Learning stratigraphic unit boundary calibration quality control method and system

Abstract

The invention belongs to the technical field of oilfield development, and provides a learning stratigraphic unit boundary calibration quality control method and system. The method comprises the steps of obtaining basic data of all wells in a research area; according to the basic data of all wells, a first model is established by adopting a Kriging algorithm, and then a second model, a third model, a fourth model and a final development geologic model are obtained by performing quality control on the first model for multiple times. According to the stratigraphic unit boundary calibration quality control method based on the random forest method and the machine learning under the constraint of the standard continental deposition mode, on the basis of fidelity algorithm modeling, the efficiency is improved through the machine learning means, the three problems that unit boundary demarcation in oilfield development is low in manual recognition efficiency and precision and cannot adapt to mass data are effectively solved, and the method is suitable for popularization and application. And rapid and automatic comparison, screening and calibration of each level of stratigraphic unit boundary errors from an oil reservoir group level, a sandstone group level to a unit level are realized, and the precision and efficiency of stratigraphic comparison and auditing work are greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of oilfield development, and particularly relates to a quality control method and system for calibrating formation unit boundaries for learning. Background Art

[0002] At present, many oil and gas fields are discovered in basins with sedimentary rocks as the main reservoirs. Under the standard continental sedimentation model, the reservoirs show strong heterogeneity characteristics with ten to hundreds of small layers developing vertically and different thicknesses. For developed oilfields, especially high-water-cut old oilfields, to achieve precise and efficient development of multi-layer heterogeneous reservoirs, the primary task is to ensure the accuracy of formation unit boundaries at all levels from oil reservoir group level, sandstone group to unit level, and eliminate the risks of perforation cross-layer, high investment but poor implementation effect caused by missed or misinterpreted boundaries. However, there are the following difficulties in the current formation boundary verification: First, the quality control of formation boundaries mainly relies on manual review. With the continuous increase in the number of wells in developed oilfields, it is increasingly difficult to rely solely on manual quality control. Second, the conventional quality control of formation boundaries mainly relies on logging curve characteristics and interpretation data. However, there are differences in logging curves and interpretation results of well patterns in different periods of developed oilfields. It is difficult to ensure the accuracy of formation boundary division only relying on logging information. Third, the quality control of sedimentary unit-level formation boundaries mainly relies on one-by-one verification, lacking macro sedimentary geological model constraints and effective quality control means for calibrating river facies reservoir boundaries.

[0003] An important means of fine geological research in development is modeling, ranging from decimeter-level sand body units to oil reservoir groups dozens or hundreds of meters in size. The basis of modeling is basic data such as logging data and development data. In the modeling at the development area level, it is difficult to ensure that the accuracy rate of tens of thousands to hundreds of thousands of data from thousands of wells reaches 100%. Therefore, the technical data and model accuracy quality control discrimination criteria determined through development data and practical experience for specific development research areas are particularly important. The development research object applicable to the present invention is the quality control of sand body unit boundary calibration under typical continental sedimentation, especially river facies sedimentation models. The input data is the oil reservoir group-level model, but the quality control research object needs to be refined to the accuracy and rationality of the division of sand body unit boundaries at the unit (small layer) level. The input data as the quality control object is modeled through interpolation and fitting by the following method, and the subsequent quality control discrimination is efficiently realized through machine learning using the random forest algorithm. The designed algorithm principle and background are as follows:

[0004] Modeling principle in the quality control method for calibrating formation unit boundaries based on the random forest rule machine learning under the constraint of the standard continental sedimentation model: Kriging is a regression algorithm for spatially modeling and predicting (interpolating) a random process / random field based on the covariance function. The original algorithm is called ordinary Kriging, and the improved algorithms include universal Kriging, co-Kriging, and disjunctive Kriging, etc. According to the first law of geography, all values in space are interconnected, and values that are closer have stronger correlations. The random field uses the covariance function to describe the above conclusion, corresponding to the kernel function in the Gaussian process regression theory. Using the Kriging method requires the random field to satisfy two assumptions: First, the mathematical expectation of the random field exists and is independent of the position; Second, for any two points in the random field, its covariance function is only a function of the vector between the points.

[0005] There are three problems in the current verification work of formation boundaries in oilfield development geological research: First, the quality control of formation boundaries mainly relies on manual review, and with the increase of development wells, the accuracy and efficiency cannot be guaranteed; Second, the conventional quality control of formation boundaries mainly relies on logging curve characteristics and interpretation data, but there are inevitable statistical effects and systematic errors in the logging curves and interpretation results of different periods of well patterns in the developed oilfields; Third, the quality control of sedimentary unit-level formation boundaries mainly relies on one-by-one verification, lacking the constraint of the macroscopic sedimentary geological model and having no effective quality control means for calibrating the boundaries of fluvial facies reservoirs. Based on the fidelity algorithm modeling and using machine learning means to improve efficiency. Summary of the Invention

[0006] In order to solve at least one problem in the background technology, the present invention proposes a quality control method and system for learning formation unit boundary calibration.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A quality control method for learning formation unit boundary calibration includes the following steps:

[0009] Obtain the basic data of all wells in the research area;

[0010] According to the basic data of all wells, establish a first model using the Kriging algorithm;

[0011] Calibrate and control the boundaries of oil reservoir groups in the first model, complete the update of the oil reservoir group-level model and all well data, and obtain a second model;

[0012] Based on the second model, perform unit-level boundary calibration and quality control, complete the correction or elimination of abnormal wells, and obtain an updated unit-level model, which is the third model;

[0013] Based on the third model, identify the abnormal sand body areas inside the unit and lock the abnormal wells according to random forest machine learning. After removing the abnormal wells, the updated model is obtained as the fourth model.

[0014] Extract the unit-level models one by one from the fourth model to obtain the corrected model for each unit. Quality control the corrected unit model of each unit with the standard layer of the adjacent sandstone group to obtain the sandstone group-level boundary calibration model.

[0015] Based on the sandstone group-level boundary calibration model and the depth-domain seismic model for quality control, the final development geological model is obtained as the fifth model.

