Rapid data assimilation method for oil reservoir parameters based on vector space and flow model
By combining vector space and flow models, the mapping from prior data to posterior parameter fields is directly performed, which solves the problem of low computational efficiency in traditional data assimilation methods and realizes rapid assimilation and efficient inversion of reservoir parameters.
Patent Information
- Application Number
- CN202510753803.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-06
- Publication Date
- 2025-09-12
AI Technical Summary
Existing model-based data assimilation methods require repeated iterative calculations, resulting in low computational efficiency and high resource consumption. Traditional proxy models rely on iterative optimization in the parameter space and cannot effectively improve the computational speed.
A method based on vector space and flow model is adopted to construct a priori model through geological modeling, cluster analysis is performed using the DBSCAN algorithm, typical curves are extracted for fitting, and a flow model is constructed for rapid inversion of reservoir parameters. Direct mapping from prior data to posterior data avoids repeated simulation.
It realizes rapid data assimilation from prior data to posterior parameter fields, simplifies the data assimilation process, improves computing efficiency, and reduces computing time and resource consumption.
Smart Images
Figure CN120633423A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of oil and gas field development, and in particular relates to a fast data assimilation method for reservoir parameters based on vector space and flow model. Background Art
[0002] Data assimilation is the process of using a numerical reservoir model to repeatedly adjust model parameters to achieve the best match between simulated production data and historically observed production data. It is an essential step in reservoir production prediction and uncertainty quantification.
[0003] Common model-based data assimilation methods require repeated iterative calculations of uncertain parameters and involve a large number of numerical simulation processes, which results in low computational efficiency and high resource consumption. To improve computational efficiency, surrogate models have been introduced to replace the original time-consuming reservoir numerical simulation process, thereby significantly increasing computational speed and have been widely used in the industry. However, from the perspective of the complete data assimilation process, current surrogate model-based data assimilation still relies on iterative optimization in the parameter space. The flow model (Normalizing Flow) constructs an explicit probabilistic association between parameters and observed data through a reversible neural network, and can directly generate a posterior parameter distribution that conforms to geological constraints in a single forward propagation. This paradigm innovation based on probabilistic reasoning effectively circumvents the computational bottleneck of repeatedly calling simulators in traditional methods, providing a new technical path for the rapid assimilation of reservoir parameters. Summary of the Invention
[0004] The problem to be solved by the present invention is to provide a method for rapid data assimilation of reservoir parameters based on vector space and flow models. This method not only avoids repeated forward simulations but also simplifies the complex process of data assimilation, achieving rapid data assimilation from prior data to posterior data and then to the posterior parameter field. Specifically, a vector space-based method is used to directly predict posterior data using prior data and observation data, without the need for model generation, thus avoiding repeated forward simulations. A flow model is used to construct an efficient inversion method for reservoir parameters, simplifying the tedious iterative model update process to a mapping relationship from production data to reservoir parameters, significantly improving data assimilation efficiency.
[0005] To solve the above technical problems, the present invention adopts a technical solution: a method for rapid data assimilation of reservoir parameters based on vector space and flow model, comprising the following steps:
[0006] S1: Apply geological modeling software to construct a priori reservoir models and obtain corresponding production data through numerical simulation;
[0007] S2: Use the DBSCAN algorithm to perform cluster analysis on the prior data and determine the optimal number of clusters;
[0008] S3: For each type of prior data, a typical curve is extracted as a basis vector to fit the observed data to obtain the posterior data;
[0009] S4: Use the prior field and corresponding production data to construct a training sample set and build a rapid inversion method for reservoir parameters based on the flow model;
[0010] S5: Use the posterior data obtained based on the vector space and input it into the trained flow model to perform rapid data assimilation of reservoir parameters.
[0011] Furthermore, the S2 includes the following steps:
[0012] S21: Set the neighborhood radius eps and minimum number of samples N for DBSCAN min , mark all production data curves as unvisited, calculate the EPS neighborhood of each curve and count the number of curves in the neighborhood;
[0013] S22: Traverse the curves. If the number of curves in the neighborhood of a curve is ≥ N min , marked as a core point and a new cluster is created, recursively expanding the unvisited curves in its neighborhood to join the cluster; otherwise, it is marked as a noise candidate point;
[0014] S23: Mark all noise candidate points that are not assigned to a cluster as noise, completing DBSCAN clustering.
