and the standard values ​​provided in the embodiments of this application;

[0051] Figure 9c A schematic diagram showing the comparison between water saturation test results and standard values ​​provided in an embodiment of this application;

[0052] Figure 10 A schematic diagram of the reservoir property parameter processing device provided in the embodiments of this application;

[0053] Figure 11 A schematic diagram of the reservoir property parameter processing device provided in the embodiments of this application.

[0054] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0055] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0056] The terms “first,” “second,” etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, products, or apparatus.

[0057] First, let me explain the terms used in this application:

[0058] Reservoir physical properties refer to a series of key indicators that quantitatively describe the physical properties of oil and gas reservoir rocks, such as porosity, permeability, and water saturation. These parameters directly determine the reservoir's ability to store fluids, such as oil, natural gas, and water, and its capacity to allow fluids to flow within it.

[0059] Traditional methods for calculating reservoir physical parameters, such as the Archie formula, the Wyllie time-averaged equation, and the Coates permeability model, typically rely on empirical formulas and simplifying assumptions, making them ill-suited to complex and variable reservoir environments. Furthermore, they suffer from limited computational accuracy, significant regional dependence, and the need for core calibration of formula parameters, requiring parameter readjustment when applied across basins, thus lacking universality. Machine learning-based methods use well logging curves as input features and physical parameters as labels to directly establish an end-to-end mapping model, learning the correlation patterns from input to output from the data. While machine learning-based methods improve computational accuracy to some extent, they heavily rely on high-quality feature engineering and extensive labeled data, neglecting the dynamic evolution of formations along depth-time, and exhibiting poor model interpretability, with black-box model outputs often contradicting geological mechanisms and lacking generalization ability. Therefore, existing reservoir parameter calculation methods suffer from low computational accuracy and a lack of reliability.

[0060] To overcome the limitations of existing reservoir parameter calculation methods, this application provides a reservoir physical property parameter processing method that integrates attention mechanisms and physical mechanisms. By combining the automatic extraction capability of key features through attention mechanisms with the in-depth understanding of reservoir characteristics through physical mechanisms, the accuracy and reliability of physical property parameter calculations are effectively improved, while enhancing the interpretability and generalization ability of the model. Specifically, by patching multi-source logging data and using attention mechanisms to capture long-distance dependencies in depth, dynamic evolution correlation between formation depth and time is achieved. By automatically weighting key spatiotemporal features of each source logging data, dynamic modeling of temporal (sequence) and spatial (variable) alignment of multi-source logging data and multivariable spatiotemporal dependencies is achieved, replacing manual feature engineering. By embedding a physical mechanism model, the calculation results are ensured to conform to geophysical laws, enhancing the model's interpretability and generalization ability.

[0061] Specifically, firstly, multi-source logging data, including gamma, resistivity, acoustic, density, and neutron data, are patched at equal intervals, dividing the data into multiple local regions along the depth direction while preserving depth-domain location information. A specially designed residual network is used to learn the representation of the patched data, mapping the extracted features to a high-dimensional feature space, and aligning the feature dimensions within this space. A reservoir property parameter processing model based on a Transformer decoder structure is constructed, using a multi-head attention mechanism to capture the complex interactions between local regions corresponding to the patched data and between local and global features, thereby achieving accurate modeling of the dynamic evolution correlation between formation depth and time. Secondly, key features are automatically identified and weighted to ensure data alignment at different time points and spatial locations, effectively capturing the spatiotemporal dependencies between multiple variables, thus avoiding the cumbersome and limited nature of traditional manual feature engineering. Furthermore, regarding model constraints, physical mechanism models such as the Kozeny-Carman equation and the Simandoux equation are embedded into the calculation framework of the loss function as regularization terms. Known geophysical laws are used to constrain the calculation process, ensuring that the calculation results are not only accurate but also consistent with actual geological conditions, significantly improving the model's interpretability. Through these technical means, this application can significantly improve the accuracy, reliability, and versatility of reservoir physical parameter processing, providing a more efficient and scientific tool for petroleum exploration and geological research.

[0062] Figure 1 This is a schematic diagram of a scenario for the reservoir property parameter processing method provided in an embodiment of this application. Figure 1 As shown, in the specific application scenario of this application, firstly, core experimental data and multi-source logging data of the target well are acquired. These data are then preprocessed, and the preprocessed data is input into a pre-trained reservoir property parameter calculation model, outputting high-precision reservoir property parameter curves. Secondly, integrating geological and seismic interpretation, high-quality reservoirs are identified through high-precision porosity and permeability calculations. The relevant data of these high-quality reservoirs are then input as "hard data" into the three-dimensional reservoir attribute model to guide decision-making, such as reserve calculation, well location optimization, and development scheme design. This three-dimensional reservoir attribute model includes a three-dimensional porosity model, a three-dimensional permeability model, and a three-dimensional water saturation model. The aforementioned property parameter calculation model can be the reservoir property parameter processing model provided in the reservoir property parameter processing method of this application embodiment.

[0063] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0064] Figure 2 This is a schematic flowchart of a reservoir property parameter processing method provided in an embodiment of this application. Figure 2 As shown, the method for processing reservoir physical property parameters includes:

[0065] S201. Obtain patch data of each source logging data in the multi-source logging data for the target well. Each patch data contains the data of the corresponding source logging data at a set depth. The multi-source logging data is a variety of logging data obtained by measuring the target well with multiple measuring devices. The multiple measuring devices perform the measurement based on different physical principles.