[0016] Preferably, obtain the basic data of all wells in the study area, including:

[0017] Determine the four-point coordinates of the study area.

[0018] Collect the layer data of the target layer drilled and the logging curve data in the entire study area.

[0019] Collect the 3D seismic data volume that completely covers the coordinate range of the study area.

[0020] Collect the depth-domain velocity model of the study area.

[0021] Collect the seismic geological interpretation results of the study area.

[0022] Preferably, according to the basic data of all wells, establish the first model using the Kriging algorithm, including the following steps:

[0023] Apply the logging curve data and the layer data to interpolate and establish the oil reservoir group-level model using the Kriging algorithm.

[0024] Refine the 3D grid of the oil reservoir group-level model to establish the sandstone group model.

[0025] Establish the unit-level model within the sandstone group model.

[0026] Determine the scales of the three directions of the 3D grid at different scales, and the number of update and iteration rounds for establishing each level of model.

[0027] Based on the quality control of the singularities of the geological model of the standard continental sedimentation pattern, if the vertical correlation coefficient of the eigenvalue in any grid of the oil reservoir group-level model is lower than 90% at the grid accuracy of 100m*100m, the quality control fails and comprehensive adjustment is required until the quality control is passed. After quality control and iterative update, the first model for subsequent quality control is obtained.

[0028] Preferably, conduct calibration quality control on the oil reservoir group-level boundaries in the first model, complete the update of the oil reservoir group-level model and all well data, and obtain the second model, including the following steps:

[0029] Apply the depth domain velocity model to transform the collected 3D seismic data volume and seismic geological interpretation results to the depth domain through time-depth conversion, obtaining seismic depth and horizon depth for subsequent quality control;

[0030] Calibrate and quality control the boundaries of oil reservoir groups in the first model based on seismic depth and horizon depth;

[0031] If the error rate * 100% between the horizon depth and the logging stratification depth of the oil reservoir group layer data greater than or equal to 30m in any grid is greater than 5%, the quality control fails, and the calibration position of the well corresponding to the logging data of the grid is defined as an unusable well. If the error rate is greater than or equal to 1% and less than or equal to 5%, adjust the logging curve oil reservoir group level stratification depth. If the error rate is less than 1%, pass the quality control without any adjustment and update, and complete the update of the oil reservoir group level model and all well data to obtain the second model.

[0032] Preferably, perform unit level boundary calibration and quality control based on the second model, complete the correction or elimination of abnormal wells, and obtain the updated unit level model, which is the third model, including the following steps:

[0033] Identify the logging curves of sandstone groups, establish sandstone group level models and threshold standards, and then obtain the standard layer to calculate the correction amount to achieve automatic correction and modification of sandstone group level boundaries;

[0034] Coarsen the spatial grid of a single unit, apply the maximum error control method under the constraint of the standard continental sedimentation model for quality control, and identify the first round of large boundaries of abnormal unit bodies;

[0035] Take the first round of large grid abnormal unit positions as input data, perform quality control on the automatic discrimination of faults after refining the grid, and use the quality control results to carve the unit level fault boundaries;

[0036] Use the standard continental sedimentation model constraint to complete the unit level boundary calibration and quality control, calculate the minimum sampling interval correction amount through variable grids, and complete the correction or elimination of abnormal wells;

[0037] If the error rate * 100% between the sand body unit greater than or equal to 2m and the logging stratification depth in any grid is greater than 5%, the quality control fails, and the calibration position of the well corresponding to the logging data of the grid is defined as an unusable well. If the error rate is greater than 1% and less than or equal to 5%, adjust the logging curve stratification depth. If the error rate is less than or equal to 1%, pass the quality control;

[0038] Eliminate the unusable wells in the second model to obtain the third model.

[0039] Preferably, based on the third model, complete the identification of abnormal sand body areas and lock abnormal wells according to random forest machine learning. After eliminating the abnormal wells, obtain the updated model as the fourth model, including the following steps:

[0040] Using the random forest algorithm, the identification and classification of different types of stacked sand bodies within the unit are realized through machine learning, and the abnormal sand body identification area is obtained;

[0041] Based on the abnormal sand body identification area, abnormal wells are locked;

[0042] Abnormal wells are removed from the third model to obtain the fourth model.

[0043] A quality control system for learning formation unit boundary calibration includes:

[0044] The basic unit is used to obtain the basic data of all wells in the study area;

[0045] The first unit is used to establish the first model using the Kriging algorithm based on the basic data of all wells;

[0046] The second unit is used to calibrate and quality control the oil reservoir group-level boundary in the first model, complete the update of the oil reservoir group-level model and all well data, and obtain the second model;

[0047] The third unit is used to calibrate and quality control the unit-level boundary based on the second model, complete the correction or removal of abnormal wells, and obtain the updated unit-level model as the third model;

[0048] The fourth unit is used to identify the abnormal sand body area within the unit based on the third model according to random forest machine learning, lock abnormal wells, and obtain the updated model as the fourth model after removing the abnormal wells;

[0049] The extraction unit is used to extract the unit-level models one by one from the fourth model, obtain the corrected models of each unit, and quality control each corrected unit model with the adjacent sandstone group standard layer to obtain the sandstone group-level boundary calibration model;

[0050] The fifth unit is used to perform quality control based on the sandstone group-level boundary calibration model and the depth domain seismic model to obtain the final development geological model as the fifth model.

[0051] Preferably, the basic unit includes:

[0052] The positioning module is used to determine the four-point coordinates of the study area;

[0053] The first collection module is used to collect the layer data and logging curve data of the target layer drilled in the entire study area;

[0054] The second collection module is used to collect the three-dimensional seismic data volume that completely covers the coordinate range of the study area;

[0055] The third collection module is used to collect the depth domain velocity model of the study area;

[0056] The fourth collection module is used to collect the seismic geological interpretation results of the research area.

[0057] Preferably, the first unit includes:

[0058] The interpolation module is used to apply well logging curve data and development data to interpolate and establish an oil reservoir group-level model using the Kriging algorithm.

