An oil and gas resource reliability quantitative characterization method, system and storage medium

Abstract

The present application relates to the technical field of oil and gas resource exploration, and particularly relates to a method and system for quantitatively characterizing the reliability of oil and gas resources and a storage medium, comprising: obtaining an oil and gas enrichment probability representing the reliability of oil and gas resources, and quantifying geological condition parameters and exploration evidence parameters of the oil and gas enrichment probability; training an error parameter identification model for identifying error parameters; and jointly training an oil and gas enrichment probability prediction model adapted to the error parameter identification model for predicting the oil and gas enrichment probability in the presence of error parameters. Through the innovative hierarchical error processing architecture, multi-level error identification system and cascade training strategy, the present application effectively solves the parameter error problem in the calculation of the oil and gas enrichment probability, fully considers the error amplification effect caused by the multiplication characteristics of the geological condition parameters, and the differentiated processing requirements of different types of errors, forming a complete solution from error identification to compensation prediction.

Description

Technical Field

[0001] This invention relates to the field of oil and gas resource exploration technology, specifically to a method, system, and storage medium for quantitative characterization of oil and gas resource reliability. Background Technology

[0002] In the field of oil and gas exploration, the probability of oil and gas enrichment is a core indicator for quantitatively characterizing resource reliability and decision-making potential. Its traditional calculation method usually relies on the integration of multiple geological parameters. One of the most classic models is to perform a weighted summation of the product of six key geological condition parameters (source rock, reservoir, caprock, trap, charging, and preservation) with exploration evidence parameters (gas measurement data).

[0003] However, this model suffers from an inherent weakness that traditional methods have failed to address effectively: multiplication acts like a magnifying glass, drastically amplifying minute errors in single or multiple input parameters. For example, multiplying six parameters between 0 and 1 results in a product that is extremely sensitive to fluctuations in any one parameter. Furthermore, the exploration evidence parameters themselves contain uncertainties. Most existing technologies either ignore these errors or employ simple data cleaning or averaging, lacking an intelligent method that can systematically identify, diagnose, and compensate for errors of varying sources and degrees. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system and storage medium for quantitative characterization of oil and gas resource reliability, so as to solve the technical problem of lack of automated identification and compensation for errors of different sources and degrees in the process of quantifying the probability of oil and gas enrichment in the prior art.

[0005] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution:

[0006] A method for quantitatively characterizing the reliability of oil and gas resources includes the following steps:

[0007] To obtain the probability of oil and gas enrichment that characterizes the reliability of oil and gas resources, as well as geological condition parameters and exploration evidence parameters that quantify the probability of oil and gas enrichment.

[0008] A first dataset is constructed based on geological condition parameters. Random errors are added to the geological condition parameters in the first dataset to form a first low-resolution dataset. A second dataset is constructed based on the first low-resolution dataset, exploration evidence parameters, and oil and gas enrichment probability. Random errors are added to the exploration evidence parameters in the second dataset to form a second low-resolution dataset.

[0009] An error parameter recognition model was trained in the first low-resolution dataset and the second low-resolution dataset to identify the error parameters.

[0010] A probability prediction model for oil and gas enrichment, adapted to the error parameter identification model, is jointly trained in the second dataset, the first low-resolution dataset, and the second low-resolution dataset to predict the probability of oil and gas enrichment in the presence of error parameters.

[0011] As a preferred embodiment of the present invention, the geological condition parameters include source rock conditions. Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions ;

[0012] Hydrocarbon source rock conditions Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions and exploration evidence parameters Values ​​are assigned by expert experience;

[0013] The probability of hydrocarbon enrichment depends on the source rock conditions. Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions product term Exploration Evidence Parameters We obtain the weighted sum.

[0014] As a preferred embodiment of the present invention, the source rock conditions Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions and the product term Combine them into the first error-free sample, and summarize all the first error-free samples to form the first dataset;

[0015] Random noise is added to any one or two of the first error-free samples to obtain low-error samples.

[0016] Random noise is added to any three or four items in the first error-free sample to obtain a high-error sample.

[0017] All low-error samples and all high-error samples are added to the first dataset to form the first low-resolution dataset.

[0018] As a preferred embodiment of the present invention, the method for training an error parameter identification model in a first low-resolution dataset includes:

[0019] Using data samples from the first low-resolution dataset as input and the first error label of the data samples as output, a classification model is trained to obtain the first error parameter recognition model. ,in, The first error label is the output of the identification model for the first error parameter. RF is a classification model. The first error label includes: no error, low error, and high error.

