A drilling leakage prediction model establishment method, a drilling leakage prediction method and a device

Abstract

The application discloses a drilling leakage prediction model establishment method, a drilling leakage prediction method and a device. The drilling leakage prediction model establishment method comprises the following steps: for a drilled well with leakage mark data, extracting a seismic attribute value corresponding to each depth value of the leakage mark from each set seismic attribute body according to a synthetic record of the well, so as to obtain a sample containing the leakage mark corresponding to the depth value and each seismic attribute value; the leakage mark data comprises a plurality of depth values of the well and the leakage mark corresponding to each depth value; a plurality of samples are combined into a sample set, the sample set is used for training a set classification model, so as to obtain a classification model used for predicting drilling leakage, and the classification model comprises at least one sub-classification model. The drilling leakage can be reasonably predicted before drilling.

Description

A drilling loss prediction model, method and device Technical Field

[0001] This invention relates to the field of oil and gas drilling engineering technology, and in particular to a drilling loss prediction model establishment, drilling loss prediction method and device. Background Technology

[0002] Carbonate formations often have well-developed pores, fractures, and cavities, resulting in complex leakage pathways. During drilling, both permeable and non-returnable losses can occur, severely hindering safe and efficient drilling. Currently, traditional drilling loss analysis methods are all "post-hoc" methods, unable to achieve pre-drilling "prediction"; or the predicted pre-drilling loss results have significant uncertainties. Summary of the Invention

[0003] Under current engineering technology conditions, only seismic data is three-dimensional data that can be directly measured. Reasonable application of seismic data can achieve true "prediction". However, the difficulty in using seismic data to predict well leakage is that obtaining the desired result through seismic waves is an inversion process. The inversion result may have infinitely many solutions. Therefore, how to impose constraints to make the inversion result a unique solution is the primary problem for all technologies that use seismic data as the basic technical means.

[0004] In view of the above problems, the present invention is proposed to provide a drilling leakage prediction model establishment, drilling leakage prediction method and apparatus to overcome or at least partially solve the above problems, and to achieve reasonable prediction of drilling leakage in three-dimensional space.

[0005] In a first aspect, embodiments of the present invention provide a method for establishing a drilling loss prediction model, comprising:

[0006] For a drilled well with missing marker data, the seismic attribute value corresponding to each depth value of the missing marker is extracted from each set seismic attribute volume based on the synthetic record of the well, so as to obtain a sample containing the missing marker corresponding to the depth value and each seismic attribute value. The missing marker data includes multiple depth values ​​of the well and the missing marker corresponding to each depth value.

[0007] The samples mentioned above are combined into a sample set;

[0008] The sample set is used to train a defined classification model to obtain a classification model for predicting well leakage, the classification model comprising at least one sub-classification model.

[0009] Secondly, embodiments of the present invention provide a drilling loss prediction method, including:

[0010] Each set seismic attribute extracted from the seismic data volume is input into the classification model obtained using the above method to obtain the well leakage prediction results.

[0011] Thirdly, embodiments of the present invention provide a drilling loss prediction model establishment apparatus, comprising:

[0012] The sample acquisition module is used to extract the seismic attribute value corresponding to each depth value of the missing marker from each set seismic attribute body based on the synthetic record of the well for a drilled well with missing marker data, so as to obtain a sample containing the missing marker corresponding to the depth value and each seismic attribute value. The missing marker data includes multiple depth values ​​of the well and the missing marker corresponding to each depth value.

[0013] A sample set acquisition module is used to assemble multiple samples acquired by the sample acquisition module into a sample set.

[0014] The model training module is used to train a set classification model using the sample set obtained by the sample set acquisition module to obtain a classification model for predicting well leakage, the classification model including at least one sub-classification model.

[0015] Fourthly, embodiments of the present invention provide a computer program product for drilling loss prediction function, including a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the above-mentioned drilling loss prediction model establishment method, or implements the above-mentioned drilling loss prediction method.