[0015] Furthermore, the S3 includes the following steps:
[0016] S31: Each set of production data is regarded as a d-dimensional vector, where d is the number of time steps, and all the data are constructed into an N l ×d-dimensional vector space;
[0017] S32: The center points of each cluster after classification are used as basis vectors in the vector space. In this space, the observation data d obs Represented as basis vectors e1,e2,...,e k The linear combination of is as follows:
[0018] d obs =a1e1+a2e2+...+a k e k
[0019] Among them, a1, a2, ..., a k is the coefficient to be determined,
[0020]
[0021] S33: Calculate the best linear combination and obtain N lposterior production data d posterior , the formula is as follows,
[0022]
[0023] Furthermore, the S4 includes the following steps:
[0024] S41: Use flow models to establish mapping relationships from production data to reservoir parameter fields;
[0025] S42: Through multiple reversible and differentiable transformations f, the prior distribution p of production data is transformed Z (z) is mapped to the target distribution p of the reservoir attribute field X (x),
[0026]
[0027] Among them, z represents production data, x represents reservoir attribute field;
[0028] S43: By maximizing the log-likelihood of the target data, the inversion model learns to generate a reservoir parameter field that approximates the true data distribution.
[0029]
[0030] S44: Using N obtained in S1 r The inversion model is trained using a training data set, the input of which is reservoir production data, and the model output is the reservoir parameter field.
[0031] Furthermore, the present invention provides a device for executing the above-mentioned data processing method.
[0032] Furthermore, the present invention provides a device comprising a memory, a processor, and an algorithm stored in the memory and executable on the processor, wherein the processor implements the above-mentioned data processing method when executing the computer program.
[0033] Furthermore, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer algorithm, and the computer algorithm implements the above-mentioned data processing when executed by a processor.
[0034] The advantages and positive effects of the present invention are:
[0035] 1. The present invention proposes a posterior production data prediction method based on vector space, which can directly predict posterior production data based on prior data and observation data without generating a reservoir model, thus avoiding time-consuming reservoir numerical simulation calculations.
[0036] 2. This invention develops an efficient reservoir parameter inversion method based on a flow model. This method, driven by data, directly establishes a nonlinear mapping relationship between production dynamics data and reservoir parameter fields. This simplifies the tedious iterative model update process into a single inversion from production data to reservoir parameters, significantly improving data assimilation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is a schematic diagram of the overall process of an embodiment of the present invention.
[0038] Figure 2 This is a permeability reference field map of a two-dimensional reservoir calculation example according to an embodiment of the present invention.
[0039] Figure 3 1 is a daily oil production curve diagram of three production wells obtained by the vector space-based posterior production data prediction method according to an embodiment of the present invention.
[0040] Figure 4 1 is a graph of daily water production curves of three production wells obtained by the vector space-based posterior production data prediction method according to an embodiment of the present invention.
[0041] Figure 5 This is a reservoir parameter field map obtained by a rapid reservoir parameter data assimilation method based on vector space and flow model in an embodiment of the present invention.
[0042] Figure 6 It is a daily oil production curve diagram of three production wells obtained by numerical simulation of oil reservoirs in an embodiment of the present invention.
[0043] Figure 7 This is a daily water production curve diagram of three production wells obtained by numerical simulation of the oil reservoir in an embodiment of the present invention. DETAILED DESCRIPTION
[0044] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0045] The present invention will be specifically described below with reference to the accompanying drawings:
[0046] like Figure 1 The figure shows the overall process of a fast data assimilation method for reservoir parameters based on vector space and flow model, which specifically includes the following steps.
[0047] S1: Apply geological modeling software to construct a priori reservoir model and obtain corresponding production data through numerical simulation. Specifically, S1 includes the following steps:
[0048] S11: In the geological modeling software, a random geostatistical modeling method is applied to obtain N r A two-dimensional reservoir model;
[0049] S12: Perform numerical simulations on all models to obtain corresponding prior production data, and randomly select a model field as a reference field.
[0050] S2: Use the DBSCAN algorithm to perform cluster analysis on the prior data to determine the optimal number of clusters. Specifically, S2 includes the following steps.