[0020] The training loss of the first error parameter identification model ,in, This represents the true value of the first error label. This is for cross-entropy calculation.

[0021] As a preferred embodiment of the present invention, the source rock conditions of the first error-free sample in the first low-resolution dataset are used. Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions As input, the product term of the first error-free sample As output, a neural network CNN is trained to obtain an error-free prediction model for the product term. ,in, The product term is the output of the error-free prediction model for the product term;

[0022] Hydrocarbon source rock conditions in low-error samples of the first low-resolution dataset Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions As input, the product term is the output of the error-free prediction model. As output, the neural network CNN is retrained to obtain a low-error prediction model for the product term. ,in, The product term is the output of the low-error prediction model for the product term;

[0023] Hydrocarbon source rock conditions of high-error samples in the first low-resolution dataset Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions As input, the product term is used to predict the output of the low-error prediction model. As output, the neural network CNN is retrained to obtain a high-error prediction model for the product term. ,in, The product term is the output of the high-error prediction model for the product term;

[0024] The training loss of the error-free prediction model for the product term ,in, This is for calculating the mean square error;

[0025] The training loss of the low-error prediction model for the product term ,in, For KL divergence calculation;

[0026] The training loss of the high-error prediction model for the product term .

[0027] As a preferred embodiment of the present invention, the product term of the first error-free sample in the first low-resolution dataset is... Exploration evidence parameters and the probability of oil and gas enrichment These are combined to form the second error-free sample, constituting the second dataset;

[0028] The product term output by the low-error prediction model of the product term The product term in the high-error prediction model output of the product term Exploration evidence parameters and the probability of oil and gas enrichment The combination forms the error sample of the product term;

[0029] Exploration evidence parameters in the second error-free sample Random noise was added to obtain exploration evidence error samples;

[0030] Add product term error samples and exploration evidence error samples to the second dataset to form a second low-resolution dataset.

[0031] As a preferred embodiment of the present invention, the method for training an error parameter identification model in a second low-resolution dataset includes:

[0032] Using data samples from the second low-resolution dataset as input and the second error label of the data samples as output, a classification model is trained to obtain a second error parameter recognition model. ,in, The second error label is the output of the second error parameter identification model. RF is a classification model. The second error label includes the second error-free error, the product term error, and the exploration evidence error.

[0033] The training loss of the second error parameter identification model ,in, This represents the true value of the second error label. This is for cross-entropy calculation.

[0034] As a preferred embodiment of the present invention, the product term of the second error-free samples in the second low-resolution dataset is used. Exploration evidence parameters As input, the oil and gas enrichment probability of the second error-free sample. As output, a neural network CNN is trained to obtain an enriched probability error-free prediction model. ,in, The enrichment probability is the output of the error-free prediction model for enrichment probability.

[0035] The product terms of the error samples in the second low-resolution dataset or Exploration evidence parameters As input, the enrichment probability is the output of the error-free prediction model. As output, the neural network CNN is retrained to obtain the error prediction model of the enriched probability product term. ,in, The enrichment probability is the output of the enrichment probability product term error prediction model.

[0036] The product term of the exploration evidence error samples in the second low-resolution dataset Exploration evidence parameters after random noise addition As input, the enrichment probability is the output of the error-free prediction model. As output, the neural network CNN is retrained to obtain the enriched probability exploration evidence error prediction model. ,in, The enrichment probability is the output of the enrichment probability exploration evidence error prediction model.

[0037] The training loss of the enrichment probability error-free prediction model ,in, This is for calculating the mean square error;

[0038] The training loss of the enriched probability product term error prediction model ,in, For KL divergence calculation;

[0039] The training loss of the enriched probability exploration evidence error prediction model .

[0040] As a preferred embodiment of the present invention, the present invention provides a quantitative characterization system for oil and gas resource reliability, applied to a method for quantitative characterization of oil and gas resource reliability, the system comprising:

[0041] The data collection unit is used to acquire the probability of oil and gas enrichment, which characterizes the reliability of oil and gas resources, as well as geological condition parameters and exploration evidence parameters that quantify the probability of oil and gas enrichment.