[0016] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:

[0017] (1) The drilling loss prediction model establishment method provided in this embodiment of the invention uses the synthetic record and loss marker data of the drilled well to mark the seismic attributes for loss, and obtains a sample containing the depth value of the drilled well, the loss marker and various seismic attributes. The synthetic record uses well logging data as a bridge connecting the seismic data and the loss data. Multiple samples are combined into a sample set, and the sample set is used to train the set classification model to obtain a classification model for predicting drilling loss. A loss risk prediction model is established by using machine learning algorithm, so that pre-drilling loss prediction can be realized, and it can also provide a reference for well location deployment optimization design.

[0018] (2) The drilling leakage prediction model establishment method provided in this embodiment of the invention uses a variety of seismic attributes that have been marked for leakage in the training samples. That is, it comprehensively analyzes a variety of seismic attributes instead of obtaining a prediction result for each seismic attribute, thus avoiding the diversity of seismic inversion results.

[0019] (3) The drilling leakage prediction model establishment method provided in the embodiments of the present invention includes at least one sub-classification model in the classification model, so that multiple factors can be comprehensively considered in the prediction results using multiple methods, thereby improving the rationality of the final leakage prediction results.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0022] Figure 1 is a flowchart of the drilling loss prediction model establishment method in Embodiment 1 of the present invention;

[0023] Figure 2 is a flowchart illustrating the specific implementation of the synthetic recording well seismic calibration in Embodiment 1 of the present invention;

[0024] Figure 3 is a schematic diagram of the synthetic recording well seismic calibration in Embodiment 1 of the present invention;

[0025] Figure 4 is a flowchart illustrating the specific implementation of classification model verification and optimization in an embodiment of the present invention.

[0026] Figure 5 is a flowchart illustrating the specific implementation of the drilling loss prediction method in Embodiment 2 of the present invention.

[0027] Figure 6 is a schematic diagram of the drilling loss prediction model establishment device in an embodiment of the present invention. Detailed Implementation

[0028] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0029] To address the problem of difficulty in predicting well leakage before drilling in existing technologies, this invention provides a well leakage prediction model establishment, a well leakage prediction method, and an apparatus that can achieve reasonable well leakage prediction before drilling.

[0030] Example 1

[0031] This invention provides a method for establishing a drilling loss prediction model, the process of which is shown in Figure 1 and includes the following steps:

[0032] Step S11: For drilled wells with missing marker data, extract the seismic attribute value corresponding to each depth value of the missing marker from each set seismic attribute volume according to the synthetic record of the well, and obtain a sample containing the missing marker corresponding to the depth value and each seismic attribute value.

[0033] The lost-in-the-well (WIN) marking data includes multiple depth values ​​for the well and a WIN marking corresponding to each depth value. Specifically, the WIN marking can be a WIN probability, i.e., the magnitude of the risk of lost-in-the-well drilling, with a value of 0–1, 0–100, or 0–100%. Optionally, the WIN marking may not be a specific numerical value, but rather the level of risk of lost-in-the-well drilling. Other marking formats are also possible; the specific format is not limited in this embodiment.

[0034] The essence of wellbore leakage is the existence of leakage channels (such as cavities, fractures, and caverns) in the formation. When the fluid column pressure in the wellbore exceeds the formation pressure, leakage will occur. Therefore, the occurrence of leakage is a probabilistic issue. Seismic data reflects the original structural information of the formation. Therefore, by linking seismic data with leakage record data, it is possible to predict the probability of wellbore leakage risk.

[0035] The data sources used in this embodiment of the invention mainly include three parts: 1. Seismic pure wave data, i.e., the original seismic data volume, from which commonly used seismic attribute volumes need to be extracted; 2. Well logging data, mainly used for the synthesis of sonic transit time and density logging curves; 3. Loss data, i.e., the loss marker data of drilled wells, mainly including loss depth and loss amount. The loss amount is either the loss level or the loss risk probability determined based on the loss amount, and the loss risk probability is the probability of loss occurring.