[0051] S21: Use DBSCAN algorithm to perform cluster analysis on prior data, set the neighborhood radius (eps) and the minimum number of samples (N min ), all production data curves are marked as unvisited. For each well, the N1 set of prior production data is calculated for each curve's eps neighborhood (all curves with a distance ≤ eps), and the number of curves in the neighborhood is counted.
[0052] S22: Calculate the number of curves in the neighborhood of each curve eps and N in turn min If the current number of curves is less than N min If the number of current curves is greater than N, then min , it is marked as a core point and a new cluster is created. Starting from the core point, all unvisited curves and their neighbors in the neighborhood of the core point are added to the current cluster through density expansion and marked as visited.
[0053] S23: All noise candidate points that are not assigned to a cluster are uniformly marked as noise (outlier curve) to complete the DBSCAN clustering analysis.
[0054] S3: For each type of prior data, a typical curve is extracted as a basis vector to fit the observed data to obtain the posterior data. Specifically, S3 includes the following steps.
[0055] S31: Each set of production data is regarded as a d-dimensional vector, where d is the number of time steps, and all the data are constructed into an N l ×d-dimensional vector space;
[0056] S32: The center points of each cluster after classification are used as basis vectors (k×d) in the vector space. In this space, the observation data d obs Represented as basis vectors e1,e2,...,e k The linear combination of is as follows:
[0057] d obs =a1e1+a2e2+...+a k ek
[0058] Among them, a1, a2, ..., a k is the coefficient to be determined,
[0059]
[0060] S33: Calculate the best linear combination and obtain N l posterior production data d posterior , the formula is as follows,
[0061]
[0062] S4: Use the prior field and the corresponding production data to construct a training sample set and build a rapid inversion method for reservoir parameters based on the flow model. Specifically, S4 includes the following steps:
[0063] S41: The core idea of the flow model is to construct a nonlinear mapping from structured latent space to high-dimensional parameter field through a hierarchical coupling mechanism of reversible transformations, breaking through the limitations of explicit probability density modeling in traditional inversion. Production data is mapped to a latent variable space with geological decoupling characteristics, and a reversible neural network is used to realize the bidirectional probabilistic association between the spatial topological structure of the parameter field and the dynamic production data. The input data is mapped to a latent variable space, and transformation operations are performed in this space to realize the construction of complex data distribution. Based on the flow model, end-to-end probabilistic reasoning from production data to parameter field can be achieved while maintaining the accuracy of heterogeneous characterization of geological structure. Compared with traditional iterative optimization methods, it has the significant advantages of not requiring gradient calculation and avoiding local optimal solutions.
[0064] S42: Through a reversible and differentiable transformation f, the prior distribution p of the production data Z (z) is mapped to the target distribution p of the reservoir attribute field X (x),
[0065]
[0066] Among them, z represents production data, x represents reservoir attribute field;
[0067] S43: By maximizing the log-likelihood of the target data, the inversion model learns to generate a reservoir parameter field that approximates the true data distribution.
[0068]
[0069] S44: Using N obtained in S1 r The inversion model is trained using a training data set, the input of which is reservoir production data, and the model output is the reservoir parameter field.
[0070] S5: Use the posterior data obtained based on the vector space and input it into the trained flow model to perform rapid data assimilation of reservoir parameters. Specifically, S5 includes the following steps:
[0071] S51: Use the flow model trained in S4 to predict the posterior data and obtain the corresponding posterior parameter field;
[0072] S52: Obtain a posteriori production data through numerical simulation and verify its fitting results.
[0073] The present invention will be specifically described below in conjunction with specific embodiments:
[0074] First, the random geostatistical modeling method was applied in the geological modeling software to obtain 801 two-dimensional reservoir models. All the fields were numerically simulated to obtain the corresponding production data. A model field was randomly selected as the reference field, e.g. Figure 2 The remaining 800 models are used as prior models. The DBSCAN algorithm is used to perform cluster analysis on the prior data. 50 geologically representative basis vectors are extracted from the original 800 models. The observed data are fitted using linear combinations in vector space to generate 100 sets of posterior production data. Figure 3 and Figure 4 The daily oil and water production curves of the three wells are shown respectively. It can be seen from the figure that the vector space-based method can accurately predict the observed data. The obtained posterior production data can cover the observed data. Compared with the prior data, the uncertainty of the posterior data is greatly reduced.
