Digital twin oil and gas reservoir damage spatio-temporal evolution 4D intelligent prediction and diagnosis method and system

Abstract

The invention provides a digital twin oil and gas reservoir damage spatio-temporal evolution 4D intelligent prediction and diagnosis method and system, and belongs to the technical field of petroleum and natural gas engineering.The method comprises the steps that digital twin bodies corresponding to well section samples in a target reservoir are converted into corresponding multi-dimensional features, the multi-dimensional features are input into an intelligent model, and the multi-dimensional features are obtained; and determining a change sequence of the reservoir damage rate corresponding to the target reservoir in the future time period. According to the method, four dimensions of'time, depth, space and dynamic evolution 'are taken as unified modeling objects, a'structure-parameter-behavior' three-in-one digital twin body is constructed, and reservoir damage rate prediction and evolution monitoring are realized through a 4D dynamic mapping and model feedback mechanism. According to the embodiment of the invention, the digital twinborn system of the full life cycle is introduced into an oil and gas reservoir damage prediction scene for the first time, the prediction precision, the temporal-spatial resolution and the response capability are remarkably improved, and the method can be widely applied to intelligent protection and efficient development decision support of complex reservoirs.

Description

Technical Field

[0001] The present invention relates to the field of petroleum and natural gas engineering technology, and specifically to a 4D intelligent prediction and diagnosis method, system, device, processor and computer program product for the spatiotemporal evolution of digital twin oil and gas reservoir damage. Background Art

[0002] Oil and gas reservoir damage is a common and pressing issue in oil and gas field development. It primarily manifests as physical or chemical changes in the reservoir's pore structure, leading to decreased permeability and productivity. The causes of reservoir damage are complex, encompassing a variety of physical and chemical processes, such as solid particle migration, pore plugging, chemical precipitation reactions, clay mineral expansion, and interfacial emulsification. These processes exhibit significant spatial and temporal dynamics. As oil and gas field development progresses toward complex well structures, deeper formations, and reservoirs with low or ultra-low permeability, reservoir damage becomes increasingly prominent, posing a serious challenge to the efficient development and management of oil and gas fields. Traditionally, reservoir damage assessment relies on detailed laboratory analysis of reservoir cores, including physical and chemical properties, as well as mineral composition, to assess reservoir damage. Laboratory core analysis is typically conducted under controlled conditions, which differ from the temperature, pressure, and fluid conditions found in actual reservoirs. These discrepancies can lead to discrepancies between the analytical results and the actual conditions. Furthermore, for complex reservoirs, such as fractured or highly heterogeneous ones, core analysis struggles to accurately capture all reservoir characteristics, increasing uncertainty in damage prediction. In addition, the interpretation of core analysis results may be affected by the analyst's experience and knowledge and may be subjective.

[0003] In recent years, with the development of the Industrial Internet and edge computing, "digital twin" technology has been gradually applied to the state modeling and evolution prediction of complex systems. It has demonstrated excellent real-time mapping and control capabilities in fields such as aviation manufacturing and smart grids. In oil and gas engineering, digital twins can establish a multidimensional correspondence between the physical state of reservoirs and virtual models, enabling virtual mirroring, evolutionary deduction, and dynamic feedback of complex geological structures and development activities.

[0004] However, existing research on digital twins of oil and gas reservoirs has mostly focused on development process simulation (such as production capacity forecasting and pressure evolution) or equipment operation monitoring. It lacks a specialized modeling system for reservoir "damage evolution," a multi-physics, multi-spatiotemporal coupled process. In particular, in reservoir damage scenarios, micro-scale evolutionary processes such as pore-throat structural destruction, fluid property changes, and mineral interactions are difficult to accurately represent in macro-scale models. This leads to a disconnect between the "digital twin" and the "physical mechanisms," making it difficult to meet the precision requirements of dynamic prediction and development control.

[0005] Therefore, there is an urgent need for a "full-process, full-scale, full-state" digital twin system that integrates image reconstruction, parameter analysis, model learning and feedback mechanism to achieve high-precision visual modeling, intelligent prediction and dynamic update of oil and gas reservoir damage, and provide scientific support for subsequent reservoir protection and development optimization. Summary of the Invention

[0006] The purpose of the embodiments of the present invention is to provide a digital twin oil and gas reservoir damage spatiotemporal evolution 4D intelligent predictive diagnosis method, system, device, processor and computer program product to solve or at least partially solve the above-mentioned defects of the prior art.