[0036] Seismic data volumes (and seismic attribute volumes extracted from them) and missing data cannot be directly linked because seismic data is time-domain data, while missing data is depth-domain data. Therefore, it is necessary to first create synthetic records for well-seismic calibration, as shown in Figure 2. The specific steps include the following:

[0037] Step S21: Use the well logging curves from the drilled well and the seismic wavelet to perform deconvolution to obtain a synthetic record.

[0038] Specifically, the aforementioned logging curves for drilled wells refer to the sonic transit time curves and density curves extracted from all logging curves of drilled wells.

[0039] The seismic wavelet is either the Rick wavelet or the zero-phase wavelet extracted from the seismic data volume. It can be obtained by first extracting the seismic volume corresponding to the drilled well from the seismic data volume, and then extracting the seismic wavelet from the corresponding seismic volume.

[0040] Step S22: Adjust the synthetic record based on the seismic bodies corresponding to the drilled wells extracted from the seismic data volume.

[0041] The synthetic record is adjusted according to the corresponding seismic body until the two reach maximum similarity. At this point, the depth domain of the well logging curves contained in the synthetic record and the time domain of the seismic body can be effectively matched.

[0042] Synthetic records unify the well logging depth domain and the seismic time domain, so the seismic attribute volume can be marked with omissions using the synthetic records of wells and the omission marker data of wells, thereby obtaining the samples required for training.

[0043] As shown in Figure 3, a synthetic record (time-depth relationship calibration) is obtained by combining the seismic volume corresponding to the well extracted from the seismic data volume with well logging data containing missing data (mainly using sonic transit time curves). This achieves effective matching between the well logging depth domain and the seismic time domain. Seismic attributes can be marked for missing data based on the well logging missing data to obtain training samples. In addition to obtaining the time-depth relationship calibration results, the layer velocity and average velocity of the seismic data can also be obtained.

[0044] Loss markers for wells primarily include loss depth and loss amount, which can be obtained from recorded data in the Final Completion Report (FWR), Daily Drilling Log (DDR), and Materlog. Loss amount can be a loss level or loss risk probability determined based on the loss amount; the loss risk probability is the probability of a loss occurring.

[0045] Taking well leakage prediction in thick carbonate rock formations as an example, the seismic attribute volume set above can be a seismic attribute volume such as variance, time-frequency decay, sweet spot or root mean square amplitude.

[0046] The sweet spot seismic attribute volume is obtained by predicting porosity based on seismic data volume, defining areas with porosity higher than a set value as sweet spot regions.

[0047] It is possible to use Python programming to convert seismic data into information related to seismic reflection waves or rock physics through some mathematical transformation methods (such as complex trace analysis, time-frequency analysis, wave impedance inversion, etc.), thereby obtaining the above-mentioned seismic attribute volume from the seismic data volume. Different seismic attribute volumes reflect the essential properties of the strata from different levels.

[0048] Step S12: Combine multiple samples into a sample set.

[0049] Step S13: Train the set classification model using the sample set to obtain a classification model for predicting well leakage.

[0050] The classification model includes at least one sub-classification model. In some embodiments, it may include at least one of the following sub-classification models:

[0051] Logistic regression subclassification model, decision tree subclassification model, and support vector machine subclassification model.

[0052] In some embodiments, training a given classification model using a sample set specifically includes performing the following iterative training steps on each sub-classification model using the sample set:

[0053] Based on the weight value and predicted omission probability of each seismic attribute, the overall omission probability is determined. The weight value of each seismic attribute is then adjusted based on the overall omission probability and the omission probability marked in the omission label of the sample.

[0054] Repeat the above steps to iteratively train the sub-classification model until the degree of agreement between the currently determined comprehensive omission probability and the labeled omission probability reaches the first set requirement.