[0066] This step acquires multi-source logging data from the target well, including data on gamma ray, resistivity, acoustic waves, density, and neutrons. Multi-source logging data refers to various types of logging data collected from the same depth range of the same target well using measuring equipment based on different physical principles; hence, it consists of multiple sources.

[0067] Patch data is obtained by dividing the multi-source logging data into multiple local regions along the depth direction. Each patch data point contains data from the corresponding source logging data within a set depth, i.e., a depth range, and retains the depth domain location information. This set depth can be a pre-set fixed depth or a depth adapted to the target well; it can be adjusted according to the actual situation.

[0068] It should be noted that patch data is generated by converting corresponding source logging data characterized by continuous logging curves into local regional sequences, facilitating the processing of time-series correlations by subsequent reservoir property parameter processing models. Due to the vertical sequentiality and local correlation of subsurface strata, rock properties such as porosity and permeability at a certain depth are not solely determined by the measured value at that point, but strongly depend on the local geological environment formed by the adjacent strata above and below. Therefore, it is necessary to capture their local context and patterns. Patch processing with equal-length intervals transforms continuous depth sequences into discrete, processable units. Subsequent reservoir property parameter processing models, based on a multi-head attention mechanism, automatically learn the dependencies between patch data, effectively modeling depth-span correlations.

[0069] S202. Input the patch data corresponding to each source logging data into the reservoir physical parameter processing model. In the reservoir physical parameter processing model, extract features from the patch data corresponding to each source logging data and map them to the same feature space. Align feature dimensions and add spatiotemporal identification information in the feature space to obtain the spatiotemporal feature vector of the corresponding patch data. Then, use a multi-head attention mechanism to perform calculations based on the spatiotemporal feature vectors corresponding to each source logging data to obtain the reservoir physical parameters of the target well.

[0070] This step uses patch data from various source logging data as input to the reservoir physical parameter processing model, where the reservoir physical parameters are pre-trained models. After inputting the patch data into the reservoir physical parameter processing model, the model extracts local features from each patch data and maps them to the same feature space, aligning the features in the spatial dimension to ensure that the feature vector dimensions of the abstract features extracted from different depths and layers are fixed and consistent.

[0071] Spatiotemporal identification information, such as time ID and spatial ID, is added to the feature vectors corresponding to each patch data to obtain spatiotemporal feature vectors for each patch data. Based on the spatiotemporal feature vectors corresponding to each patch data, a multi-head attention mechanism is used to capture the spatiotemporal dependencies between multi-source logging data. These spatiotemporal dependencies reflect the complex interaction relationships within each patch data and between it and other patch data, i.e., the relationships between the logging curves of each source, and the reservoir physical parameters of the target well are predicted.

[0072] For example, multi-source logging data such as gamma, resistivity, acoustic, density, and neutron can be used to calculate reservoir porosity. After calculating the porosity, the logging data such as gamma, resistivity, acoustic, density, and neutron can be integrated to calculate reservoir physical parameters such as permeability and water saturation.

[0073] The reservoir property parameter processing method provided in this application obtains patch data of each source logging data in the multi-source logging data for a target well. Each patch data contains data of the corresponding source logging data at a set depth. Through patching processing, continuous multi-source logging data is converted into a local regional sequence so that the reservoir property parameter processing model can handle the temporal correlation between the multi-source logging data. In addition, the patch data corresponding to each source logging data is input into the reservoir property parameter processing model. In the reservoir property parameter processing model, features are extracted from the patch data corresponding to each source logging data and mapped to the same feature space. Feature dimension alignment and spatiotemporal identification information are added in the feature space to obtain the spatiotemporal feature vector containing time and space information of the corresponding patch data. Finally, a multi-head attention mechanism is used to perform calculations based on the spatiotemporal feature vectors corresponding to each source logging data to obtain the reservoir property parameters of the target well. By integrating an attention mechanism into the reservoir property parameter processing model, the complex interaction relationships within each patch data corresponding to each source logging data and between it and other patch data are captured. Long-distance dependency features in the depth-time dimension are extracted to achieve accurate characterization of formation dynamic evolution. Based on this accurate characterization, reservoir property parameters are predicted, thereby improving the computational accuracy and reliability of the reservoir property parameter processing method.

[0074] Based on the above embodiments, the reservoir property parameter processing model includes an encoding module, a fusion module, and a decoding module, wherein:

[0075] The encoding module is used to extract local features from the patch data corresponding to each source logging data and map them to the same feature space. In the feature space, feature dimension alignment is performed and spatiotemporal identification information is added to obtain the spatiotemporal feature vector of the corresponding patch data.

[0076] The fusion module is used to concatenate the spatiotemporal feature vectors of the corresponding source logging data in depth order to obtain the spatiotemporal feature sequence of the corresponding source logging data, and input the spatiotemporal feature sequence of each source logging data into the decoding module;

[0077] The decoding module is used to perform calculations on the spatiotemporal feature sequences corresponding to the multi-source logging data based on the spatiotemporal dependencies of each source logging data using a multi-head attention mechanism, in order to obtain the reservoir physical parameters of the target well.