[0059] The construction module is used to refine the 3D grid of the oil reservoir group-level model, establish the sandstone group model, and is also used to establish the unit-level model within the sandstone group model, and is also used to determine the scales of the three directions of the 3D grid at different scales, and the update and iteration rounds for establishing each level of model.

[0060] The first quality control module is used to perform quality control on the singularities of the geological model based on the standard continental sedimentation pattern. If the vertical correlation coefficient of the eigenvalue in any grid of the oil reservoir group-level model is less than 90% at a grid accuracy of 100m * 100m, the quality control fails and comprehensive adjustment is required until the quality control passes. After quality control and iterative update, the first model for subsequent quality control is obtained.

[0061] Preferably, the second unit includes:

[0062] The conversion module is used to apply the depth domain velocity model to perform time-depth conversion on the collected 3D seismic data volume and seismic geological interpretation results to the depth domain, obtaining the seismic depth and horizon depth for subsequent quality control.

[0063] The second quality control module is used to perform calibration quality control on the oil reservoir group-level boundaries in the first model based on the seismic depth and horizon depth; if the error rate * 100% of the horizon depth and well logging stratification depth of the oil reservoir group layer data greater than or equal to 30m in any grid is greater than 5%, the quality control fails, and the calibrated position of the well corresponding to the well logging data of the grid is an unusable well. If the error rate is greater than or equal to 1% and less than or equal to 5%, the well logging curve oil reservoir group-level stratification depth is adjusted. If the error rate is less than 1%, the quality control passes without any adjustment and update, and the oil reservoir group-level model and all well data are updated to obtain the second model.

[0064] Preferably, the third unit includes:

[0065] The identification module is used to identify the well logging curves of the sandstone group, establish the sandstone group-level model and threshold criteria, and then obtain the standard layer to calculate the correction amount to achieve automatic correction and modification of the sandstone group-level boundaries.

[0066] The coarsening module is used to coarsen the spatial grid of a single unit, perform quality control using the maximum error control method under the constraint of the standard continental sedimentation pattern, and identify the first-round large boundary of the abnormal unit body.

[0067] A carving module, which uses the abnormal unit positions of the first-round large grid as input data to perform automatic discrimination and quality control of tomograms after refining the grid, and carves the unit-level tomogram boundaries using the quality control results;

[0068] A third quality control module, which is used to complete the quality control of unit-level boundary calibration by using the standard continental sedimentation pattern constraint, calculate the minimum sampling interval correction amount through variable grids, and complete the correction or elimination of abnormal wells; if the error rate * 100% of the sand body units with a thickness greater than or equal to 2m and the logging stratification depth in any grid is greater than 5%, the quality control fails, and the calibrated position of the well corresponding to the logging data of the grid is an unusable well. If the error rate is greater than 1% and less than or equal to 5%, the logging curve stratification depth is adjusted. If the error rate is less than or equal to 1%, the quality control passes; finally, the unusable wells in the second model are eliminated to obtain the third model.

[0069] Preferably, the fourth unit includes:

[0070] A fourth quality control module, which uses the random forest algorithm to realize the identification and classification of different types of superimposed sand bodies inside the unit through machine learning, obtains the abnormal sand body identification area, then locks the abnormal wells based on the abnormal sand body identification area, and finally eliminates the abnormal wells from the third model to obtain the fourth model.

[0071] Advantages of the present invention:

[0072] 1. The quality control method for calibrating formation unit boundaries based on the random forest rule under the constraint of the standard continental sedimentation pattern of the present invention is based on the fidelity algorithm modeling and uses machine learning means to improve efficiency. It effectively solves the three problems of low efficiency, low accuracy, and inability to adapt to massive data in the identification of unit boundaries in oilfield development, realizes the rapid automatic comparison, screening, and calibration of the errors of formation unit boundaries at all levels from the oil reservoir group level, sandstone group to unit level, greatly improves the accuracy and efficiency of formation correlation review work, and can be widely applied in the basic geological research field with high timeliness requirements to support the precise development of conventional, unconventional, and new energy sources, and has good application and promotion prospects;

[0073] 2. The present invention realizes the rapid automatic comparison, screening, and calibration of the errors of formation unit boundaries at all levels from the oil reservoir group level, sandstone group to unit level, greatly improves the accuracy and efficiency of formation correlation review work, and can be widely applied in the basic geological research field with high timeliness requirements.

[0074] Other features and advantages of the present invention will be described in the following description, and, in part, will be obvious from the description, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structure pointed out in the description and the drawings. Description of the Drawings

[0075] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0076] Figure 1 The flowchart of a method for calibrating and quality controlling the boundaries of learning stratigraphic units of the present invention is shown;

[0077] Figure 2 The block diagram of a system for calibrating and quality controlling the boundaries of learning stratigraphic units of the present invention is shown. Detailed implementation manners

[0078] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0079] A method for calibrating and quality controlling the boundaries of learning stratigraphic units, as Figure 1 shown, includes the following steps:

[0080] S1: Obtain the basic data of all wells in the study area; S2: According to the basic data of all wells, establish a first model using the Kriging algorithm; S3: Calibrate and quality control the boundaries at the oil reservoir group level in the first model, complete the update of the oil reservoir group level model and all well data, and obtain a second model; S4: Based on the second model, perform unit-level boundary calibration and quality control, complete the correction or elimination of abnormal wells, and obtain an updated unit-level model as the third model; S5: Based on the third model, complete the identification of abnormal sand body areas within the unit and lock abnormal wells according to random forest machine learning. After eliminating the abnormal wells, obtain an updated model as the fourth model; S6: Extract the unit-level models one by one from the fourth model to obtain the corrected model for each unit. Quality control each corrected unit model with the standard layer of the adjacent sandstone group to obtain a sandstone group level boundary calibration model; S7: Based on the sandstone group level boundary calibration model and the depth domain seismic model, perform quality control to obtain the final development geological model as the fifth model.