[0055] When there is more than one sub-classification model, the specified classification model is trained using a sample set, which may include:

[0056] After training each sub-classification model in the set classification model using the sample set, the weight value of each sub-classification model is iteratively optimized until the predicted omission probability determined by the current weight value of each sub-classification model and the output comprehensive omission probability matches the omission probability marked in the omission label of the sample to the second set requirement.

[0057] The overall omission probability determined based on all sub-classification models can be determined using the following formula:

[0058]

[0059] in, To calculate the overall omission probability; w j p represents the weight value of the j-th seed classification model, where j = 1, 2, ..., m, and m is the number of subclassification models, for example, m = 3; ij Let be the omission probability predicted by the j-th seed classification model based on the i-th earthquake attribute.

[0060] The drilling loss prediction model establishment method provided in Embodiment 1 of this invention utilizes the synthetic logging data and loss marker data of drilled wells to mark seismic attributes for loss, obtaining a sample containing the depth value of the drilled well, loss markers, and various seismic attributes. The synthetic logging data serves as a bridge connecting seismic data and loss data. Multiple samples are combined into a sample set, which is used to train a set classification model to obtain a classification model for predicting drilling loss. A loss risk prediction model is established using machine learning algorithms, enabling pre-drilling loss prediction and providing a reference for well location deployment optimization design.

[0061] The training samples contain a variety of seismic attributes that have been omitted from the labeling, that is, a comprehensive analysis of multiple seismic attributes is performed, rather than obtaining a prediction result for each seismic attribute, thus avoiding the diversity of seismic inversion results.

[0062] The classification model includes at least one sub-classification model, which allows multiple methods to be used to comprehensively consider multiple factors in the prediction results, thereby improving the rationality of the final omission prediction results.

[0063] In some embodiments, new wells that were not involved in training and validation can also be used to test the model, as shown in Figure 4, mainly including:

[0064] Step S41: For other drilled wells with missing marker data, extract the seismic attribute value corresponding to each depth value of the missing marker from each set seismic attribute volume according to the synthetic record of the well, and obtain a test sample containing the missing marker corresponding to the depth value and each seismic attribute value; combine multiple test samples into a test sample set.

[0065] Other drilled wells can be wells newly drilled after model training that have missing data; or wells that are not newly drilled but were not used as sample data during model training. For example, the total sample set can be divided into two subsets: one subset for training samples and one subset for testing samples.

[0066] Step S42: Use the trained classification model and test sample set to predict drilling leakage and obtain the predicted leakage probability for each test sample.

[0067] Specifically, this can include inputting the omission markers (i.e. omission probabilities) contained in each sample in the test sample set into the trained classification model after removing them, and obtaining the predicted omission probability for each test sample based on the model's output.

[0068] Step S43: Determine whether the consistency between the predicted omission probability and the labeled omission probability for each test sample meets the third set requirement.

[0069] If yes, it means that the current classification model has high reliability and can be directly used for region omission prediction; if no, proceed to step S44 to obtain new samples to enrich the existing sample set and further optimize the current classification model.

[0070] Step S44: Add the test sample set to the sample set and train the current classification model using the current sample set.

[0071] The final classification model can be a three-dimensional risk leakage probability prediction model with standard seismic data format (SGY). It can be imported into any working platform that supports SGY data volumes and can arbitrarily slice and truncate the data to achieve regional leakage risk prediction.

[0072] The principles of the three algorithms mentioned above—logistic regression subclassification model, decision tree subclassification model, and support vector machine subclassification model—are as follows:

[0073] ①Logistic Regression

[0074] The basic idea is to fit a logistic / hypothesis function to predict the probability of an event occurring; the predicted probability output will inevitably be between 0 and 1. Since the result of linear regression is generally a continuous value with an uncertain range, a step function, specifically the sigmoid function, can be chosen to convert the continuous value to a value between 0 and 1. The function and probability expression are as follows:

[0075]

[0076] p(y=0|w,x)=1-g(z)

[0077] p(y=1|w,x)=g(z)

[0078] In the formula, e is Euler's constant; p is the probability.