[0078] Optionally, the encoding module is based on a residual network, and the decoding module adopts the Transformer decoder structure. The residual network, by introducing a residual learning mechanism, can effectively solve the gradient vanishing problem in deep network training, improving the training efficiency and performance of the model. It is important to note that the Transformer decoder structure itself does not concern itself with the order of the input sequence by default. However, for well logging data, spatial information, i.e., depth order, and temporal information, i.e., depositional order, are crucial. Therefore, the reservoir physical parameters must be explicitly told the model the "location" of each data point. Thus, the essence of adding spatiotemporal identification information is to create a location code for each data point, which uniquely identifies the data point's position in space and time.

[0079] For example, Figure 3 This is a schematic diagram illustrating the feature alignment process for multi-source logging data provided in an embodiment of this application. For example... Figure 3 As shown, the patch data corresponding to each source logging data, in this example, density logging (i.e., density logging data), sonic logging (i.e., sonic logging data), resistivity logging (i.e., resistivity logging data), and porosity (i.e., porosity logging data), are input into the reservoir physical property parameter processing model. Figure 4 This is a schematic diagram of the residual network structure provided in an embodiment of this application. Figure 4As shown, the residual network consists of multiple structures including activation functions, linear layers, regularization modules, skip connections, and normalization modules. The residual network first enhances the nonlinear expressive power of the patch data through activation functions, then extracts local features from each patch data point through linear layers, and finally maps each patch data point to a high-dimensional feature space through the residual network. For example, if the patch data is 8-dimensional, the high-dimensional feature space is 1024-dimensional. The regularization module is used to prevent overfitting and improve generalization ability. The feature representation is further enhanced by introducing residual learning. Furthermore, the normalization module aligns the feature dimensions in the high-dimensional feature space, ensuring that feature vectors of abstract features extracted from different depths and levels can be received and processed by subsequent modules in a consistent and efficient manner.

[0080] Additionally, feature vectors are generated for each patch data. In this example, density logging corresponds to vector V1, sonic logging to vector V2, resistivity logging to vector V3, and porosity logging to vector V4. Vectors V1, V2, V3, and V4 all have the dimension `out_dim`, and spatiotemporal identifiers, such as time ID and spatial ID, are added to each to obtain the spatiotemporal feature vectors for the corresponding patch data. As an example, the time ID indicates the position of the patch data in the depth sequence; "0" indicates the first patch, "1" indicates the second patch, and so on. The spatial ID indicates the logging curve type; for example, density logging has a spatial ID of 0, sonic logging has a spatial ID of 1, resistivity logging has a spatial ID of 2, and porosity logging has a spatial ID of 3.

[0081] Fusion module ( Figure 3 (Not shown) The spatiotemporal feature vectors of each patch data are arranged in depth order, that is, the time IDs are arranged from shallow to deep, to form a sequence, which is used as the input of the decoding module.

[0082] The decoding module employs a multi-head attention mechanism to calculate the relationship between patch data within the spatiotemporal feature sequence, thereby understanding the longitudinal evolution pattern of the entire target well.

[0083] Alternatively, the encoding module is based on a multilayer perceptron, and the decoding module adopts a Transformer decoder structure.

[0084] The patch data is processed by the encoding and fusion modules to generate a 2DToken matrix. This 2DToken matrix can include multiple curves, and each curve can be decomposed into multiple tokens.

[0085] Common attention mechanisms are used to handle 1D token sequences, where tokens are only temporally related, i.e., sequentially related. However, in reservoir property parameter calculations, there is a correlation between multi-source logging curves and expert-annotated property parameter curves, known as spatial correlation, and temporally related between tokens of the same measurement item. For example, porosity can be calculated from acoustic and density curves. Figure 5 This is a schematic diagram illustrating the 2D tokenization of well logging data provided in an embodiment of this application. Figure 5 As shown, the acoustic wave curve, density curve, and porosity curve, after being processed by the encoding and fusion modules, are converted into a 2D token matrix. This matrix contains three curves. Taking the porosity, density, and acoustic wave curves as examples, each curve is decomposed into three tokens, which are interconnected and have spatiotemporal dependencies. The acoustic wave curve is decomposed into tokens 0-2, the density curve into tokens 3-5, and the porosity curve into tokens 6-8.

[0086] It should be noted that the example matrix may contain multiple curves, and each curve may be decomposed into multiple tokens.

[0087] Table 1

[0088] Tk_0 0 0 0 0 0 0 1 1 1 Tk_1 0 0 0 0 0 0 0 1 1 Tk_2 0 0 0 0 0 0 0 0 1 TK_3 0 0 0 0 0 0 1 1 1 Tk_4 0 0 0 0 0 0 0 1 1 Tk_5 0 0 0 0 0 0 0 0 1 Tk_6 0 0 0 0 0 0 1 1 1 Tk_7 0 0 0 0 0 0 0 1 1 Tk_8 0 0 0 0 0 0 0 0 1

[0089] Figure 6 This is a schematic diagram illustrating the modeling and association between tokens provided in an embodiment of this application. For example... Figure 6 As shown, this example calculates porosity by combining density curves and acoustic curves. This is a typical covariate scenario where the correlation between variables is unbalanced. An correlation matrix M is created, where 1 in the correlation matrix indicates that the correlation requires attention mechanism computation, and 0 indicates that the correlation does not require attention mechanism computation. The correlation matrix M is shown in Table 1.