[0007] To achieve the above objectives, a first aspect of an embodiment of the present invention provides a 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage, the method comprising: Converting the digital twins corresponding to the well section samples in the target reservoir into corresponding multidimensional features, and inputting the multidimensional features into the intelligent model to determine the change sequence of the reservoir damage rate corresponding to the target reservoir in the future time period; The digital twin corresponding to the well section sample includes three-dimensional structural features, multi-source input parameters, and historical time series data corresponding to the reservoir damage rate; the model parameters corresponding to the intelligent model are determined during the training process by minimizing the loss function, and the loss function is determined according to the following method: For each training data in the plurality of training data, the following processing steps are performed: inputting a multidimensional feature corresponding to the training data into a preset empirical model to determine an empirical value of a reservoir damage rate corresponding to the training data; and determining a deviation corresponding to the training data based on the empirical value and an observed value of the reservoir damage rate corresponding to the training data; Determining a penalty coefficient according to the deviation and hyperparameter corresponding to each training data in the plurality of training data; and A loss function is determined based on the penalty coefficient, the deviation corresponding to each training data, the observation value, and the model prediction value corresponding to the intelligent model.

[0008] Preferably, determining the deviation corresponding to each training data includes: According to the empirical value of the reservoir damage rate corresponding to each training data and observations , determine the deviation corresponding to each training data : .

[0009] Preferably, the penalty coefficient is determined by: determining the penalty coefficient according to the deviation corresponding to each training data in the training data. and hyperparameters , determine the penalty coefficient :

[0010] Wherein, N is the total number of training data required by the loss function during the training process.

[0011] Preferably, the loss function is determined by: And the deviation corresponding to each training data , the observed value And the model prediction value corresponding to the intelligent model Determine the loss function : .

[0012] Preferably, the preset empirical model is a multiple linear regression model, and the intelligent model is a light gradient boosting machine LightGBM.

[0013] Preferably, the training data is established based on the SMOTE artificial sampling technique.

[0014] Preferably, the modeling features corresponding to the intelligent model are determined according to the following method: Using multiple evaluation models to evaluate the importance of different feature combinations corresponding to the multidimensional features, and generating a feature importance ranking list corresponding to each evaluation model; Determining a comprehensive ranking of feature importance corresponding to each feature combination based on the feature importance ranking lists corresponding to the multiple evaluation models; and The feature combination corresponding to the minimum value of the comprehensive ranking of feature importance is used as the modeling feature.

[0015] Preferably, the multi-source input parameters include: lithology, burial depth, initial porosity, initial permeability, cementation type, montmorillonite content, illite content, chlorite content, kaolinite content, illite-montmorillonite mixed layer content, chlorite-montmorillonite mixed layer content, clay mineral content, quartz content, feldspar content and at least two of the formation water mineralization.

[0016] A second aspect of an embodiment of the present invention provides a digital twin oil and gas reservoir damage spatiotemporal evolution 4D intelligent prediction and diagnosis system, the system comprising: Feature extraction module: used to obtain the digital twins corresponding to the well section samples in the target reservoir and convert the digital twins corresponding to the well section samples in the target reservoir into corresponding multidimensional features; Prediction module: used for inputting the multi-dimensional features into the intelligent model to determine the change sequence of the reservoir damage rate corresponding to the target reservoir in the future time period; The digital twin corresponding to the well section sample includes three-dimensional structural features, multi-source input parameters, and historical time series data corresponding to the reservoir damage rate; the model parameters corresponding to the intelligent model are determined during the training process by minimizing the loss function, and the loss function is determined according to the following method: For each training data in the plurality of training data, the following processing steps are performed: inputting a multidimensional feature corresponding to the training data into a preset empirical model to determine an empirical value of a reservoir damage rate corresponding to the training data; and determining a deviation corresponding to the training data based on the empirical value and an observed value of the reservoir damage rate corresponding to the training data; Determining a penalty coefficient according to the deviation and hyperparameter corresponding to each training data in the plurality of training data; and A loss function is determined based on the penalty coefficient, the deviation corresponding to each training data, the observation value, and the model prediction value corresponding to the intelligent model.

[0017] The third aspect of an embodiment of the present invention provides a 4D intelligent predictive diagnosis device for the spatiotemporal evolution of digital twin oil and gas reservoir damage, comprising: a memory configured to store instructions; and a processor configured to call the instructions from the memory and to implement the 4D intelligent predictive diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage when executing the instructions.

[0018] A fourth aspect of an embodiment of the present invention provides a processor for running a program, wherein the program, when run, is used to execute: the 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage.

[0019] A fifth aspect of an embodiment of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage.