[0042] The data preprocessing unit is used to construct a first dataset based on geological condition parameters, add random errors to the geological condition parameters in the first dataset to form a first low-resolution dataset, and construct a second dataset based on the first low-resolution dataset, exploration evidence parameters and oil and gas enrichment probability, add random errors to the exploration evidence parameters in the second dataset to form a second low-resolution dataset.

[0043] The machine learning unit is used to train an error parameter identification model in the first low-resolution dataset and the second low-resolution dataset to identify error parameters; and to jointly train an oil and gas enrichment probability prediction model adapted to the error parameter identification model in the second dataset, the first low-resolution dataset, and the second low-resolution dataset to predict the oil and gas enrichment probability in the presence of error parameters.

[0044] As a preferred embodiment of the present invention, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement a method for quantitative characterization of oil and gas resource reliability.

[0045] Compared with the prior art, the present invention has the following advantages:

[0046] This invention effectively solves the parameter error problem in calculating the probability of oil and gas enrichment through an innovative hierarchical error processing architecture, a multi-level error identification system, and a cascaded training strategy. It fully considers the error amplification effect caused by the multiplication characteristics of geological condition parameters, as well as the differentiated processing requirements of different types of errors, forming a complete solution from error identification to compensation prediction. It achieves significant improvements in the accuracy of error processing, system robustness, and computational efficiency, providing more scientific and reliable technical support for the reliability assessment of oil and gas resources. Attached Figure Description

[0047] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0048] Figure 1 A flowchart of the quantitative characterization method for oil and gas resource reliability provided in an embodiment of the present invention;

[0049] Figure 2 This is a structural diagram of the oil and gas enrichment probability prediction model provided in an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of nine favorable oil and gas zones provided in an embodiment of the present invention;

[0051] Figure 4 This is a comparison chart of the prediction performance of the oil and gas enrichment probability prediction model provided in the embodiments of the present invention;

[0052] Figure 5 A system block diagram of the quantitative characterization method for oil and gas resource reliability provided in an embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] like Figure 1 As shown, this invention provides a method for quantitatively characterizing the reliability of oil and gas resources, comprising the following steps:

[0055] To obtain the probability of oil and gas enrichment that characterizes the reliability of oil and gas resources, as well as geological condition parameters and exploration evidence parameters that quantify the probability of oil and gas enrichment.

[0056] A first dataset is constructed based on geological condition parameters. Random errors are added to the geological condition parameters in the first dataset to form a first low-resolution dataset. A second dataset is constructed based on the first low-resolution dataset, exploration evidence parameters, and oil and gas enrichment probability. Random errors are added to the exploration evidence parameters in the second dataset to form a second low-resolution dataset.

[0057] An error parameter recognition model was trained in the first low-resolution dataset and the second low-resolution dataset to identify the error parameters.

[0058] A probability prediction model for oil and gas enrichment, adapted to the error parameter identification model, is jointly trained in the second dataset, the first low-resolution dataset, and the second low-resolution dataset to predict the probability of oil and gas enrichment in the presence of error parameters.

[0059] To accurately predict the probability of hydrocarbon enrichment even when there are errors in geological condition parameters and exploration evidence parameters (i.e., parameters related to hydrocarbon gas measurement and testing), this invention first establishes a machine learning model for automatic error identification to determine whether the error originates from geological condition parameters or exploration evidence parameters. Furthermore, it establishes targeted error compensation prediction models based on the error type: when the error originates from geological condition parameters, a machine learning model for accurately predicting the enrichment probability under the condition of errors in geological condition parameters is established (i.e., an enrichment probability product term error prediction model); when the error originates from exploration evidence data, a machine learning model for accurately predicting the enrichment probability under the condition of errors in exploration evidence data is established (i.e., an enrichment probability exploration evidence error prediction model).

[0060] In the expert empirical model for enrichment probability calculation, six parameters of source rock conditions are included in the geological condition parameters. Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions The overall product term formed by multiplication , and exploration evidence parameters A weighted summation is performed. Therefore, this invention uses the product term formed by the six parameters in the geological condition parameters. Perform overall error identification and determine if the error data is a product term. Or exploration evidence parameters When the error data is a product term At that time, a targeted prediction model using the enriched probability product term is employed, when the error data is exploration evidence parameters. At that time, the enrichment probability exploration evidence error prediction model was adopted.

[0061] In this invention, there are six parameters in the geological condition parameters, and the error of each parameter will affect the product term. The above shows that one or two parameters have errors, which affect the product term. The above produces a low-level error in the product term, but when 3-4 parameters have errors, the multiplication affects the product term. This will result in a high degree of error in the product term.