[0081] It should be noted that Kriging is an exact estimation algorithm. From the solution system, it can be seen that the values of the estimated random field at the sample points are consistent with the corresponding observed values. The advantage of this property is that the Kriging estimation is always close to the observed values and will not deviate too far from the actual situation. However, the estimated values are always between the observed values and cannot simulate drastic changes. Similar to Gaussian process regression, Kriging does not require the samples to follow a specific probability distribution. However, in practice, when the samples do not have skewed data, Kriging often has good results.

[0082] Furthermore, in S1, it includes: determining the four-point coordinates of the study area; collecting the layer data and logging curve data of the target layer drilled in the entire study area; collecting the 3D seismic data volume that completely covers the coordinate range of the study area; collecting the depth-domain velocity model of the study area; collecting the seismic geological interpretation results of the study area.

[0083] Furthermore, in S2, it includes the following steps:

[0084] S201: Apply the logging curve data and development data to interpolate and establish an oil reservoir group-level model using the Kriging algorithm; S202: Refine the 3D grid of the oil reservoir group-level model to establish a sand group model; S203: Establish a unit (sub-layer) level model within the sand group model according to the application conditions and applicable problems of the disjunctive Kriging algorithm; S204: Determine the scales of the three directions of the 3D grid at different scales and the number of update and iteration rounds for establishing each level of model; S205: Based on the singularity quality control of the geological model of the standard continental sedimentation pattern, if the vertical correlation coefficient of the eigenvalue (the type and order of magnitude of the eigenvalue are determined according to the actual situation) in any grid of the oil reservoir group-level model is lower than 90% at a grid accuracy of 100m * 100m, then it fails the quality control and needs to be comprehensively adjusted until it passes the quality control. After quality control and iterative update, the first model for subsequent quality control is obtained.

[0085] It should be noted that in the modeling and interpolation stage of the present invention, the disjunctive Kriging algorithm is adopted, and the principle is as follows: Kriging is a method of weighted average estimation of an unknown function for a known function. Its prediction theory is close to linear regression, so there is also a BLUP theory similar to the Gauss-Markov theorem. Express the estimation of variables in the random field as a linear system containing random errors, then BLUP can be expressed as selecting the parameters of the linear system to minimize the variance between the estimated value and the true value.

[0086] Ordinary Kriging is the BLUP of the random field, but when there is a non-linear relationship between the random field and the exponential set, the linear estimation result is often not optimal. Disjunctive Kriging generalizes the weight coefficient in ordinary Kriging to a function, thus realizing the non-linear estimation of the random field. Disjunctive Kriging can be defined as the following optimization problem:

[0087]

[0088] wherein is the estimator at s0 to be estimated, the function f is a pre-given non-linear function, and Y(s i ) is the data at the known data point s i (i = 1, 2,..., n). Usually when the indicator function is given, disjunctive kriging is also called indicator kriging. Solving the given function makes become the orthogonal projection of the true value Y in the vector space composed of f[Y(s i )]. Thus, the mathematical formulation of the above problem is as follows:

[0089]

[0090] wherein is the mathematical expectation of the difference between the estimator of the function at s0 and the orthogonal projection of the true value of the function at the actual data at s j in the vector space. Y(s0) is the estimator of the function at s0, Y(s i ) is the true value of the actual data of the function at s j , Y(s i ) is the orthogonal projection of the true value of the function in the vector space and the actual data at s j , is the summation operation of the mathematical expectation of the true value of the function at s i and the orthogonal projection in the vector space at s j .

[0091] Disjunctive kriging requires that the sample set is of the same factor. However, in applications, it is usually directly assumed that the sample set follows a joint normal distribution. Expanding the function f with the N-order Hermite polynomial can transform the above formula into the following form:

[0092]

[0093] wherein, η ik is the k-th order Hermite polynomial, and f ik is the undetermined parameter satisfying the following relationship:

[0094]

[0095] wherein, ρ ij is the correlation coefficient of Y(s i ) and Y(s j ), and b kThe expansion coefficient of the k-th order Hermite polynomial. For the disjunctive Kriging problem after the Hermite polynomial expansion, ordinary Kriging is used to solve for b in each Hermite polynomial expansion k and f ik The specific solution process depends on numerical calculations. Considering the computational complexity, the Hermite polynomial expansion series is usually not too large. The disjunctive Kriging variance is obtained as follows:

[0096]

[0097] In the formula, σ 2 is the disjunctive Kriging variance, C is the coefficient of the Hermite polynomial, b k is the expansion coefficient of the k-th order Hermite polynomial, and ρ ij is the correlation coefficient between Y(s i ) and Y(s j ).

[0098] Kriging method is widely used for spatial interpolation of various observations and can be used as a surrogate model to interpolate the limited simulation results in numerical experiments of engineering problems. In other words, if a deterministic simulation method, such as the finite element method, is used globally for the problem, it will consume a large amount of computing resources and lead to low timeliness. However, it can be efficiently interpolated globally by using the Kriging method, and only the results of local individual points need to be simulated to achieve this.

[0099] It should be further noted that in the quality control method for calibrating the boundaries of stratigraphic units based on the random forest rule under the constraint of the standard continental sedimentation model, the machine learning method used to achieve the rapid discrimination and classification of sand body units and thus the calibration of unit boundaries is as follows: Random Forest is a supervised machine learning algorithm constructed by the decision tree algorithm and is used by data scientists to solve regression and classification problems. Random Forest uses ensemble learning, combining many classifiers to provide solutions for complex problems. Random Forest is an algorithm that integrates multiple or various decision trees through the idea of ensemble learning. Its basic unit is still the decision tree, and its essence belongs to the ensemble learning method in machine learning. From an intuitive perspective, each decision tree is a classifier (assuming we are dealing with a classification problem now). Then, for an input sample, N trees will have N classification results. Random Forest integrates all the classification voting results and designates the category with the most votes as the final output, which is the simplest Bagging idea. The advantages of Random Forest are as follows: First, Random Forest is good at dealing with high-dimensional data because the subsets of features are randomly selected. Second, the performance or effect of Random Forest is better than that of a single decision tree algorithm. Third, when the amount of data increases sharply for distributed machine learning, the training of each tree is an independent process. If it is a single machine for parallelization or a cluster for distributed training, big data training can be appropriately processed at this time. Fourth, because Random Forest selects features, the influence between features will be reduced during the learning process.