[0075] 800 sets of production data and their corresponding permeability fields were used as training data to train the flow model. The input was reservoir production data, including single-well oil production and single-well water production. The model output was the reservoir permeability field. By using the flow model for rapid data assimilation, this method can reduce the time required to generate a single parameter field from 25 minutes in traditional numerical simulation to 0.8 seconds. 100 sets of posterior production data were input into the trained flow model to obtain the corresponding 100 posterior parameter fields. Figure 5 The figure shows some of the posterior parameter fields generated by the flow model. As can be seen from the figure, the generated posterior fields have realistic visual effects. To further verify the fitting effect, the generated posterior fields are numerically simulated and the daily oil and water production curves are plotted. Figure 6-7 As shown, compared with the prior data, the fitted data are more concentrated around the reference data and cover most of the observed data, indicating that the generated parameter field meets the requirements.
[0076] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.
Claims
1. A method for rapid data assimilation of reservoir parameters based on vector space and flow model, characterized by: The following steps are included: S1: Apply geological modeling software to construct a priori reservoir models and obtain corresponding production data through numerical simulation; S2: Use the DBSCAN algorithm to perform cluster analysis on the prior data and determine the optimal number of clusters; S3: For each type of prior data, a typical curve is extracted as a basis vector to fit the observed data to obtain the posterior data; S4: Use the prior field and corresponding production data to construct a training sample set and build a rapid inversion method for reservoir parameters based on the flow model; S5: Use the posterior data obtained based on the vector space and input it into the trained flow model to perform rapid data assimilation of reservoir parameters.
2. The method for rapid reservoir parameter data assimilation based on vector space and flow model according to claim 1, characterized in that: Said S2 comprises the following steps, S21: Set the neighborhood radius eps and minimum number of samples N for DBSCAN min , mark all production data curves as unvisited, calculate the EPS neighborhood of each curve and count the number of curves in the neighborhood; S22: Traverse the curves. If the number of curves in the neighborhood of a curve is ≥ N min , marked as a core point and a new cluster is created, recursively expanding the unvisited curves in its neighborhood to join the cluster; otherwise, it is marked as a noise candidate point; S23: Mark all noise candidate points that are not assigned to a cluster as noise, completing DBSCAN clustering.
3. The method for rapid reservoir parameter data assimilation based on vector space and flow model according to claim 1 or 2, characterized in that: Said S3 comprises the following steps, S31: Each set of production data is regarded as a d-dimensional vector, where d is the number of time steps, and all the data are constructed into an N l ×d-dimensional vector space; S32: The center points of each cluster after classification are used as basis vectors in the vector space. In this space, the observation data d obs Represented as basis vectors e1,e2,...,e k The linear combination of is as follows: d obs =a1e1+a2e2+...+a k and k Among them, a1, a2, ..., a k is the coefficient to be determined, S33: Calculate the best linear combination and obtain N l posterior production data d posterior , the formula is as follows, 4. The method for rapid reservoir parameter data assimilation based on vector space and flow model according to claim 1 or 2, characterized in that: Said S4 comprises the following steps, S41: Use flow models to establish mapping relationships from production data to reservoir parameter fields; S42: Through multiple reversible and differentiable transformations f, the prior distribution p of production data is transformed Z (z) is mapped to the target distribution p of the reservoir attribute field X (x), Among them, z represents production data, x represents reservoir attribute field; S43: By maximizing the log-likelihood of the target data, the inversion model learns to generate a reservoir parameter field that approximates the true data distribution. S44: Using N obtained in S1 r The inversion model is trained using a training data set, the input of which is reservoir production data, and the model output is the reservoir parameter field.
5. A device, characterized in that: Run the data processing method according to any one of claims 1 to 4.
6. A device comprising a memory, a processor, and an algorithm stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the data processing method according to any one of claims 1 to 4 is implemented.
7. A computer-readable storage medium storing a computer algorithm, characterized in that: When the computer algorithm is executed by a processor, the data processing according to any one of claims 1 to 4 is realized.