[0020] The 4D intelligent predictive diagnostic method for the spatiotemporal evolution of digital twin oil and gas reservoir damage, provided by this embodiment of the invention, firstly designs the penalty coefficient and deviation of the loss function based on the corresponding predicted values ​​of the intelligent model and the empirical model. This integrates knowledge in the field of reservoir damage into the intelligent model, forming a knowledge-guided intelligent framework. Secondly, by inputting multidimensional features into the intelligent model, it can simulate the evolution of damage during different development stages, such as waterflooding, fracturing, and acidizing, and optimize operational parameters. The intelligent model also provides real-time reservoir damage rate predictions, enabling real-time adjustments to development plans.

[0021] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the embodiments of the present invention, but do not constitute a limitation of the embodiments of the present invention. In the accompanying drawings: Figure 1 4D intelligent prediction and diagnosis method for spatiotemporal evolution of digital twin oil and gas reservoir damage provided by an embodiment of the present invention; Figure 2 is a schematic diagram of CT scan results provided by an embodiment of the present invention; Figure 3 Schematic diagram of NMR scanning results provided by an embodiment of the present invention; Figure 4 Schematic diagram of the maximum sphere algorithm provided by an embodiment of the present invention; Figure 5 is a schematic diagram of pore radius provided by an embodiment of the present invention; Figure 6 Schematic diagram of throat length provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following describes the specific implementation of the embodiment of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the embodiment of the present invention and is not used to limit the embodiment of the present invention.

[0024] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application are in compliance with the relevant provisions of laws and regulations. In the embodiments of this application, certain software, components, models, and other existing solutions in the industry may be mentioned. These should be considered as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of this application, but it does not mean that the applicant has or will necessarily use such solutions.

[0025] To overcome the bottlenecks of traditional reservoir damage analysis methods in terms of time response lag, insufficient spatial resolution, and missing prediction dimensions, this paper proposes a 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage. The four dimensions of "time, depth, space, and dynamic evolution" are used as unified modeling objects to construct a digital twin of the "structure-parameter-behavior" trinity, and reservoir damage rate prediction and evolution monitoring are achieved through 4D dynamic mapping and model feedback mechanism. Figure 1 This is a flow chart of the 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage provided by an embodiment of the present invention. Figure 1As shown, the method may include: converting a digital twin corresponding to a well section sample in a target reservoir into a corresponding multidimensional feature, and inputting the multidimensional feature into an intelligent model to determine a change sequence of a reservoir damage rate corresponding to the target reservoir in a future time period; Among them, the digital twin corresponding to the well section sample includes three-dimensional structural characteristics, multi-source input parameters and historical time series data corresponding to the reservoir damage rate.

[0026] In practical applications, for any well section sample in the target reservoir S i , and define its digital twin as:

[0027] in, M i Represents three-dimensional structural characteristics, including: pore radius , Roar length and equivalent pore-throat ratio . Pore radius It can be constructed through CT scanning and the maximum sphere algorithm, which is specifically achieved through the following formula: ; In the above formula, represents the entire coordinate space, x represents the scanning coordinates, represents the radius of the sphere corresponding to the scanning coordinate, represents the entire pore space.

[0028] In some embodiments, online computed tomography (CT) and nuclear magnetic resonance (NMR) technologies are first used to obtain the dynamic changes of the reservoir pore structure and fluid distribution corresponding to the target reservoir in real time. Then, an equivalent pore-fracture network model is constructed based on the maximum sphere algorithm. Finally, the three-dimensional structural characteristics are obtained by analyzing the constructed pore-fracture network model.

[0029] Specifically, for the target reservoir, each sample is scanned a total of 3,600 times. The scanning pixel is 200nm and the spatial resolution is 1μm. The main steps of X-ray CT scanning are: S1. Fix the coal sample to be scanned on the precision sample stage and turn on the X-ray source switch; S2. The high-resolution detector detects the X-rays absorbed and attenuated by the scanned sample; S3. The computer software automatically records and stores the X-rays converted into electronic signals; S4. After a successful scan, rotate the sample holder on the sample stage to a certain angle and repeat a new scan. When the rotation angle of the sample holder reaches 360°, the entire scanning of a coal sample is completed. The CT pore structure diagram and NMR fluid distribution diagram are shown as follows. Figure 2 and Figure 3 shown.