[0062] This invention first establishes a machine learning model (first error parameter identification model) to identify whether a product term belongs to a low-level error or a high-level error. Then, it specifically establishes machine learning models for error compensation for low-level and high-level errors, namely a low-error prediction model and a high-error prediction model for the product term. Additionally, it establishes an error-free prediction model for the product term, which can automatically predict the product term under six error-free parameter data conditions. The training loss... This ensures that the error-free prediction model output of the product term is close to the true value, thus guaranteeing the model's prediction accuracy.

[0063] The low-error prediction model for product terms can predict product terms even with one or two error parameters, and the training loss is low. It consists of two parts, the first part It measures the difference between the prediction results of product terms based on 1 to 2 error parameters (the output of the low-error product term prediction model) and the prediction results of product terms based on 6 error-free parameters (the output of the no-error product term prediction model). It forces the low-error product term prediction model to mimic the output of the no-error product term prediction model on high-quality data as closely as possible, even when the input data contains 1-2 errors (low-level errors). Essentially, it distills the knowledge or judgment logic learned by the no-error product term model into the low-error product term prediction model, enabling it not only to learn to fit the data but also to think like the no-error product term prediction model. This allows it to ignore or correct noise in the input to a certain extent, significantly improving the low-error product term prediction model's tolerance to 1-2 error parameters.

[0064] Part Two The difference between the prediction results of the product term based on 1 to 2 error parameters (the output of the low-error prediction model of the product term) and the true value was measured. This ensures that the prediction of the low-error prediction model of the product term will not completely deviate from the final correct target by simply imitating the error-free model of the product term. It is used as an anchor point to directly pull the low-error prediction model of the product term toward the true value. This prevents the low-error prediction model from amplifying the deviation when there is a small deviation in the prediction of the error-free model. It ensures the basic correctness of the learning process.

[0065] Furthermore, while high-error prediction models for product terms can predict product terms even with 3 to 4 error parameters, their large number and significant error mean that directly fitting them to error-free prediction models for product terms yields low reliability. Therefore, this invention utilizes these models to fit a low-error prediction model for product terms, employing a training loss... It consists of two parts, the first part This method measures the difference between the prediction results of product terms based on 3 to 4 error parameters (the output of the high-error prediction model) and the prediction results of product terms based on 1 to 2 error parameters (the output of the low-error prediction model). It forces the high-error prediction model to mimic the output of the low-error prediction model with 1 to 2 errors as closely as possible, even when the input data has 3-4 errors (high-level errors). Essentially, it distills the knowledge or judgment logic learned by the low-error model and transfers it to the high-error prediction model, enabling it not only to learn to fit the data but also to think like the low-error model, thus ignoring or correcting noise in the input to some extent. This significantly improves the high-error prediction model's tolerance for 3-4 error parameters.

[0066] Part Two The difference between the prediction results of the product term based on 3 to 4 error parameters (the output of the high-error prediction model of the product term) and the prediction results of the product term based on 6 error-free parameters (the output of the error-free prediction model of the product term) was measured. This ensures that the prediction of the high-error prediction model of the product term will not completely deviate from the final correct target by simply imitating the low-error prediction model of the product term. It uses the low-error prediction model as an anchor point to directly pull the high-error prediction model of the product term towards error-free prediction. This prevents the high-error prediction model from amplifying the deviation when there is a small deviation in the low-error prediction model. It ensures the basic correctness of the learning process.

[0067] This invention employs a layer-by-layer cascaded training method, which ensures the matching between the compensation strategy and the severity of the error. It fine-tunes minor problems and activates a more powerful compensation mechanism for severe problems. At the same time, the cascaded structure allows complex models dealing with high errors to build upon low-error models rather than starting from scratch, thus accelerating convergence and improving performance.

[0068] Therefore, this invention can compensate for the error of the product term under both low and high error conditions, that is, it can obtain a relatively accurate prediction result of the product term under both low and high error conditions.