[0090] The correlation matrix is ​​a matrix based on geologically prior definitions of interaction rules between variables. In the correlation matrix, 1 indicates that an association requires attention mechanism computation, and 0 indicates that no association requires attention mechanism computation. During attention calculation, the correlation matrix forces irrelevant positions to have a weight of 0. By constructing the correlation matrix, the attention interaction between different patches is controlled, modeling the dependencies between space and depth. For example, the density curve patch only interacts with the acoustic curve patch, i.e., its corresponding position in the M matrix is ​​1; the gamma curve patch does not participate in permeability calculation, i.e., its corresponding position in the M matrix is ​​0. In self-attention computation, the M matrix forces irrelevant positions to have an attention weight of 0.

[0091] For example, Tk_6 is only related to Tk_0 and TK_3, and not to other tokens.

[0092] Figure 7 This is a schematic diagram of the decoder structure of the Transformer provided in an embodiment of this application. Figure 7 As shown, the main architecture of the Transformer decoder includes a linear projection module, a space-time attention module, a normalization module, a feedforward module, an activation function, and skip connections. The projection module projects the spatiotemporal feature sequence obtained from the fusion module through a linear layer into a query (Q), key (K), and value (V) matrix, i.e.:

[0093] Q = XW Q

[0094] K = XW K

[0095] V = XW V

[0096] Where X is the spatiotemporal feature sequence of the input decoding module, i.e., the input of the Transformer decoder, and W... Q W K W V These are learnable parameters. Q, K, and V change dynamically according to the input sequence.

[0097] The linear projection module inputs the Q, K, and V matrices into the spatiotemporal attention module. The multi-head attention mechanism within the spatiotemporal attention module performs operations on the Q, K, and V matrices. An attention mask mechanism is used to determine whether attention concentration operations are performed between tokens in the spatiotemporal feature sequence. The expression for this is:

[0098]

[0099] Where Q is the query matrix, K is the key matrix, V is the value matrix, and d k K is the dimension; T represents the matrix transpose operation; M is the incidence matrix, and Mask(M) is the mask of the incidence matrix.

[0100] The values ​​in the mask (M) are carefully set. For locations that need to be observed, the mask value is 0. For locations that need to be masked, the mask value is a very large negative number.

[0101] Key features are extracted through steps such as normalization module, residual connection, and feedforward module. Then, the linear projection module and activation function output the correlation weights between patches to control the range of attention interaction between patches.

[0102] The temporal-spatial attention weights calculated by the multi-head attention mechanism are multiplied with the input of the Transformer decoder, i.e., the spatiotemporal feature sequence. This multiplication refers to matrix multiplication, and the key features are weighted and summed. The predicted physical property parameters, such as porosity-related key features like acoustic waves and density, are then weighted to automatically identify and weight key features in the multi-source logging data, ensuring data alignment at different time points and spatial locations.

[0103] The weighted feature representation is passed through a linear projection layer, such as a fully connected layer, to obtain the output of the physical property parameter prediction model.

[0104] Based on the above embodiments, patch data for each source logging data in the multi-source logging data for the target well is obtained, including:

[0105] Acquire multi-source logging data for the target well;

[0106] Preprocessing is performed on the logging data from each source in the multi-source logging data to obtain the target data of the corresponding source logging data. The preprocessing is used to eliminate the influence of dimensions and amplitude.

[0107] The target data corresponding to each source logging data is patched at equal intervals to obtain patch data for the corresponding source logging data.

[0108] Specifically, various measuring devices are prepared, including natural gamma ray logging tools, resistivity logging tools, sonic logging tools, density logging tools, and neutron logging tools. These devices are lowered into the target well, and data is collected from the wellhead at specific sampling intervals until the bottom of the well. The acquired multi-source logging data includes depth information and corresponding logging curve data, such as resistivity values, sonic velocity values, formation density values, and neutron porosity values.

[0109] Outlier detection and removal are performed on the logging data from each source in the multi-source logging dataset. For example, for natural gamma ray logging data, outliers are identified and removed using the 3σ principle. Missing data points are filled using interpolation methods. The logging data from each source in the multi-source logging dataset are standardized to make different types of logging data comparable. For example, resistivity logging data and sonic logging data are standardized to the [0,1] interval, respectively.

[0110] Based on the geological data and previous analysis of the target well, multiple depth intervals are defined. For example, according to pre-set length intervals, the target data corresponding to each source logging data is patched at equal length intervals to divide the depth range of the target layer into multiple local regions. That is, from the starting depth to the ending depth, the standardized source logging data is divided into fixed-length patches, such as 8 patches, preserving local features and depth location information. For each depth interval, the logging curve data corresponding to the selected source logging data within that interval is extracted.

[0111] Repeat the above steps to extract patch data for the depth range.

[0112] Furthermore, the reservoir property parameter processing model is obtained through the following methods:

[0113] (1) Obtain existing core data and expert-annotated physical property data to construct a training set. This includes: multi-source logging data, such as gamma, resistivity, acoustic wave, density, and neutron; and physical property data that have been corrected by experts, such as porosity, permeability, and water saturation.

[0114] For example, physical property data, including porosity, permeability, and water saturation, are obtained from existing wells that have been expert-calibrated and labeled. Simultaneously, multi-source logging data, including natural gamma, resistivity, sonic, density, and neutron logging data, are obtained from existing wells.