[0030] The maximum sphere algorithm is used to construct the equivalent pore-fracture network model: First, any point in the pore space is taken as the base point ( Figure 4 The red part in the figure), continuously searching for the largest inscribed sphere with this point as the center and tangent to the skeleton boundary ( Figure 4 ); Secondly, when all inscribed spheres are found, the other inscribed spheres contained in the inscribed sphere will be removed, and the remaining inscribed spheres will form a sphere set; then, a clustering algorithm can be used to classify and merge the largest spheres and identify pores ( Figure 5 ) and throat ( Figure 6 Finally, the pores can be represented by larger spheres, and the throats can be represented by a series of smaller spheres. Based on the maximum sphere algorithm for extracting the pore-fracture network model, the pores can be defined as the equivalent maximum inscribed sphere extracted from the pore-fracture network model. The radius of the inscribed sphere can be obtained using the equal diameter expansion method of the equivalent sphere, which is the pore radius ( Figure 5 The pore volume can be calculated based on the pore radius.

[0031] P i Represents multi-source input parameters, including reservoir properties (such as porosity , permeability K ), mineral components C clay , formation water mineralization σ w wait; H i Represents the historical evolution path, that is, at different moments t ∈[ t 0, t n ], the change sequence of reservoir damage rate is: ; In the above formula, Indicates that the sample is t k Reservoir damage rate at time.

[0032] In some embodiments, the multi-source input parameters include: lithology, burial depth, initial porosity, initial permeability, cementation type, montmorillonite content, illite content, chlorite content, kaolinite content, illite-montmorillonite mixed layer content, chlorite-montmorillonite mixed layer content, clay mineral content, quartz content, feldspar content and at least two of the formation water mineralization.

[0033] In some embodiments, the model parameters corresponding to the intelligent model are determined during the training process by minimizing a loss function, wherein the loss function is determined according to S101 to S103: S101. For each training data in a plurality of training data, execute processing steps S201-S202: S201, inputting the multidimensional features corresponding to the training data into a preset empirical model to determine an empirical value of the reservoir damage rate corresponding to the training data; In some embodiments, the preset empirical model is a multiple linear regression model. The multiple linear regression method is to find the most appropriate regression function with at least two independent variables from a large amount of data. Mathematically, it can be written as:

[0034] In the above formula, X yes n × k The independent variable matrix of Y is the dependent variable n dimensional vector, β yes k dimensional regression coefficient vector, ε is the random error associated with the regression.

[0035] The above formula can be further transformed into:

[0036] The regression coefficients can be estimated using the least squares method β The optimal regression equation was established by using the F test and R test.

[0037] when X When the rank is full, the regression coefficient can also be obtained by the following formula β：

[0038] In order to make the multivariate linear regression model applicable to a wider range of different oilfield reservoirs, the embodiment of the present invention expands the input parameters to the parameters described in Table 1, and the observation value is the reservoir damage rate. β The values ​​correspond to F0 to F14 in Table 1. In order to avoid introducing unnecessary errors and improve the usability of the empirical correlation formula, the embodiment of the present invention will bias ε Introduce the regression equation and calculate the specific value. Finally, the regression coefficient of the multivariate linear regression model is determined according to the embodiment of the present invention. β and bias ε as follows: ; .

[0039] Table 1 Dataset statistical indicators

[0040] In some embodiments, the intelligent model is a Light Gradient Boosting Machine (LightGBM). LightGBM is a new implementation of the gradient boosting machine algorithm, an optimization based on the XGBoost algorithm. It uses the same gradient descent approach as XGBoost, performing a second-order Taylor expansion of the loss function term in the objective function at the current point before solving for the optimal point. This approach retains more information about the objective function and improves prediction accuracy.

[0041] After performing a second-order Taylor expansion on the loss function, the objective function can be written as: (1) (2) (3) (4) Where, x i are input features, y i is the true value, It is t -1-step forecast value, is the loss function, T is the number of leaves, w is the weight vector, Ω( f t ) is a regularization term, and γ and λ are user-defined hyperparameters.

[0042] Define a I j ={ i | q ( x i )= j}, represents the input sample index set assigned to leaf node j, then the objective function can be converted to: (5) (6) (7) Where, G j and H j is the first and second order derivative of each leaf node sample. Each term in ∑ is about w jThe quadratic equation of , let its reciprocal be 0, and the optimal weight can be obtained .

[0043] (8) Substituting the optimal weight into the objective function eliminates the weight variable. After eliminating the weight variable, the objective function (Formula 13) can guide the splitting of the t-th decision tree, just like the Gini index and squared error.

[0044] (9) We hope that each node split can reduce the objective function as much as possible. The expression of the reduction degree is as follows: (10) The first, second, and third items in the brackets are the scores of the left child node, the right child node, and the parent node, respectively. Therefore, the division of the left and right nodes is completely determined by L split Decide.