[0079] ② Decision Tree

[0080] The basic idea is to use a tree model to make decisions (classification or regression) on data. The entire process can be viewed as starting from the root node and gradually moving to the leaf nodes. That is, a metric is chosen to calculate the classification results after branching based on different features, and the best result is selected as the root node. This process is repeated to determine the splitting criteria for all nodes. This metric can be described using "entropy (a measure of the uncertainty of a random variable)," as shown in the following formula:

[0081]

[0082] In the formula, n represents the n different discrete values ​​of x. And p i This represents the probability that x takes the value i, and log is the logarithm to the base 2 or e.

[0083] ③Support Vector Machine

[0084] The basic idea is to establish a hyperplane as a decision surface that maximizes the isolation boundary between positive and negative examples, thereby distinguishing different types of data. The hyperplane equation and the discriminant equation are as follows:

[0085] w T x+b=0

[0086] g(x) = w T x+b

[0087] In the formula, w and b are constants to be determined.

[0088] Example 2

[0089] Embodiment 2 of the present invention provides a drilling loss prediction method, comprising:

[0090] Each set seismic attribute extracted from the seismic data volume is input into the classification model obtained using the above method to obtain the well leakage prediction results.

[0091] Taking the drilling loss prediction of a severely leaky, thick carbonate reservoir as an example, the specific implementation process is shown in Figure 5, including the following steps:

[0092] Step S51: For drilled wells with leakage marker data, classify the leakage levels of the drilled wells and determine the leakage risk probability based on the leakage levels.

[0093] Twenty wells with different degrees of leakage were selected, and the leakage levels were classified according to leakage rate and leakage amount. The specific classification results are shown in Table 1. Different leakage levels (leakage rate or leakage amount) correspond to different leakage risk probabilities. The risk probability of no leakage is 0, the risk probability of permeable leakage is 25%, the risk probability of partial leakage is 50%, the risk probability of severe leakage is 75%, and the leakage probability of non-returning leakage is 100% (1).

[0094] Table 1 Classification of Leakage Level

[0095] Classification leakage rate, m 3 / h leakage, m 3 No leakage 0 Permeable leakage <1.59 <38.16 Partial leakage 1.59~7.59 38.16~190.8 Severe leakage 7.59~15.9 190.8~381.6 Regressive leakage >15.9 >381.6 surface

[0096] Step S52: For each drilled well, extract the seismic attribute value corresponding to each depth value from the variance, time-frequency decay, sweet spot or root mean square amplitude seismic attribute volume according to the well's synthetic record, and obtain a sample containing the missing markers corresponding to the depth value and each seismic attribute value.

[0097] Step S53: Add the obtained samples to the training sample subset or the test sample subset.

[0098] The obtained samples can be added to the training sample subset or the test sample subset by random sampling, so that the ratio of the number of samples in the training sample subset to the number of samples in the test sample subset is 3:1.

[0099] Step S54: Train the classification model, which includes a logistic regression subclassification model, a decision tree subclassification model, and a support vector machine subclassification model, using a subset of training samples to obtain a classification model for predicting well leakage.

[0100] Four seismic attributes and leakage probabilities were analyzed separately. The leakage probabilities calibrated from a single well were point data. Through machine learning and planar interpolation methods, leakage risk data at different depths / layers could be obtained, thus achieving three-dimensional risk leakage prediction. At the same depth point, the four attributes yielded four different leakage risk probabilities. A "voting" algorithm was used to perform weight analysis on the probabilities obtained from the four attributes, with different attributes having different weight values. Then, the probabilities were sorted and superimposed according to their different weights to obtain the overall comprehensive risk probability.

[0101] Step S55: Evaluate the trained classification model using a subset of test samples.

[0102] If the accuracy of the verification classification model prediction meets the set requirements, proceed to step S57; if the accuracy of the test classification model prediction does not meet the set requirements, combine the training sample subset and the test sample subset into one sample set, and retrain and optimize the model.