[0115] (2) The multi-source logging data and the physical property parameter data annotated by experts are cleaned and standardized, and divided into multiple patches along the depth direction. Each patch contains the data of each multi-source logging curve and physical property parameter curve in the same depth range.

[0116] For example, outlier detection and removal are performed on multi-source logging data and expert-annotated physical property parameter data. The multi-source logging data is then standardized to ensure uniform dimensions and amplitude.

[0117] Based on existing well geological data and preliminary analysis, multiple depth intervals are defined, such as eight. Multi-source logging data and expert-annotated physical property parameter data are divided into fixed-length patches, such as eight patches. For each depth interval, selected multi-source logging data and expert-annotated physical property parameter data within that interval are extracted to form a patch. Each patch contains data from all multi-source logging curves and expert-annotated physical property parameter curves within the same depth interval.

[0118] Repeat the above steps to extract patch data for the depth range.

[0119] (3) Input the patch data into the reservoir physical parameter processing model, train the fusion model using the training set, and optimize the model parameters using the joint loss function. At the same time, introduce an adaptive learning mechanism to dynamically adjust the model parameters according to different reservoir environments, and finally obtain the trained reservoir physical parameter processing model.

[0120] For example, the patch data is input into the reservoir physical parameter processing model, and the patch data is processed based on the multi-head attention mechanism to obtain the reservoir physical parameters of the target well.

[0121] As one possible implementation, a joint loss function is used when training the reservoir property parameter processing model. The joint loss function consists of a first part and a second part. The first part is used to calculate the model loss, and the second part is used to constrain and correct the model.

[0122] Specifically, a hybrid modeling approach is adopted to construct a joint loss function, and the model parameters are optimized based on the loss function to train the final reservoir physical property parameter processing model.

[0123] The joint loss of the hybrid model is:

[0124] minLoss = Loss Attention +βLoss f(x)

[0125] Where β is an adjustable parameter, β∈[0,1].

[0126] In one implementation, the first part calculates the mean square error; the second part calculates the loss based on the physical mechanism of reservoir physical parameters, using a loss constraint and correction model.

[0127] In the first part, the loss of the data model is obtained by calculating the mean squared error (MSE) between the output of the reservoir physical property parameter processing model and the expert-annotated physical property parameter data, that is:

[0128] Loss Attention =MSE Loss

[0129] In the second part, the physical mechanism model is as follows:

[0130] f(x) = f(x1,x2,…x) n )

[0131] Where x1, x2, ... x n These are variables related to the calculated reservoir physical properties. For example, porosity calculation is related to density and acoustic properties.

[0132] Loss corresponding to f(x) f(x)The Minimum Segregation of Porosity (MSE) is the difference between the model output, such as the calculated porosity, and the value obtained by directly calculating using existing formulas. For example, from raw input data, such as density and acoustic waveforms, density porosity and acoustic porosity are calculated using physical formulas, and porosity is calculated using existing formulas. Minimizing this MSE serves to constrain and correct the data model. f(x) The calculation formula is:

[0133] Loss f(x) =MSE(model output - f(x))

[0134] For porosity, the physical mechanism adopts density porosity. Harmony and acoustic porosity Integration methods:

[0135]

[0136] Where k1+k2=1.

[0137] Density Porosity The calculation is performed by measuring the bulk density of the formation, using the following formula:

[0138]

[0139] Where, ρ b ρ is the total formation density directly measured by a density logging instrument. f ρ is the pore fluid density. ma This represents the density of the rock skeleton.

[0140] acoustic porosity The speed of sound waves propagating through the earth's strata can be calculated using time difference, which can be achieved through the Wyllie time averaging formula:

[0141]

[0142] Where Δt is the sound wave time difference, Δt f Δt represents the porosity fluid travel time. ma This refers to the time difference of the rock skeleton.

[0143] For permeability, the physical mechanism is estimated through empirical or existing equations based on parameters such as porosity, specifically using the Kozeny-Carman equation:

[0144]

[0145] Where k is the absolute penetration rate. S is porosity, S is specific surface area, and C is the Kozeny constant.

[0146] For water saturation, the physical mechanism is estimated using a saturation model for argillaceous sandstone, which can be achieved through the Simandoux equation, typically a quadratic Simandoux equation:

[0147]

[0148] Among them, S w R represents the water saturation level. t R is the true resistivity of the formation. w R is the resistivity of formation water. sh V is the resistivity of mudstone. sh The content of clay, The effective porosity is given by , m is the cementation index, and a is the rock coefficient.

[0149] During model training, an adaptive learning mechanism is introduced to dynamically adjust model parameters according to different reservoir environments, enhancing the model's generalization ability. For example, a tradeoff coefficient β is set, with a value ranging from [0,1]. When β→0, it means the reservoir property parameter processing model trusts the data itself more. In areas with high data quality and reliable labels, such as areas with dense core calibration, the reservoir property parameter processing model is allowed to learn freely from the data, even discovering patterns beyond physical formulas. When β→1, it means the reservoir property parameter processing model trusts physical mechanisms more. In areas with sparse data, high noise, or complex conditions, such as fractured regions, physical laws are used to constrain and guide the model, preventing it from making absurd predictions that violate physical common sense.

[0150] It should be noted that β is not a fixed hyperparameter, but a dynamic variable related to the input data, i.e., the current predicted depth points or data samples.