[0045] LightGBM abandons the level-wise decision tree growth strategy used by most gradient boosting decision tree (GBDT) tools and instead employs a leaf-wise algorithm with a depth limit. XGBoost uses a level-wise growth strategy, which allows for simultaneous splitting of leaves at the same level within a single data pass. This facilitates multi-threaded optimization, effectively controls model complexity, and reduces overfitting. However, the level-wise strategy is inefficient because it treats leaves at the same level indiscriminately. Many leaves have low split gains, making search and selection unnecessary, thus increasing computational overhead. LightGBM's leaf-wise strategy finds the leaf with the highest split gain from all current leaves, then splits it, repeating the cycle. Compared to the level-wise strategy, its advantage is that, given the same number of splits, the leaf-wise algorithm achieves lower leaf-wise error and higher accuracy. However, its disadvantage is that it can lead to deeper decision trees, which can cause overfitting. Therefore, LightGBM adds a maximum depth limit to the leaf-wise algorithm to ensure high efficiency while preventing overfitting.

[0046] LightGBM's scalability is reflected in three key aspects: using smaller samples, less memory, and fewer features. This is achieved through three techniques: Gradient-based one-side sampling (GOSS), a histogram algorithm, and exclusive feature bundling (EFB). GOSS is a sampling algorithm that discards samples that do not contribute to information gain, significantly reducing computational effort and time.

[0047] Furthermore, compared to XGBoost's pre-sorting algorithm, LightGBM's histogram-based algorithm reduces its time complexity from O(Data × Features) to O(Bins × Features). In terms of memory, the histogram algorithm saves approximately seven times the memory consumption of the pre-sorting algorithm. The EFB algorithm is used to reduce feature dimensionality. It converts a large number of mutually exclusive features into low-dimensional dense features, effectively avoiding the calculation of redundant zero-valued features.

[0048] Therefore, LightGBM is faster and uses less memory while maintaining high accuracy. As oilfield datasets continue to expand in the future, LightGBM has potential application value in reservoir damage prediction and even the entire oil and aviation fuel industry.

[0049] S202: determining a deviation corresponding to the training data based on an empirical value and an observed value of a reservoir damage rate corresponding to the training data; Specifically, the determination of the deviation corresponding to each training data includes: according to the empirical value of the reservoir damage rate corresponding to each training data and observations , determine the deviation corresponding to each training data : . (11) S102, determining a penalty coefficient according to the deviation and hyperparameter corresponding to each training data in the plurality of training data; Specifically, the penalty coefficient is determined by: according to the deviation corresponding to each training data in the training data and hyperparameters , determine the penalty coefficient : (12) Wherein, N is the total number of training data required by the loss function during the training process.

[0050] It should also be noted that the penalty coefficient While maintaining the sample weight, it is convenient to further train samples with large deviations from the empirical correlation formula; hyperparameters In the range of 0 to 1, the selection will be made in Bayesian optimization.

[0051] S103. Determine a loss function based on the penalty coefficient, the deviation corresponding to each training data, the observation value, and the model prediction value corresponding to the intelligent model.

[0052] Specifically, the loss function is determined by: And the deviation corresponding to each training data , the observed value And the model prediction value corresponding to the intelligent model Determine the loss function : (13) Furthermore, according to formulas (4) and (5), the loss function can be determined The first-order gradient and the second-order gradient as follows: (14) (15) The loss function The first-order gradient and the second-order gradient Integrate into LightGBM's custom loss function for intelligent model training.

[0053] It should be noted that the gradient of the loss function, that is, the derivative of the loss function with respect to the model parameters, is of great significance in machine learning and optimization. Its core role is to guide the update direction of the model parameters to minimize the loss function.

[0054] For steps S101 to S103, in order to integrate the knowledge of reservoir damage into the intelligent model and form a knowledge-guided intelligent framework, the embodiment of the present invention uses the empirical model as a reference for model prediction. Traditional data-driven models often use mean square error as the loss function, and observe the changes in the loss function on the data set to adjust the model parameters to achieve the purpose of training. For this purpose, a deviation , penalty coefficient β DK The loss function is introduced as a weight factor and a scale.

[0055] In some embodiments, by constructing the following 4D mapping operator , complete the process from twin input to reservoir damage rate Prediction output:

[0056] in, θ Represents the training parameters of the LightGBM intelligent model; M i Represents three-dimensional structural characteristics, including: pore radius , Roar length and equivalent pore-throat ratio ; P i Represents multi-source input parameters, including reservoir properties (such as porosity , permeability K ), mineral components C clay , formation water mineralization σ w wait; H i Represents the historical evolution path, that is, at different moments t ∈[ t 0, t n ], the corresponding change sequence of reservoir damage rate.