[0103] Step S56: Use the trained classification model and four seismic attributes—variance, time-frequency decay, sweet spot, and root mean square amplitude—to perform two-dimensional or three-dimensional well leakage prediction for the region.

[0104] Embodiment 2 of the present invention realizes pre-drilling risk prediction, reduces the risk of leakage, and has good results.

[0105] Based on the inventive concept of this invention, this embodiment of the invention also provides a drilling loss prediction model establishment device, the structure of which is shown in Figure 6, including:

[0106] The sample acquisition module 61 is used to extract the seismic attribute value corresponding to each depth value of the missing marker from each set seismic attribute body according to the synthetic record of the well for a drilled well with missing marker data, so as to obtain a sample containing the missing marker corresponding to the depth value and each seismic attribute value. The missing marker data includes multiple depth values ​​of the well and the missing marker corresponding to each depth value.

[0107] The sample set acquisition module 62 is used to assemble a sample set from multiple samples acquired by the sample acquisition module 61.

[0108] The model training module 63 is used to train a set classification model using the sample set obtained by the sample set acquisition module 62 to obtain a classification model for predicting well leakage, the classification model including at least one sub-classification model.

[0109] In some embodiments, the model training module 63, in training the set classification model using the sample set, specifically performs the following iterative training steps on each sub-classification model using the sample set:

[0110] Based on the weight value and predicted omission probability of each earthquake attribute, a comprehensive omission probability is determined. The weight value of each earthquake attribute is then adjusted based on the comprehensive omission probability and the omission probability marked in the omission markers of the samples. This process continues until the degree of agreement between the currently determined comprehensive omission probability and the marked omission probability reaches a first set requirement.

[0111] In some embodiments, when there is more than one sub-classification model, the model training module 63, specifically for training the set classification model using the sample set, is used to:

[0112] After training each sub-classification model in the set classification model using the sample set, the weight value of each sub-classification model is iteratively optimized until the predicted omission probability determined based on the current weight value of each sub-classification model and the output comprehensive omission probability matches the omission probability marked in the omission label of the sample to the second set requirement.

[0113] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0114] Based on the inventive concept of this invention, embodiments of this invention also provide a computer program product with drilling loss prediction function, including a computer program / instruction, wherein when the computer program / instruction is executed by a processor, it implements the above-mentioned drilling loss prediction model establishment method, or implements the above-mentioned drilling loss prediction method.

[0115] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process may be rearranged without departing from the scope of this disclosure. The appended method claims provide elements of various steps in an exemplary order and are not intended to limit the scope to the specific order or hierarchy described.

[0116] In the detailed description above, various features are combined together in a single embodiment to simplify this disclosure. This approach to disclosure should not be construed as reflecting an intention that embodiments of the claimed subject matter require more features than are explicitly stated in each claim. Rather, as reflected in the appended claims, the invention is presented with fewer features than all of the features in a single disclosed embodiment. Therefore, the appended claims are hereby explicitly incorporated into the detailed description, with each claim representing a separate preferred embodiment of the invention.

[0117] The foregoing description includes examples of one or more embodiments. It is certainly impossible to describe all possible combinations of components or methods in order to describe the above embodiments, but those skilled in the art will recognize that further combinations and arrangements of the various embodiments are possible. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. Furthermore, the term "comprising" as used in the specification or claims is interpreted in a manner similar to the term "including," as interpreted when used as a conjunction in the claims. Additionally, the use of any term "or" in the specification of the claims is intended to mean "non-exclusive or."