[0151] Reservoir confidence can be determined in the following ways:

[0152] High-confidence regions, such as areas with dense core calibration, can have their confidence level determined by the density of core data. For example, if the number of core samples within a certain depth range exceeds a specific number, that area can be considered a high-confidence region. Low-confidence regions, such as fracture-developed sections, can be identified using fracture identification algorithms, such as machine learning-based fracture detection models. For example, if the fracture probability of a fracture identification algorithm within a certain depth range exceeds a specific probability, that area can be considered a low-confidence region. Depth ranges other than high-confidence and low-confidence regions can be considered intermediate-confidence regions.

[0153] In the early stages of model training, the β parameter is initialized to a small value, such as 0.1, so that the model learns data-driven features first. During each training epoch, the β parameter is dynamically adjusted based on the confidence level of the current depth interval. Specifically:

[0154] In the high-confidence region, as the model trains to this level, the parameter β is gradually decreased (e.g., β→0) to minimize the MSE loss and improve the model's prediction accuracy. In the low-confidence region, as the model trains to this level, the β parameter is gradually increased (e.g., β→1) to minimize the loss of the physical mechanism model and improve the model's generalization ability. In the medium-confidence region, the β parameter is kept at a moderate value, such as 0.5, to balance the MSE loss and the loss of the physical mechanism model.

[0155] For example, the adjustment strategy for the β parameter can be implemented in the following two ways:

[0156] The first adjustment method is to adjust the β parameter based on the confidence level of the current depth interval during each training cycle. For example, if the current depth interval is a high-confidence region, decrease the β parameter; if the current depth interval is a low-confidence region, increase the β parameter.

[0157] The second adjustment method is to adjust the rate of change of the β parameter appropriately during training as the model gradually converges. For example, in the early stages of training, the rate of change of the β parameter can be larger so that the model can quickly adapt to regions with different confidence levels; in the later stages of training, the rate of change of the β parameter can be smaller so that the model can achieve a better balance between different regions.

[0158] Next, specific examples will be used to verify the application of the technology provided in this application for reservoir physical property parameter processing. For example, complete logging data from 30 wells were selected as training samples, in which experts have already estimated and core-corrected the reservoir physical property parameters. Logging data from another 10 wells were used as test samples for model training and testing. The model parameters are as follows:

[0159] The parameters for the prediction task are as follows:

[0160] seq_len = 48, so the viewport size is 48.

[0161] patch_len = 8 The patch length is 8.

[0162] label_len = 24, the prior sequence length is 24.

[0163] pred_len = 8, the predicted sequence length is 8.

[0164] Figure 8 This is a schematic diagram illustrating the prediction of reservoir physical property parameters provided in an embodiment of this application. For example... Figure 8As shown, the data from the top 48 data points is used to predict the data for the next 8 data points. When predicting any data point, the reservoir property parameter processing model uses the current data point as a reference and looks back at 48 data points as input context. These 48 data points contain the historical information, patterns, and trends required for the reservoir property parameter processing model to make predictions. Furthermore, the reservoir property parameter processing model does not predict only one data point at a time, but rather predicts the next 8 data points all at once. The patch data is 8 data points in size (equivalent to 1 meter of logging data).

[0165] The residual network maps the patch data to a 1024-dimensional feature space, with the following parameters:

[0166] c_in = 8, where the input is the data for a patch.

[0167] hidden_dim = 512, the feature dimension of the hidden layer is 512.

[0168] = out_dim=1024 outputs a 1024-dimensional feature space.

[0169] The parameters of the Transformer decoder are as follows:

[0170] d_model = 1024, the feature dimension of the input transformer is 1024.

[0171] e_layers=4 indicates four Decoder Layers

[0172] n_heads = 8, the number of multi-heads is 8.

[0173] d_ff = 2048, the feature dimension of the feed-forward layer.

[0174] The reservoir property parameter processing model and the parameters defined above were used to train the model using logging data from 30 wells, totaling 88,350 data records. Data from another 10 wells, totaling 22,964 test data records, were also used. The porosity, permeability, and water saturation calculated using this method were compared with the core-corrected results to obtain the following... Figures 9a-9c The results show a comparison between the reservoir physical property parameter test results and the standard values. Figure 9a This is a schematic diagram comparing the porosity test results with standard values ​​provided in an embodiment of this application. Figure 9aAs shown, the horizontal axis corresponds to the well depth, with the unit being meters (m), and the vertical axis corresponds to porosity, with the unit being percentage (%). Here, POR represents the standard porosity value, and POR_Pre represents the porosity test result. For example, at a depth of 3100m, the corresponding porosity test result is approximately 9.5%, indicating that the pore volume accounts for 9.5% of the total volume. Figure 9b This is a schematic diagram comparing the permeability test results with standard values ​​provided in an embodiment of this application. Figure 9b As shown, the horizontal axis corresponds to the well depth, with the unit being meters (m), and the vertical axis corresponds to permeability, with the unit being percentage (%). Here, PERM represents the standard permeability value, and PERM_Pre represents the permeability test result. For example, at a depth of 3052m, the corresponding permeability test result is approximately 3.4%, indicating that the permeated volume accounts for 3.4% of the total volume. Figure 9c This is a schematic diagram comparing the water saturation test results with standard values ​​provided in an embodiment of this application. Figure 9c As shown, the horizontal axis corresponds to the well depth, with the unit being meters (m), and the vertical axis corresponds to the water saturation, with the unit being percentage (%). Here, SW represents the standard value of water saturation, and SW_Pre represents the water saturation test result. For example, at a depth of 3051m, the corresponding water saturation test result is approximately 26%, indicating that the water-bearing volume accounts for 26% of the total volume.