[0100] Further, in S3, the following steps are included:

[0101] S301: Apply the depth-domain velocity model to transform the collected 3D seismic data volume and seismic geological interpretation results from time domain to depth domain through time-depth conversion to obtain seismic depth and horizon depth for subsequent quality control; S302: Calibrate and control the boundaries of oil reservoir group levels in the first model based on the seismic depth and horizon depth. If the error rate of the horizon depth and well logging stratification depth of the oil reservoir group layer data greater than or equal to 30m in any grid ((error value / vertical depth top depth) * 100%) is greater than 5%, the quality control fails, and the well location corresponding to this grid is designated as an unavailable well. If the error rate is greater than or equal to 1% and less than or equal to 5%, adjust the well logging curve oil reservoir group level stratification depth. If the error rate is less than 1%, pass the quality control without any adjustment and update, and complete the update of the oil reservoir group level model and all well data to obtain the second model.

[0102] Further, in S4, the following steps are included:

[0103] S401: Identify the well logging curves of the sandstone group through machine learning, establish the sandstone group-level model and threshold standard, and then obtain the standard layer calculation correction amount to realize the automatic correction and correction of the sandstone group-level boundary; S402: Coarsen the spatial grid of a single unit (small layer), apply the maximum error control method under the constraint of the standard continental sedimentary model for quality control, and identify the first round of large boundaries of the abnormal unit body; S403: Use the position of the abnormal unit body of the first round of large grid as input data, perform automatic fault identification quality control after the refined grid, and use the quality control results to carve the unit-level fault boundary. The data within the fault does not participate in the subsequent quality control and correction processing; S404: Use the standard continental sedimentary model to constrain the completion of the unit The level boundary is calibrated for quality control, and the minimum sampling interval correction amount is calculated by changing the grid to complete the correction or elimination of abnormal wells. In this embodiment, the quality control input data has reached the minimum sampling interval, and the error rate can be directly calculated by adaptive subtraction; if the error rate of the sand body unit and the logging layer depth greater than or equal to 2m in any grid (error value / vertical depth top depth)*100% is greater than 5%, the quality control fails, and the well corresponding to the grid is marked as an unusable well. If the error rate is greater than 1% and less than or equal to 5%, the logging curve layer depth is adjusted. If the error rate is less than or equal to 1%, the quality control is passed; S405: Eliminate the unusable wells in the second model to obtain the third model.

[0104] It should be noted that in S401, the sandstone group-level boundary calibration quality control constrained by the standard terrestrial sedimentary model, the quality control object is the sandstone group model in the second model, the quality control reference is the standard layer data in the oil layer group, and the standard layer data here also needs to be converted to the depth domain to complete the quality control to obtain the calculated correction amount. Therefore, it is necessary to obtain the standard layer calculation correction amount to realize the automatic correction and correction of the sandstone group-level boundary, and then the static step S402 is performed.

[0105] It should be further explained that in S403, the unit-level fault boundaries are some offset parts that appear in the unit (sub-layer) spatial interface. They need to be identified first, and the influence of the fault on the spatial distribution of the unit should be eliminated (that is, the corresponding abnormal wells in the subsequent fault area should be eliminated), and then the anomalies in other non-displaced areas should be processed (that is, the corresponding abnormal wells in the subsequent abnormal sand body area should be eliminated).

[0106] Furthermore, in step S5, the following steps are included:

[0107] S501: Use the random forest algorithm to achieve the identification and classification of different types of superimposed sand bodies within the unit through machine learning, and obtain the abnormal sand body identification area. In addition, the type of characteristic values ​​and discrimination criteria selected for dimensionality reduction identification should be determined in the sample training stage, and the timeliness requirement should be considered in dimensionality increase identification. At the same time, the sand body grade discrimination criteria should be determined based on development data and practical experience; S502: Lock the abnormal wells based on the abnormal sand body identification area; S503: Eliminate the abnormal wells from the third model to obtain the fourth model.

[0108] Further, in step S6, taking the unit models extracted one by one from the fourth model as input data, the quality control of the sand body unit boundary calibration is completed by using the standard continental sedimentation pattern constraint. Under the discrimination criteria determined by the method of iterative adaptive subtraction to reduce the residual in the per-unit quality control, the adjustment that meets the highest residual tolerance threshold is completed to obtain the corrected model Model for each unit layer1 , Model layer2 , …, Model layern . Using the standard continental sedimentation pattern, each unit model Model obtained layer1 , Model layer2 , …, Model layern is quality-controlled with the standard layer of the nearest sandstone group. After passing the quality control, a sandstone group-level boundary calibration model Model for the multi-unit merger is formed layer(1+2+…) , …, Model layer[…+(n-1)+n] .

[0109] Further, in S7, using the standard continental sedimentation pattern, the sandstone group-level models Model layer(1+2+…) , …, Model layer[…+(n-1)+n] ; are quality-controlled with the depth-domain seismic model. After passing the quality control, a final merged oil reservoir group-level boundary calibration model Model f (i.e., the fifth model) outputs the final development geological model.

[0110] It should be noted that the final calibration model Model obtained by iterative quality control using the standard continental sedimentation pattern f can be used to quickly realize the rapid correction, fusion analysis or 4D monitoring of well logging, development and other types of data at the oil reservoir group level, sandstone group level, and unit level.

[0111] The following is an example of a certain development area of a certain oilfield in combination with S1-S7. The area of the development area of this oilfield (i.e., the research area) is 120 km 2 , the main target layer is the Sa-Pu-Gao oil reservoir group. At present, the comprehensive water cut exceeds 95%. The average formation thickness of the overall Sartu oil reservoir group is about 110 m, the burial depth ranges from 890 m to 1000 m, it is internally divided into 3 oil reservoir groups and 9 sandstone groups, 33 units can be divided, there are 10,195 effective modeling well logging data, and 516,880 formation boundary data.