Claims

1. A method for establishing a drilling loss prediction model, characterized in that, include: For drilled wells with varying degrees of leakage in their leakage-marked data, seismic attribute values ​​corresponding to each depth value of the leakage marker are extracted from each set seismic attribute volume based on the well's synthetic record. This yields a sample containing the leakage marker corresponding to each depth value and each seismic attribute value. The leakage-marked data includes multiple depth values ​​of the well and the leakage marker corresponding to each depth value. The set seismic attribute volume is a variance, time-frequency decay, sweet spot, or root mean square amplitude seismic attribute volume. The leakage marker is a leakage probability determined based on the leakage amount. Multiple samples are combined into a sample set. After training each sub-classification model in the set classification model using the sample set, the weight values ​​of each sub-classification model are iteratively optimized until the predicted leakage probability determined based on the current weight value and the output comprehensive leakage probability of each sub-classification model reaches a second set requirement in agreement with the leakage probability marked in the sample's leakage marker. This yields a classification model for predicting drilling leakage in carbonate formations. The classification model includes a logistic regression sub-classification model, a decision tree sub-classification model, and a support vector machine sub-classification model.

2. The method as described in claim 1, characterized in that, The step of training the classification model using the sample set specifically includes performing the following iterative training steps on each sub-classification model using the sample set: determining the comprehensive omission probability based on the weight value and predicted omission probability corresponding to each earthquake attribute; adjusting the weight value corresponding to each earthquake attribute based on the comprehensive omission probability and the omission probability marked in the omission labels of the samples; until the degree of agreement between the currently determined comprehensive omission probability and the marked omission probability reaches a first set requirement.

3. The method as described in claim 1, characterized in that, Also includes: A synthetic record is obtained by deconvolution using the logging curves of the drilled well and the seismic wavelet; the synthetic record is then adjusted based on the seismic volume corresponding to the drilled well extracted from the seismic data volume.

4. The method as described in claim 3, characterized in that, The logging curves are sonic transit time curves and density curves.

5. The method as described in claim 3, characterized in that, The seismic wavelet is either a Ricker wavelet or a zero-phase wavelet extracted from the seismic data volume.

6. The method as described in claim 1, characterized in that, Also includes: For other drilled wells with missing data, the seismic attribute values ​​corresponding to each depth value of the missing data are extracted from each set seismic attribute volume based on the synthetic record of the well, resulting in a test sample containing the missing data corresponding to the depth value and each seismic attribute value; multiple test samples are combined into a test sample set; drilling missing prediction is performed using the trained classification model and the test sample set to obtain the predicted missing probability corresponding to each test sample; it is determined whether the consistency between the predicted missing probability and the marked missing probability of each test sample meets the third set requirement; if not, the test sample set is added to the sample set, and the current sample set is used to train the current classification model.

7. A method for predicting well leakage, characterized in that, include: Each set seismic attribute extracted from the seismic data volume is input into a classification model obtained using the method described in any one of claims 1 to 6 to obtain well leakage prediction results.

8. A drilling loss prediction model establishment device, characterized in that, include: The sample acquisition module is used to extract seismic attribute values ​​corresponding to each depth value of the missing data from each set seismic attribute volume for drilled wells with different degrees of missing data, based on the synthetic record of the well, to obtain a sample containing the missing data corresponding to each depth value and each seismic attribute value. The missing data includes multiple depth values ​​of the well and the missing data corresponding to each depth value. The set seismic attribute volume is a variance, time-frequency decay, sweet spot, or root mean square amplitude seismic attribute volume, and the missing data is a missing probability determined based on the amount of missing data. The sample set acquisition module is used to process the data acquired by the sample acquisition module. Multiple samples constitute a sample set; the model training module is used to train each sub-classification model in the set classification model using the sample set obtained by the sample set acquisition module, and then iteratively optimizes the weight value of each sub-classification model until the predicted leakage probability determined by the current weight value and the output comprehensive leakage probability of each sub-classification model meets the second set requirement of the degree of agreement with the leakage probability marked in the leakage label of the sample, so as to obtain a classification model for predicting drilling leakage in carbonate rock formations, wherein the classification model includes a logistic regression sub-classification model, a decision tree sub-classification model and a support vector machine sub-classification model.

9. A computer program product with drilling loss prediction function, comprising a computer program / instructions, wherein, When the computer program / instruction is executed by the processor, it implements the drilling loss prediction model establishment method according to any one of claims 1 to 6, or the drilling loss prediction method according to claim 7.

Images