[0175] like Figures 9a-9c As shown, the test results are highly consistent with the expert's predictions and meet the business requirements.

[0176] In addition, Table 2 illustrates the test results of reservoir property parameter processing using this reservoir property parameter processing model. It should be noted that the evaluation indicators differ depending on the production requirements. In property parameter production, greater emphasis is placed on testing the absolute mean error (MSE); therefore, the MSE indicator will not be discussed in detail here.

[0177] As shown in Table 2, the absolute average error of porosity in the test results reached 0.3422, which is less than the production requirement, and the calculated porosity can meet the production requirements; the absolute average error of permeability was 0.0331, which is less than the production requirement, and the calculated permeability can meet the production requirements; the absolute average error of water saturation was 1.3775, which is less than 5.0, and the water saturation can meet the production requirements.

[0178] Table 2

[0179]

[0180] In summary, the embodiments of this application have at least the following advantages:

[0181] First, by acquiring multi-source logging data from the target well, preprocessing and patching the logging data from each source within the multi-source logging data with equal-length intervals are performed to obtain patch data for each source of the multi-source logging data for the target well. The equal-length interval patching process converts continuous logging curves into local regional sequences, facilitating the processing of time-series correlations by the reservoir property parameter processing model and automatically learning the dependencies between patches.

[0182] Second, a reservoir property parameter processing model based on attention mechanisms and physical principles is constructed. Long-term dependencies along formation depth are established through patching and a multi-head attention mechanism, enabling dynamic correlation modeling. A dynamic evolution model of formation depth-time is established using the multi-head attention mechanism of the Transformer decoder, achieving accurate characterization of formation dynamics. This multi-head attention mechanism captures the complex interactions within each patch and between patches, extracting long-distance dependency features in the depth-time dimension while automatically weighting key features related to the target reservoir property parameters. This ensures data alignment at different time points and spatial locations, completing dynamic modeling of spatiotemporal relationships, eliminating the limitations of manual feature engineering, and further achieving high-precision calculations.

[0183] Third, during the training process of the reservoir property parameter processing model, a geophysical mechanism model is embedded. Known geophysical laws constrain the calculation process, ensuring that the calculation results are not only accurate but also consistent with actual geological conditions. This reduces reliance on regional calibration and enhances the model's interpretability and generalization ability. Simultaneously, an adaptive learning mechanism is introduced. Based on the input multi-source logging data, the model parameters are dynamically adjusted according to different reservoir environments, improving the cross-regional generalization ability of the reservoir property parameter processing model and ensuring its reliability.

[0184] Through the aforementioned technical means, this application not only overcomes the shortcomings of traditional methods and existing machine learning methods in the calculation of reservoir physical parameters, but also significantly improves the accuracy and reliability of the calculation, providing a more efficient and scientific tool for oil exploration and geological research.

[0185] Figure 10 A schematic diagram of the reservoir property parameter processing device provided in the embodiments of this application is shown below. Figure 10 As shown, the reservoir property parameter processing device 30 provided in this embodiment includes:

[0186] The acquisition module 301 is used to acquire patch data of each source logging data in the multi-source logging data for the target well. Each patch data contains the data of the corresponding source logging data at a set depth. The multi-source logging data is a variety of logging data obtained by measuring the target well with multiple measuring devices. The multiple measuring devices perform the measurement based on different physical principles.

[0187] The processing module 302 is used to input the patch data corresponding to each source logging data into the reservoir physical parameter processing model. In the reservoir physical parameter processing model, the patch data corresponding to each source logging data are subjected to feature extraction and mapped to the same feature space. Feature dimension alignment and spatiotemporal identification information are added in the feature space to obtain the spatiotemporal feature vector of the corresponding patch data. Furthermore, a multi-head attention mechanism is used to perform calculations based on the spatiotemporal feature vectors corresponding to each source logging data to obtain the reservoir physical parameters of the target well.

[0188] In one possible implementation, the reservoir property parameter processing model includes an encoding module, a fusion module, and a decoding module, wherein:

[0189] The encoding module is used to extract local features from the patch data corresponding to each source logging data and map them to the same feature space. In the feature space, feature dimension alignment is performed and spatiotemporal identification information is added to obtain the spatiotemporal feature vector of the corresponding patch data.

[0190] The fusion module is used to concatenate the spatiotemporal feature vectors of the corresponding source logging data in depth order to obtain the spatiotemporal feature sequence of the corresponding source logging data, and input the spatiotemporal feature sequence of each source logging data into the decoding module;

[0191] The decoding module is used to perform calculations on the spatiotemporal feature sequences corresponding to the multi-source logging data based on the spatiotemporal dependencies of each source logging data using a multi-head attention mechanism, in order to obtain the reservoir physical parameters of the target well.

[0192] In one possible implementation, the encoding module uses a residual network or a multilayer perceptron, and the decoding module uses a Transformer decoder structure.

[0193] In one possible implementation, the acquisition module 301 is specifically used for:

[0194] Acquire multi-source logging data for the target well;

[0195] Preprocessing is performed on the logging data from each source in the multi-source logging data to obtain the target data of the corresponding source logging data. The preprocessing is used to eliminate the influence of dimensions and amplitude.