[0112] Corresponding to S1, within the four-point coordinates of this development area, the length of the longwall is 600 km 2The three-dimensional seismic data can be effectively covered. The vertical sampling interval of the three-dimensional seismic data is 1ms, the bin size is 20m*20m, and the Sartu oil layer group is about 850ms to 930ms on the seismic section. Corresponding to S3, since the interval of the most dense well network in the study area is 125m, in order to ensure that there is at least one well data in each coarsening grid, the grid size in the x-direction and y-direction of the initial iteration should not be less than 200m. In the embodiment, if the oil layer group horizon data of 30m or more in a grid is greater than or equal to 10m, the grid size of the oil layer group is about 10m. Depth And the logging stratification depth error rate (error value / vertical depth top depth) * 100%, if it is greater than 5%, the quality control fails, and the well corresponding to the grid is marked as an unusable well. If the error rate is greater than or equal to 1% and less than or equal to 5%, the logging curve oil layer group level stratification depth is updated and adjusted. If the error rate is less than 1%, the quality control passes without any adjustment or update. The oil layer group level model and all well data are updated to obtain the second model. In the embodiment, the 60m thick Sartu oil layer group has an error rate of 0.6m corresponding to 1%. If the oil layer group level quality control is less than this well error, it passes the quality control. If the well error range is less than 3m, it needs to be adjusted. If it is greater than 3m, the well at the quality control point position is marked as an unusable well.

[0113] The above whole set of actual data development and application is based on fidelity algorithm modeling and uses machine learning to improve efficiency. It invents an effective solution to the three problems of low efficiency, low accuracy and inability to adapt to massive data in manual identification of unit boundary delineation in oil field development. It realizes rapid and automatic comparison, screening and calibration of errors of stratigraphic unit boundaries at all levels from oil layer group level, sandstone group to unit level, which greatly improves the accuracy and efficiency of stratigraphic comparison and audit work, and can be widely used in the field of basic geological research with high timeliness requirements.

[0114] A quality control system for learning stratigraphic unit boundary calibration, such as Figure 2As shown, it includes a basic unit, a first unit, a second unit, a third unit, a fourth unit, an extraction unit, and a fifth unit. The basic unit is used to obtain the basic data of all wells in the study area; the first unit is used to establish a first model using the Kriging algorithm based on the basic data of all wells; the second unit is used to calibrate and quality-control the oil reservoir group-level boundaries in the first model, complete the update of the oil reservoir group-level model and all well data, and obtain a second model; the third unit is used to perform unit-level boundary calibration and quality-control based on the second model, complete the correction or elimination of abnormal wells, and obtain an updated unit-level model, which is the third model; the fourth unit is used to identify abnormal sand body areas and lock abnormal wells based on the third model according to random forest machine learning, and obtain an updated model after eliminating abnormal wells as the fourth model; the extraction unit is used to extract unit models one by one from the fourth model, obtain the corrected model for each unit, and perform quality-control on each corrected unit model with the standard sandstone group standard layer of the adjacent layer to obtain a sandstone group-level boundary calibration model; the fifth unit is used to perform quality-control based on the sandstone group-level boundary calibration model and the depth-domain seismic model to obtain the final development geological model as the fifth model.

[0115] Furthermore, the basic unit includes a positioning module, a first collection module, a second collection module, a third collection module, and a fourth collection module. The positioning module is used to determine the four-point coordinates of the study area; the first collection module is used to collect the layered data of the target layer drilled in the entire study area and the logging curve data; the second collection module is used to collect the three-dimensional seismic data volume that completely covers the coordinate range of the study area; the third collection module is used to collect the depth-domain velocity model of the study area; the fourth collection module is used to collect the seismic geological interpretation results of the study area.

[0116] Furthermore, the first unit includes an interpolation module, a construction module, and a first quality-control module. The interpolation module is used to establish an oil reservoir group-level model by interpolation using the logging curve data and development data with the Kriging algorithm; the construction module is used to refine the three-dimensional grid of the oil reservoir group-level model, establish a sandstone group model, and is also used to establish a unit (sub-layer) level model within the sandstone group model, and is also used to determine the scales of the three directions of the three-dimensional grid at different scales and the update iteration rounds for establishing each level of model; the first quality-control module is used to perform singularity quality-control on the geological model based on the standard continental sedimentation pattern. If the vertical correlation coefficient of the eigenvalue in any grid of the oil reservoir group-level model is lower than 90% at a grid accuracy of 100m * 100m, the quality-control fails and comprehensive adjustment is required until the quality-control passes. After quality-control and iterative update, the first model for subsequent quality-control is obtained.

[0117] Further, the second unit includes a conversion module and a second quality control module. The conversion module is used to apply the depth-domain velocity model to transform the collected 3D seismic data volume and seismic geological interpretation results from time domain to depth domain through time-depth conversion, obtaining seismic depth and horizon depth for subsequent quality control. The second quality control module is used to calibrate and control the boundaries of oil reservoir groups in the first model based on the seismic depth and horizon depth. If the error rate * 100% between the horizon depth and the logging stratification depth of the oil reservoir group layer data greater than or equal to 30m in any grid is greater than 5%, the quality control fails, and the calibration position of the well corresponding to this grid is defined as an unusable well. If the error rate is greater than or equal to 1% and less than or equal to 5%, the logging curve stratification depth of the oil reservoir group is adjusted. If the error rate is less than 1%, the quality control passes without any adjustment or update, and the oil reservoir group-level model and all well data are updated to obtain the second model.