[0057] Time dimension (T): The system can integrate historical data and real-time acquisition parameters from multiple stages (drilling, completion, injection and production, acidizing, etc.) to construct a time series of damage evolution. The LightGBM model is combined with historical status to predict future evolution trends. The prediction results at each time point will be mapped to the current reservoir damage rate, forming an evolution path with memory characteristics. Construct state difference sequence as input variable and construct time series input as follows:

[0058] Depth dimension (D): Based on the acquisition of reservoir CT scan images and mineral distribution characteristics, the system establishes the corresponding relationship between the reservoir longitudinal section and geological stratification, and can identify the mineral distribution, porosity and permeability structure and damage rate changes in different depth layers. Upper: Porosity , permeability , mineral components and formation water mineralization Mapping into vectors X D , defined as follows: ; Spatial dimension (S): Combine CT / NMR image processing with the maximum sphere algorithm to reconstruct the three-dimensional pore and fracture network, extract the equivalent pore throat structure, seepage channel and damage sensitive area, and realize micron / millimeter level three-dimensional spatial mapping. At the same time, it supports projecting the predicted output damage rate data into the 3D model according to the wellbore position and grid coordinates to construct a reservoir damage rate distribution map. Mapped into a three-dimensional reservoir damage distribution function:

[0059] In the above formula, Indicates that time t is at coordinate The reservoir damage rate, Represents the distribution cross section of the reservoir damage rate at time t.

[0060] Dynamic Dimension (Δ): By training a LightGBM model with dynamic memory capabilities and introducing the "empirical model deviation-penalty factor" mechanism as a loss function constraint, the model not only fits the current input but also has the ability to judge evolutionary trends. The system can dynamically update the model structure and parameters based on new input data, achieving continuous learning and iterative correction of damage expansion. Introducing a dynamic penalty loss function based on deviation control:

[0061] In the above formula, represents the penalty coefficient; Represents the deviation corresponding to each training data; represents the observed value; Indicates the model prediction value corresponding to the intelligent model.

[0062] In summary, the proposed 4D intelligent predictive diagnostic method for the spatiotemporal evolution of digital twin oil and gas reservoir damage, through a combination of static and dynamic analysis, image-structure fusion, and data-model collaboration, completes a complete closed-loop process from reservoir physical property acquisition, image modeling, feature extraction, intelligent prediction, to visual output. This system transcends the limitations of existing technologies, which rely on static assessment, two-dimensional prediction, and empirical analysis, providing a novel solution for damage monitoring and intelligent control throughout the lifecycle of oil and gas reservoirs.

[0063] In some embodiments, the training data is created based on the synthetic minority oversampling technique (SMOTE).

[0064] SMOTE is a synthetic artificial sampling technique, an improved oversampling algorithm based on random sampling. It is used to overcome the problem of imbalanced data. The SMOTE algorithm examines and simulates samples from several categories and synthesizes data by oversampling samples from these categories.

[0065] The SMOTE algorithm's generation process is as follows: First, the K-nearest neighbor method is applied to each sample point. Then, N adjacent points are randomly selected from these K adjacent points. The difference between the sample and its nearest neighbor is multiplied by a random number between 0 and 1 to achieve the goal of synthesizing the data. The algorithm essentially samples in the feature space rather than in the data space. All adjacent points in the feature space have comparable characteristics. Therefore, the SMOTE algorithm avoids the problem of undersampling, which can lead to the loss of some implicit information, nor does it suffer from the simple replication that can cause overfitting of the model, as occurs with oversampling.

[0066] Its mathematical expression can be written as:

[0067] in, represents the generated synthetic samples, It represents the original minority class sample, Indicates from of K A randomly selected sample from the nearest neighbors, Represents a random number in the interval [0,1].

[0068] During their research, the inventors of this application discovered that feature selection, or data dimensionality reduction, is a key step in addressing high-dimensional nonlinear problems. Appropriate feature selection can remove redundant and irrelevant features, thereby enabling the extraction of useful feature information and significantly impacting classification and regression results. However, in previous reservoir damage prediction methods, feature selection was often arbitrary, which easily led to inaccurate model predictions.

[0069] Based on this, an embodiment of the present invention proposes to determine the modeling features corresponding to the intelligent model in the following manner: S301, using multiple evaluation models to evaluate the importance of different feature combinations corresponding to the multidimensional features, and generating a feature importance ranking list corresponding to each evaluation model; Specifically, the multiple evaluation models include: random forest, extreme gradient boosting tree and light gradient boosting machine.