[0196] The target data corresponding to each source logging data is patched at equal intervals to obtain patch data for the corresponding source logging data.

[0197] In one possible implementation, a joint loss function is used when training the reservoir property parameter processing model. The joint loss function consists of a first part and a second part. The first part is the model loss, and the second part is used to constrain and correct the model.

[0198] In one possible implementation, the first part calculates the mean square error; the second part calculates the loss based on the physical mechanism of reservoir properties, using a loss constraint and correction model.

[0199] The reservoir property parameter processing device provided in this application embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0200] Figure 11 This is a schematic diagram of a reservoir property parameter processing device provided in an embodiment of this application. Figure 11 As shown in the embodiment of this application, the reservoir property parameter processing device 40 includes at least one processor 401 and a memory 402. Optionally, the reservoir property parameter processing device 40 further includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus.

[0201] In a specific implementation, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to perform the above-described method.

[0202] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0203] In the above embodiments, it should be understood that the processor 401 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0204] The memory 402 may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0205] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0206] This application also provides a reservoir physical property parameter processing system, including a measuring device and the aforementioned reservoir physical property parameter processing device. The measuring device can measure the target well based on different physical principles to obtain the aforementioned multi-source logging data. Optionally, the measuring device can be a single measuring device or multiple measuring devices. When the measuring device is a single measuring device, it supports different physical principles for obtaining multi-source logging data; when there are multiple measuring devices, each of these multiple measuring devices supports at least one physical principle for obtaining multi-source logging data.

[0207] This application also provides a computer program product, including a computer program that, when executed, implements the above-described method.

[0208] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed, implement the above-described method.

[0209] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0210] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0211] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0212] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0213] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0214] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0215] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0216] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for processing reservoir physical property parameters, characterized in that, include: Obtain patch data for each source logging data in the multi-source logging data for the target well. Each patch data contains data of the corresponding source logging data at a set depth. The multi-source logging data is a variety of logging data obtained by measuring the target well with multiple measuring devices based on different physical principles. Patch data corresponding to each source logging data are input into a reservoir physical parameter processing model. This model includes an encoding module, a fusion module, and a decoding module. In the encoding module, local features are extracted from the patch data corresponding to each source logging data and mapped to the same feature space. Feature dimension alignment and spatiotemporal identification information are added in the feature space to obtain the spatiotemporal feature vector of the corresponding patch data. The fusion module concatenates the spatiotemporal feature vectors of the corresponding source logging data according to depth order to obtain the spatiotemporal feature sequence of the corresponding source logging data. This spatiotemporal feature sequence is then input into the decoding module. Finally, the decoding module employs a multi-head attention mechanism to perform calculations on the spatiotemporal feature sequences corresponding to the multi-source logging data based on their spatiotemporal dependencies, thereby obtaining the reservoir physical parameters of the target well.

2. The method for processing reservoir physical property parameters according to claim 1, characterized in that, The encoding module uses a residual network or a multilayer perceptron, and the decoding module uses a Transformer decoder structure.

3. The method for processing reservoir physical property parameters according to claim 1 or 2, characterized in that, The process of acquiring patch data for each source logging data in the multi-source logging data for the target well includes: Acquire multi-source logging data for the target well; The well logging data from each source in the multi-source logging data are preprocessed to obtain the target data of the corresponding source logging data. The preprocessing is used to eliminate the influence of dimensions and amplitude. The target data corresponding to each source logging data is patched at equal intervals to obtain patch data for the corresponding source logging data.

4. The method for processing reservoir physical property parameters according to claim 1 or 2, characterized in that, When training the reservoir property parameter processing model, a joint loss function is used. The joint loss function includes a first part and a second part. The first part is the model loss, and the second part is used to constrain and correct the model.

5. The reservoir property parameter processing method according to claim 4, characterized in that, The first part calculates the mean square error; the second part calculates the loss based on the physical mechanism of reservoir properties, using a loss constraint and correction model.

6. A reservoir physical property parameter processing device, characterized in that, include: The acquisition module is used to acquire patch data of each source logging data in the multi-source logging data for the target well. Each patch data contains the data of the corresponding source logging data at a set depth. The multi-source logging data is a variety of logging data obtained by measuring the target well with a variety of measuring devices based on different physical principles. The processing module is used to input the patch data corresponding to each source logging data into the reservoir physical parameter processing model. The reservoir physical parameter processing model includes an encoding module, a fusion module, and a decoding module. In the reservoir physical parameter processing model, the encoding module extracts local features from the patch data corresponding to each source logging data and maps them to the same feature space. Feature dimension alignment and spatiotemporal identification information are added in the feature space to obtain the spatiotemporal feature vector of the corresponding patch data. The fusion module concatenates the spatiotemporal feature vectors of the corresponding source logging data according to depth order to obtain the spatiotemporal feature sequence of the corresponding source logging data, and inputs the spatiotemporal feature sequence of each source logging data into the decoding module. The decoding module uses a multi-head attention mechanism to calculate the spatiotemporal feature sequence corresponding to each source logging data based on the spatiotemporal dependency of each source logging data to obtain the reservoir physical parameters of the target well.

7. A reservoir physical property parameter processing device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the method as described in any one of claims 1 to 5.

9. A computer program product, characterized in that, Includes a computer program, which, when executed, implements the method of any one of claims 1 to 5.

Images