[0069] After obtaining the error-compensated prediction result of the product term, since it still corresponds to the error source of geological condition parameters, this invention trains an enriched probability product term error prediction model. This model can utilize the error-compensated prediction result of the product term, that is, obtain an accurate enriched probability prediction result based on the error-compensated prediction result of the product term. Its training loss... It consists of two parts, the first part The measurement of the difference between the enriched probabilities output by the enriched probability product term error prediction model and the enriched probability error-free prediction model forces the enriched probability product term error prediction model to mimic the output of the enriched probability error-free prediction model on high-quality data as closely as possible, even when there are errors in the input product terms (low or high levels of error). This is equivalent to distilling the knowledge or judgment logic learned by the enriched probability error-free model into the enriched probability product term error prediction model, enabling it not only to learn to fit the data but also to think like the enriched probability error-free prediction model. This allows it to ignore or correct noise in the input to a certain extent, significantly improving the enriched probability product term error prediction model's tolerance to product term error parameters.

[0070] Part Two The difference between the enriched probability output by the enriched probability product term error prediction model and the true value is measured to ensure that the prediction of the enriched probability product term error prediction model will not completely deviate from the final correct target by simply imitating the enriched probability error-free model. It is used as an anchor point to directly pull the enriched probability product term error prediction model toward the true value. That is, it prevents the enriched probability product term error prediction model from amplifying the deviation when there is a small deviation in the prediction of the enriched probability error-free model. It ensures the basic correctness of the learning process.

[0071] Therefore, based on the compensated product terms, enrichment probability prediction results that are similar to or have high accuracy with the true value can be obtained.

[0072] Similarly, this invention trains an enrichment probability exploration evidence error prediction model, which can obtain accurate enrichment probability prediction results from exploration evidence data containing errors. Its training loss... It consists of two parts, the first part Measuring the difference between the enriched probability output of the enriched probability exploration evidence error prediction model and the enriched probability error-free prediction model forces the error prediction model to mimic the output of the enriched probability error-free prediction model on high-quality data as closely as possible, even when the input exploration evidence contains errors. This is equivalent to distilling the knowledge or judgment logic learned by the enriched probability error-free model into the enriched probability exploration evidence error prediction model, enabling it not only to learn to fit the data but also to think like the enriched probability error-free prediction model. This allows it to ignore or correct noise in the input to a certain extent, significantly improving the enriched probability exploration evidence error prediction model's tolerance to exploration evidence error parameters.

[0073] Part Two The difference between the enriched probability output by the enriched probability exploration evidence error prediction model and the true value is measured to ensure that the prediction of the enriched probability exploration evidence error prediction model will not completely deviate from the final correct target by simply imitating the enriched probability error-free model. It is used as an anchor point to directly pull the enriched probability exploration evidence error prediction model toward the true value. That is, it prevents the enriched probability exploration evidence error prediction model from amplifying the deviation when there is a small deviation in the prediction of the enriched probability error-free model. It ensures the basic correctness of the learning process.

[0074] Therefore, enrichment probability prediction results that are similar to or have high accuracy can be obtained based on exploration evidence with errors in the prediction of enrichment probability.

[0075] After establishing a targeted enrichment probability prediction model, this invention adapts it to the labels output by the error data identification model, as follows:

[0076] When the label output by the first error data identification model is that there is a low degree of error in the geological condition parameters, the geological condition parameters with 1-2 errors are input into the low error prediction model of the product term to obtain the compensated product term under the error scenario.

[0077] When the label output by the first error data identification model is that there is a high degree of error in the geological condition parameters, the geological condition parameters with 3-4 errors are input into the high error prediction model of the product term to obtain the compensated product term under the error scenario.

[0078] Furthermore,

[0079] When the label output by the second error data identification model is "product term error", the product term is input into the enrichment probability product term error prediction model to obtain the oil and gas enrichment probability corresponding to the product term.

[0080] When the label output by the second error data identification model is "exploration evidence error exists", the exploration evidence is input into the enrichment probability exploration evidence error prediction model to obtain the oil and gas enrichment probability corresponding to the exploration evidence error term.

[0081] This invention dynamically selects the corresponding hierarchical prediction model for value prediction based on the output of the error data identification model, achieving adaptive error prediction. This mechanism ensures that the optimal model is used under specific error scenarios, maximizing prediction accuracy while reducing computational resource waste and improving method efficiency.

[0082] Error-free prediction models are oil and gas resource value prediction models that predict the value of oil and gas resources in the absence of error data.

[0083] Geological parameters include source rock conditions Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions ;

[0084] Hydrocarbon source rock conditions Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions and exploration evidence parameters The values ​​were assigned based on expert experience, and the standards for assigning expert experience values ​​are shown in Table 1.

[0085] The probability of hydrocarbon enrichment depends on the source rock conditions. Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions product term Exploration Evidence Parameters We obtain the weighted sum.