[0070] For each evaluation model: the feature importance ranking list corresponding to the evaluation model includes feature importance ranking positions and scores corresponding to different feature combinations.

[0071] Random Forest evaluates feature importance based on out-of-bag (OOB) error by calculating the reduction in out-of-bag accuracy after adding noise to a feature. When building a decision tree, XGBoost calculates the gain of each feature when splitting a node. The greater the gain, the higher the feature importance. LightGBM, also based on the gradient boosting framework, evaluates feature importance by calculating the gain when splitting a feature. Its histogram optimization technology makes feature evaluation more efficient.

[0072] S302: Determine the comprehensive ranking of feature importance corresponding to any feature combination based on the feature importance ranking lists corresponding to the multiple evaluation models; Specifically, for any of the feature combinations, the feature importance rankings determined by the random forest are as follows: , Random Forest determines the ranking of feature importance 、 LightGBM Determine the ranking of feature importance And the number of features corresponding to this feature combination , determine the comprehensive ranking of feature importance corresponding to the feature combination ;

[0073] S303: The feature combination corresponding to the minimum value of the comprehensive ranking of feature importance is used as the modeling feature.

[0074] For S303 , the smaller the comprehensive ranking of feature importance corresponding to the feature combination is, the higher the importance corresponding to the feature combination is.

[0075] The method for determining modeling features corresponding to the intelligent model, proposed in this embodiment of the present invention, assesses the importance of feature combinations using three tree ensemble models (random forest, extreme gradient boosting tree, and light gradient boosting machine). This ensures feature distinguishability and the accuracy of evaluation results. Compared to unsupervised feature selection, it more accurately reflects the relationship between features and targets. Furthermore, it employs set sorting to provide a more comprehensive ranking of factors influencing reservoir damage.

[0076] The embodiment of the present invention further provides a digital twin oil and gas reservoir damage spatiotemporal evolution 4D intelligent prediction and diagnosis system, the system comprising: Feature extraction module: used to obtain the digital twins corresponding to the well section samples in the target reservoir and convert the digital twins corresponding to the well section samples in the target reservoir into corresponding multidimensional features; Prediction module: used for inputting the multi-dimensional features into the intelligent model to determine the change sequence of the reservoir damage rate corresponding to the target reservoir in the future time period; The digital twin corresponding to the well section sample includes three-dimensional structural features, multi-source input parameters, and historical time series data corresponding to the reservoir damage rate; the model parameters corresponding to the intelligent model are determined during the training process by minimizing the loss function, and the loss function is determined according to the following method: For each training data in the plurality of training data, the following processing steps are performed: inputting a multidimensional feature corresponding to the training data into a preset empirical model to determine an empirical value of a reservoir damage rate corresponding to the training data; and determining a deviation corresponding to the training data based on the empirical value and an observed value of the reservoir damage rate corresponding to the training data; Determining a penalty coefficient according to the deviation and hyperparameter corresponding to each training data in the plurality of training data; and A loss function is determined based on the penalty coefficient, the deviation corresponding to each training data, the observation value, and the model prediction value corresponding to the intelligent model.

[0077] An embodiment of the present invention provides a 4D intelligent predictive diagnosis device for the spatiotemporal evolution of digital twin oil and gas reservoir damage, comprising: a memory configured to store instructions; and a processor configured to call the instructions from the memory and implement the 4D intelligent predictive diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage when executing the instructions.

[0078] An embodiment of the present invention provides a processor for running a program, wherein the program, when run, is used to execute: the 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage.

[0079] An embodiment of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage.

[0080] The 4D intelligent predictive diagnostic method for the spatiotemporal evolution of digital twin oil and gas reservoir damage, provided by this embodiment of the invention, firstly designs the penalty coefficient and deviation of the loss function based on the corresponding predicted values ​​of the intelligent model and the empirical model. This integrates knowledge in the field of reservoir damage into the intelligent model, forming a knowledge-guided intelligent framework. Secondly, by inputting multidimensional features into the intelligent model, it can simulate the evolution of damage during different development stages, such as waterflooding, fracturing, and acidizing, and optimize operational parameters. Through intelligent model prediction, reservoir damage rate prediction results can be obtained in real time, thus supporting real-time adjustments to development plans.