[0118] Further, the third unit includes an identification module, a coarsening module, a carving module, and a third quality control module. The identification module is used to identify the logging curves of sandstone groups, establish the sandstone group-level model and threshold standards, and then obtain the standard layer to calculate the correction amount to achieve automatic correction and modification of the sandstone group-level boundaries. The coarsening module is used to coarsen the spatial grid of a single unit (sub-layer), and use the maximum error control method under the constraint of the standard continental sedimentation pattern for quality control to identify the first-round large boundary of abnormal unit bodies inside the unit. The carving module is used to take the first-round large-grid abnormal unit body position as the input data, perform automatic discrimination and quality control of faults after refining the grid, and use the quality control results to carve the unit-level fault boundary. The third quality control module is used to complete the calibration and quality control of the unit-level boundaries using the constraint of the standard continental sedimentation pattern, calculate the minimum sampling interval correction amount through variable grids, and complete the correction or rejection of abnormal wells. If the error rate * 100% between the sand body unit greater than or equal to 2m and the logging stratification depth in any grid is greater than 5%, the quality control fails, and the calibration position of the well corresponding to this grid is defined as an unusable well. If the error rate is greater than 1% and less than or equal to 5%, the logging curve stratification depth is adjusted. If the error rate is less than or equal to 1%, the quality control passes. Finally, the unusable wells in the second model are removed to obtain the third model.

[0119] Further, the fourth unit includes a fourth quality control module, which is used to use the random forest algorithm to realize the identification and classification of different types of stacked sand bodies inside the unit through machine learning, obtain the abnormal sand body identification area, then lock the abnormal wells based on the abnormal sand body identification area, and finally remove the abnormal wells from the third model to obtain the fourth model.

[0120] It should be noted that for the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial descriptions of the method embodiments. The various units and modules of the formation unit boundary calibration quality control system are only divided according to the functional logic, but are not limited to the above division as long as the corresponding functions can be achieved. In addition, the specific names of the various units are only for the convenience of mutual distinction and are not used to limit the protection scope of the present invention.

[0121] Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A quality control method for calibrating the boundaries of stratigraphic units, characterized in that, Including the following steps: Obtain the basic data of all wells in the study area; Based on the basic data of all wells, establish the first model using the Kriging algorithm; Calibrate and quality control the oil reservoir group-level boundaries in the first model, complete the update of the oil reservoir group-level model and all well data, and obtain the second model; Based on the second model, conduct unit-level boundary calibration and quality control, complete the correction or elimination of abnormal wells, and obtain the updated unit-level model, which is the third model; Based on the third model, complete the identification of abnormal sand body areas within the unit and lock abnormal wells according to random forest machine learning. After eliminating the abnormal wells, obtain the updated model as the fourth model; Extract the unit-level models one by one from the fourth model to obtain the corrected models for each unit. Quality control each corrected unit model with the standard layer of the adjacent sandstone group to obtain the sandstone group-level boundary calibration model; Based on the sandstone group-level boundary calibration model and the depth-domain seismic model, conduct quality control to obtain the final development geological model as the fifth model.

2. The quality control method for calibrating the boundary of a learning stratigraphic unit according to claim 1, characterized in that Obtain the basic data of all wells in the study area, including: Determine the four-point coordinates of the study area; Collect the layer data and logging curve data of the target layer drilled in the entire study area; Collect the three-dimensional seismic data volume that completely covers the coordinate range of the study area; Collect the depth-domain velocity model of the study area; Collect the seismic geological interpretation results of the study area.

3. A quality control method for calibrating the boundaries of stratigraphic units according to claim 2, characterized in that, Based on the basic data of all wells, establish the first model using the Kriging algorithm, including the following steps: Apply the logging curve data and layer data to interpolate and establish the oil reservoir group-level model using the Kriging algorithm; Refine the three-dimensional grid of the oil reservoir group-level model to establish the sandstone group model; Establish the unit-level model within the sandstone group model; Determine the scales of the three directions of the three-dimensional grid at different scales, and the number of update and iteration rounds for the establishment of each level model; Based on the singularity quality control of the geological model of the standard continental sedimentation pattern, if the vertical correlation coefficient of the eigenvalue in any grid of the oil reservoir group-level model is less than 90% at a grid accuracy of 100m * 100m, the quality control fails and comprehensive adjustment is required until the quality control passes. After quality control and iterative update, obtain the first model for subsequent quality control.

4. A quality control method for calibrating the boundary of a learning stratigraphic unit according to claim 3, characterized in that Calibrate and quality control the oil reservoir group-level boundaries in the first model, complete the update of the oil reservoir group-level model and all well data, and obtain the second model, including the following steps: Apply the depth-domain velocity model to time-depth convert the collected three-dimensional seismic data volume and seismic geological interpretation results to the depth domain to obtain the seismic depth and horizon depth for subsequent quality control; Based on the seismic depth and horizon depth, calibrate and quality control the oil reservoir group-level boundaries in the first model; If the error rate * 100% of the horizon depth and logging layer depth of the oil reservoir group layer data greater than or equal to 30m in any grid is greater than 5%, the quality control fails, and the calibrated position of the well corresponding to the logging data of the grid is an unusable well. If the error rate is greater than or equal to 1% and less than or equal to 5%, adjust the oil reservoir group-level layer depth of the logging curve. If the error rate is less than 1%, the quality control passes and no adjustment is required for update. Complete the update of the oil reservoir group-level model and all well data to obtain the second model.

5. A quality control method for calibrating the boundaries of stratigraphic units according to claim 3, characterized in that Perform unit-level boundary calibration quality control based on the second model, complete the correction or rejection of abnormal wells, and obtain the updated unit-level model, which is the third model, including the following steps: Identify the logging curves of the sandstone formations, establish the sandstone formation-level model and threshold criteria, and then obtain the standard layer to calculate the correction amount to achieve automatic correction and modification of the sandstone formation-level boundary; Coarsen the spatial grid of a single unit, perform quality control using the maximum error control method under the constraint of the standard continental sedimentation model, and identify the first-round large boundaries of abnormal unit bodies; Take the first-round large-grid abnormal unit body positions as input data, perform quality control for automatic fault discrimination after refining the grid, and use the quality control results to carve the unit-level fault boundaries; Use the standard continental sedimentation model constraint to complete the unit-level boundary calibration quality control, calculate the minimum sampling interval correction amount through variable grids, and complete the correction or rejection of abnormal wells; If the sand body unit with a size greater than or equal to 2m in any grid and the logging layer depth error rate * 100% is greater than 5%, the quality control fails, and the calibrated position of the well corresponding to the logging data of the grid is an unusable well. If the error rate is greater than 1% and less than or equal to 5%, adjust the logging curve layer depth. If the error rate is less than or equal to 1%, the quality control passes; Reject the unusable wells in the second model to obtain the third model.