[0086] The probability of oil and gas enrichment primarily reflects the reliability of oil and gas resource enrichment. It is a parameter jointly determined by the probability of existence of various controlling factors, exploration level, and oil and gas discovery status within the survey area. The resource reliability coefficient is calculated by weighting factors according to different weights. The calculation method is shown in the formula:

[0087] P = (P hydrocarbon generation × P reservoir × P caprock × P trap × P charging × P preservation) × A1 + P gas meter readings and test results × A2;

[0088] In the formula:

[0089] P—Probability of oil and gas enrichment, dimensionless, ranging from 0 to 1;

[0090] P generates hydrocarbons ( — The values ​​of the source rock condition evaluation parameters are assigned, dimensionless, and range from 0 to 1;

[0091] P reservoir ( — The values ​​of reservoir condition evaluation parameters are assigned, dimensionless, and range from 0 to 1;

[0092] P capping layer ( — The values ​​of the evaluation parameters for caprock conditions are assigned, dimensionless, and range from 0 to 1;

[0093] P-loop ( — The values ​​of the trap condition evaluation parameters are assigned, dimensionless, and range from 0 to 1;

[0094] P-filling ( — The values ​​of the filling condition evaluation parameters are assigned, dimensionless, and range from 0 to 1;

[0095] P saves ( — The value of the evaluation parameter for the storage condition is dimensionless and ranges from 0 to 1.

[0096] P-gas measurement display and test status ( — The value assigned to the oil and gas measurement and testing status, dimensionless, ranging from 0 to 1.

[0097] A1 and A2 have different weights, and their sum is 1.

[0098] Table 1. Parameter Assignment Table for Oil and Gas Condition Evaluation in Favorable Oil and Gas Areas

[0099] Assignment range Hydrocarbon source rock conditions reservoir conditions cap layer conditions Trap conditions Filling conditions Storage conditions Gas measurement display and test status [0.75，1.0] Large effective source rock area and thickness, high organic carbon content The reservoir is thick and has a stable distribution. The lithology consists of gypsum-salt and argillaceous rocks, with a large thickness, no well-developed fractures, and good sealing properties. Data confirms the existence of the trap, and that the trap is large in scale. The source rocks are of good quality and are in the peak period of hydrocarbon generation. The traps are formed before the main hydrocarbon expulsion phase, and the spatial configuration of the traps and source rocks is favorable. Data confirms the existence and integrity of the caprock, which consists of gypsum-salt rock, mudstone, shale, carbonate rock, etc.; faults are not well developed, and no faults have been found to cut through traps or caprock. The gas meter readings are good, and the test yielded an industrial oil and gas flow. [0.5，0.75] The effective source rock has a large area and thickness, and a high organic carbon content. The reservoir is relatively thick and its distribution is relatively stable. The lithology is mudstone, with a large thickness and relatively underdeveloped faults, resulting in good sealing properties. Data suggests that a trap may exist, and the trap is relatively large. The source rocks are of good quality and are in the oil and gas generation stage. The traps are formed before the main hydrocarbon expulsion phase, and the spatial configuration of the traps and source rocks is favorable. Data suggests that the caprock may exist and is relatively intact; the degree of fault development is low, and no obvious faults are observed cutting through the traps or caprock. Gas analysis showed good results, and the test yielded a low-yield oil flow. [0，0.5] The effective source rock area and thickness are generally moderate, with a medium organic carbon content. The reservoir is relatively thin and has poor distribution stability. The lithology consists of mudstone and sandy mudstone, with moderate thickness, well-developed faults, and generally poor sealing properties. Data suggests that a trap may exist, but its size is relatively small. The source rocks are of poor quality, have not entered the oil and gas generation stage or are in an over-mature stage, the formation period of the traps cannot be determined or is later than the main hydrocarbon expulsion period, and the spatial configuration relationship between the traps and the source rocks is poor. The data cannot confirm the existence of a caprock; faults are well-developed, and there are instances of faults cutting through traps or caprocks. The gas meter readings are good, and the test yielded an oil and gas flow rate.

[0100] Hydrogen source rock conditions Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions Sum of product terms Combine them into the first error-free sample, and summarize all the first error-free samples to form the first dataset;

[0101] Random noise is added to any one or two items in the first error-free sample to obtain a low-error sample.

[0102] Random noise is added to any three or four items in the first error-free sample to obtain a high-error sample.