[0081] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0082] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0083] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0085] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0086] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0087] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can be implemented using any method or technology for information storage. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0088] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0089] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage, characterized by: The method comprises: Converting the digital twins corresponding to the well section samples in the target reservoir into corresponding multidimensional features, and inputting the multidimensional features into the intelligent model to determine the change sequence of the reservoir damage rate corresponding to the target reservoir in the future time period; The digital twin corresponding to the well section sample includes three-dimensional structural features, multi-source input parameters, and historical time series data corresponding to the reservoir damage rate; the model parameters corresponding to the intelligent model are determined during the training process by minimizing the loss function, and the loss function is determined according to the following method: For each training data in the plurality of training data, the following processing steps are performed: inputting a multidimensional feature corresponding to the training data into a preset empirical model to determine an empirical value of a reservoir damage rate corresponding to the training data; and determining a deviation corresponding to the training data based on the empirical value and an observed value of the reservoir damage rate corresponding to the training data; Determining a penalty coefficient according to the deviation and hyperparameter corresponding to each training data in the plurality of training data; and A loss function is determined based on the penalty coefficient, the deviation corresponding to each training data, the observation value, and the model prediction value corresponding to the intelligent model.

2. The method according to claim 1, characterized in that Determining the deviation corresponding to each training data includes: According to the empirical value of the reservoir damage rate corresponding to each training data and observations , determine the deviation corresponding to each training data : 。 3. The method according to claim 1, characterized in that Determining the penalty coefficient includes: According to the deviation corresponding to each training data in the training data and hyperparameters , determine the penalty coefficient : Wherein, N is the total number of training data required by the loss function during the training process.

4. The method according to claim 1, wherein Determining the loss function includes: According to the penalty coefficient And the deviation corresponding to each training data , the observed value And the model prediction value corresponding to the intelligent model Determine the loss function : 。 5. The method according to claim 1, wherein The preset empirical model is a multiple linear regression model, and the intelligent model is a light gradient boosting machine LightGBM.

6. The method according to claim 1, characterized in that The training data is created based on the SMOTE artificial sampling technique.

7. The method according to claim 1, characterized in that Determine the modeling features corresponding to the intelligent model according to the following method: Using multiple evaluation models to evaluate the importance of different feature combinations corresponding to the multidimensional features, and generating a feature importance ranking list corresponding to each evaluation model; Determine the comprehensive ranking of feature importance corresponding to each feature combination according to the feature importance ranking lists corresponding to the multiple evaluation models; as well as The feature combination corresponding to the minimum value of the comprehensive ranking of feature importance is used as the modeling feature.

8. The method according to claim 1, characterized in that The multi-source input parameters include: lithology, burial depth, initial porosity, initial permeability, cementation type, montmorillonite content, illite content, chlorite content, kaolinite content, illite-montmorillonite mixed layer content, chlorite-montmorillonite mixed layer content, clay mineral content, quartz content, feldspar content and at least two of the formation water salinity.

9. A digital twin oil and gas reservoir damage spatiotemporal evolution 4D intelligent prediction and diagnosis system, characterized by: The system comprises: Feature extraction module: used to obtain the digital twins corresponding to the well section samples in the target reservoir and convert the digital twins corresponding to the well section samples in the target reservoir into corresponding multidimensional features; Prediction module: used for inputting the multi-dimensional features into the intelligent model to determine the change sequence of the reservoir damage rate corresponding to the target reservoir in the future time period; The digital twin corresponding to the well section sample includes three-dimensional structural features, multi-source input parameters, and historical time series data corresponding to the reservoir damage rate; the model parameters corresponding to the intelligent model are determined during the training process by minimizing the loss function, and the loss function is determined according to the following method: For each training data in the plurality of training data, the following processing steps are performed: inputting a multidimensional feature corresponding to the training data into a preset empirical model to determine an empirical value of a reservoir damage rate corresponding to the training data; and determining a deviation corresponding to the training data based on the empirical value and an observed value of the reservoir damage rate corresponding to the training data; Determining a penalty coefficient according to the deviation and hyperparameter corresponding to each training data in the plurality of training data; and A loss function is determined based on the penalty coefficient, the deviation corresponding to each training data, the observation value, and the model prediction value corresponding to the intelligent model.

10. A digital twin oil and gas reservoir damage spatiotemporal evolution 4D intelligent prediction and diagnosis device, characterized by: include: a memory configured to store instructions; And a processor, configured to call the instructions from the memory and to implement the 4D intelligent prediction and diagnosis method of the spatiotemporal evolution of digital twin oil and gas reservoir damage according to any one of claims 1 to 8 when executing the instructions.

11. A processor, characterized in that: Used to run a program, wherein the program, when run, is used to execute: the 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage as described in any one of claims 1 to 8.

12. A computer program product, characterized in that It includes a computer program, which, when executed by a processor, implements the 4D intelligent prediction and diagnosis method for the spatiotemporal evolution of digital twin oil and gas reservoir damage as described in any one of claims 1 to 8.

Images