6. A quality control method for calibrating the boundaries of stratigraphic units according to any one of claims 3-5, characterized in that Based on the third model, complete the identification of abnormal sand body areas and lock abnormal wells according to random forest machine learning. After completing the rejection of abnormal wells, obtain the updated model as the fourth model, including the following steps: Use the random forest algorithm to achieve the identification and classification of different types of stacked sand bodies inside the unit through machine learning, and obtain the abnormal sand body identification area; Lock abnormal wells based on the abnormal sand body identification area; Reject the abnormal wells from the third model to obtain the fourth model.

7. A quality control system for calibrating the boundaries of stratigraphic units, characterized in that, Including: The basic unit is used to obtain the basic data of all wells in the study area; The first unit is used to establish the first model using the Kriging algorithm based on the basic data of all wells; The second unit is used to perform calibration quality control on the oil reservoir formation-level boundaries in the first model, complete the update of the oil reservoir formation-level model and all well data, and obtain the second model; The third unit is used to perform unit-level boundary calibration quality control based on the second model, complete the correction or rejection of abnormal wells, and obtain the updated unit-level model, which is the third model; The fourth unit is used to complete the identification of abnormal sand body areas inside the unit, lock abnormal wells based on the third model according to random forest machine learning, and obtain the updated model as the fourth model after completing the rejection of abnormal wells; The extraction unit is used to extract the unit-level models one by one from the fourth model to obtain the corrected models for each unit, and perform quality control on the corrected unit models for each unit with the standard layers of adjacent sandstone formations to obtain the sandstone formation-level boundary calibration model; The fifth unit is used to perform quality control based on the sandstone formation-level boundary calibration model and the depth-domain seismic model to obtain the final development geological model as the fifth model.

8. A quality control system for calibrating the boundaries of stratigraphic units according to claim 7, characterized in that, The basic unit includes: The positioning module is used to determine the four-point coordinates of the study area; The first collection module is used to collect the formation data and logging curve data of the target layer drilled in the entire study area; The second collection module is used to collect the three-dimensional seismic data volume that completely covers the coordinate range of the study area; The third collection module is used to collect the depth-domain velocity model of the research area; The fourth collection module is used to collect the seismic-geological interpretation results of the research area.

9. The quality control system for calibrating the boundary of a stratigraphic unit according to claim 8, wherein, The first unit includes: The interpolation module is used to establish an oil reservoir group-level model by interpolation using logging curve data and development data with the Kriging algorithm; The construction module is used to refine the 3D grid of the oil reservoir group-level model, establish a sand reservoir group model, and also establish a unit-level model within the sand reservoir group model. It is also used to determine the scales in three directions of the 3D grid at different scales and the number of update and iteration rounds for establishing each level of model; The first quality control module is used for quality control of the singularities of the geological model based on the standard continental sedimentation pattern. If the vertical correlation coefficient of the eigenvalue in any grid of the oil reservoir group-level model is less than 90% at a grid accuracy of 100m * 100m, the quality control fails and comprehensive adjustment is required until the quality control passes. After quality control and iterative update, the first model for subsequent quality control is obtained.

10. A quality control system for calibrating formation unit boundaries in learning, as claimed in claim 8, wherein The second unit includes: The conversion module is used to convert the collected 3D seismic data volume and seismic-geological interpretation results to the depth domain through time-depth conversion using the depth-domain velocity model, obtaining the seismic depth and horizon depth for subsequent quality control; The second quality control module is used for calibration quality control of the oil reservoir group-level boundaries in the first model based on the seismic depth and horizon depth; if the error rate * 100% of the horizon depth and logging stratification depth of the oil reservoir group layer data greater than or equal to 30m in any grid is greater than 5%, the quality control fails, and the calibration position of the well corresponding to the logging data of the grid is an unusable well. If the error rate is greater than or equal to 1% and less than or equal to 5%, the logging curve oil reservoir group-level stratification depth is adjusted. If the error rate is less than 1%, the quality control passes without any adjustment and update, and the oil reservoir group-level model and all well data are updated to obtain the second model.

11. A quality control system for calibrating the boundaries of stratigraphic units according to claim 8, characterized in that, The third unit includes: The identification module is used to identify the logging curves of the sandstone group, establish a sandstone group-level model and threshold criteria, and then obtain the standard layer to calculate the correction amount to realize automatic correction and modification of the sandstone group-level boundaries; The coarsening module is used to coarsen the spatial grid of a single unit, and perform quality control using the maximum error control method under the constraint of the standard continental sedimentation pattern to identify the first-round large boundaries of abnormal unit bodies; The carving module is used to use the first-round large-grid abnormal unit body position as input data, perform quality control for automatic discrimination of faults after refining the grid, and use the quality control results to carve the unit-level fault boundaries; The third quality control module is used to complete the calibration quality control of the unit-level boundaries using the constraint of the standard continental sedimentation pattern, calculate the minimum sampling interval correction amount through variable grids, and complete the correction or elimination of abnormal wells; if the error rate * 100% of the sand body unit and logging stratification depth greater than or equal to 2m in any grid is greater than 5%, the quality control fails, and the calibration position of the well corresponding to the logging data of the grid is an unusable well. If the error rate is greater than 1% and less than or equal to 5%, the logging curve stratification depth is adjusted. If the error rate is less than or equal to 1%, the quality control passes; finally, the unusable wells in the second model are eliminated to obtain the third model.

12. A quality control system for calibrating formation unit boundaries in learning according to claim 8, characterized in that, The fourth unit includes: The fourth quality control module is used to realize the identification and classification of different types of stacked sand bodies inside the unit through machine learning using the random forest algorithm, obtain the abnormal sand body identification area, then lock the abnormal wells based on the abnormal sand body identification area, and finally remove the abnormal wells from the third model to obtain the fourth model.

Images