[0103] All low-error samples and all high-error samples are added to the first dataset to form the first low-resolution dataset.

[0104] Methods for training an error parameter identification model on a first low-resolution dataset include:

[0105] Using data samples from the first low-resolution dataset as input and the first error label of the data samples as output, a classification model is trained to obtain the first error parameter recognition model. ,in, The first error label is the output of the identification model for the first error parameter. RF is a classification model. The first error label includes: no error, low error, and high error.

[0106] The training loss of the first error parameter identification model ,in, This represents the true value of the first error label. This is for cross-entropy calculation.

[0107] Hydrocarbon source rock conditions of the first error-free sample in the first low-resolution dataset Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions As input, the product term of the first error-free sample As output, a neural network CNN is trained to obtain an error-free prediction model for the product term. ,in, The product term is the output of the error-free prediction model for the product term;

[0108] Hydrocarbon source rock conditions in low-error samples of the first low-resolution dataset Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions As input, the product term is the output of the error-free prediction model. As output, the neural network CNN is retrained to obtain a low-error prediction model for the product term. ,in, The product term is the output of the low-error prediction model for the product term;

[0109] Hydrocarbon source rock conditions of high-error samples in the first low-resolution dataset Reservoir conditions Cover conditions Trap conditions Filling conditions and storage conditions As input, the product term is the product term predicted by the low-error prediction model. As output, the neural network CNN is retrained to obtain a high-error prediction model for the product term. ,in, The product term is the output of the high-error prediction model for the product term;

[0110] Training loss of a product-term error-free prediction model ,in, This is for calculating the mean square error;

[0111] Training loss of low-error prediction models for product terms ,in, For KL divergence calculation;

[0112] Training loss of high-error prediction models with product terms .

[0113] The product term of the first error-free sample in the first low-resolution dataset Exploration evidence parameters and the probability of oil and gas enrichment These are combined to form the second error-free sample, constituting the second dataset;

[0114] The product term output by the low-error prediction model of the product term The product term in the high-error prediction model output of the product term Exploration evidence parameters and the probability of oil and gas enrichment The combination forms the error sample of the product term;

[0115] Exploration evidence parameters in the second error-free sample Random noise was added to obtain exploration evidence error samples;

[0116] Add product term error samples and exploration evidence error samples to the second dataset to form a second low-resolution dataset.

[0117] Methods for training error parameter identification models on a second low-resolution dataset include:

[0118] Using data samples from the second low-resolution dataset as input and the second error label of the data samples as output, a classification model is trained to obtain a second error parameter recognition model. ,in, These represent the actual values ​​of the product term and the exploration evidence parameter in the data samples of the second low-resolution dataset, respectively. The second error label is the output of the second error parameter identification model. RF is a classification model. The second error label includes the second error-free error, the product term error, and the exploration evidence error.

[0119] The training loss of the second error parameter identification model ,in, This represents the true value of the second error label. This is for cross-entropy calculation.

[0120] The product term of the second error-free sample in the second low-resolution dataset Exploration evidence parameters As input, the oil and gas enrichment probability of the second error-free sample. As output, a neural network CNN is trained to obtain an enriched probability error-free prediction model. ,in, The enrichment probability is the output of the error-free prediction model for enrichment probability.

[0121] The product terms of the error samples in the second low-resolution dataset or Exploration evidence parameters As input, the enrichment probability is the output of the error-free prediction model. As output, the neural network CNN is retrained to obtain the error prediction model of the enriched probability product term. ,in, The enrichment probability is the output of the enrichment probability product term error prediction model.

[0122] The product term of the exploration evidence error samples in the second low-resolution dataset Exploration evidence parameters after random noise addition As input, the enrichment probability is the output of the error-free prediction model. As output, the neural network CNN is retrained to obtain the enriched probability exploration evidence error prediction model. ,in, The enrichment probability is the output of the enrichment probability exploration evidence error prediction model.

[0123] Training loss of enriched probability error-free prediction model ,in, This is for calculating the mean square error;

[0124] Training loss of the enriched probability product term error prediction model ,in, For KL divergence calculation;

[0125] Training loss of the enrichment probability exploration evidence error prediction model .

[0126] Example:

[0127] Nine favorable oil and gas areas are identified within five favorable zones in the Tarim Basin: Keping Fault Uplift, eastern Tarim, western Maigati Slope, northeastern Tarim, and northern Awati Depression. These areas include Shajingzi, Alar, eastern Yuli, northern Ruoqiang 1, northern Ruoqiang 2, Yuepuhu, Shache, southern Korla, and Yuli. Figure 3 The reliability assessment of oil and gas resources (as shown in the figure), the expert experience assessment values ​​and model prediction values ​​are shown in Table 2 and 2. Figure 4 As shown in Table 2, Phydrogen, Preservoir, Pcaprock, Ptrap, Pcharge, and Ppreservation are values ​​assigned by experts, and the values ​​in parentheses are the values ​​after error processing. Figure 4In the diagram, the solid line represents the true value of the probability of oil and gas enrichment, and the dashed line represents the predicted value of the probability of oil and gas enrichment. Using the same oil and gas resource value, the classification results of the favorable oil and gas areas obtained by the two methods into three major categories, I, II, and III, are also similar. This shows that the oil and gas resource value prediction model established in this invention can accurately predict the value of oil and gas resources, and thus has effectiveness and accuracy when applied to the classification of favorable oil and gas areas.

[0128] Table 2 Quantitative Evaluation Table of Favorable Oil and Gas Areas in the Tarim Basin

[0129] Advantageous area Shajingzi Alar Weili East Ruoqiang North 1 Ruoqiang North 2 Yuepu Lake Shache Korla South Yuli Hydrocarbon source rocks 1 1 0.9 0.9(0.8) 0.9(1) 1(0.9) 1 1 1 reservoir 0.9（0.8） 0.9 0.9 0.9(0.65) 0.9 0.9(0.8) 0.9 0.8(0.9) 0.8 cap layer 0.85 0.9 0.85(0.5) 0.8 0.8(0.7) 0.85 0.85(0.8) 0.9 0.9(0.85) Enclosed 0.95 0.85（0.7） 0.85 0.8 0.8 0.8(0.75) 0.8 0.85 0.85 Filling 0.9（0.4） 0.85 0.9 0.9 0.9(0.85) 0.85 0.85(0.9) 0.85 0.85(0.7) save 0.9 0.95 0.7(0.6) 0.75 0.75 0.8 0.8 0.75(0.95) 0.75 A1 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 Gas meter display 1 1 1 1(0.9) 1(0.8) 0.9(1) 0.9(1) 0.9 0.9 A2 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3 Enrichment probability truth value 0.712 0.689 0.558 0.545 0.545 0.561 0.561 0.543 0.543 Enrichment probability prediction value 0.731 0.681 0.567 0.602 0.521 0.548 0.523 0.561 0.558

[0130] like Figure 5 As shown, this invention provides a quantitative characterization system for oil and gas resource reliability, applied to a method for quantitative characterization of oil and gas resource reliability. The system includes:

[0131] The data collection unit is used to acquire the probability of oil and gas enrichment, which characterizes the reliability of oil and gas resources, as well as geological condition parameters and exploration evidence parameters that quantify the probability of oil and gas enrichment.

[0132] The data preprocessing unit is used to construct a first dataset based on geological condition parameters, add random errors to the geological condition parameters in the first dataset to form a first low-resolution dataset, and construct a second dataset based on the first low-resolution dataset, exploration evidence parameters and oil and gas enrichment probability, add random errors to the exploration evidence parameters in the second dataset to form a second low-resolution dataset.

[0133] The machine learning unit is used to train an error parameter identification model in the first low-resolution dataset and the second low-resolution dataset to identify error parameters; and to jointly train an oil and gas enrichment probability prediction model adapted to the error parameter identification model in the second dataset, the first low-resolution dataset, and the second low-resolution dataset to predict the oil and gas enrichment probability in the presence of error parameters.

[0134] This invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement a method for quantitative characterization of the reliability of oil and gas resources.

[0135] This invention effectively solves the parameter error problem in calculating the probability of oil and gas enrichment through an innovative hierarchical error processing architecture, a multi-level error identification system, and a cascaded training strategy. It fully considers the error amplification effect caused by the multiplication characteristics of geological condition parameters, as well as the differentiated processing requirements of different types of errors, forming a complete solution from error identification to compensation prediction. It achieves significant improvements in the accuracy of error processing, system robustness, and computational efficiency, providing more scientific and reliable technical support for the reliability assessment of oil and gas resources.

[0136] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A method for quantitatively characterizing the reliability of oil and gas resources, characterized in that, Includes the following steps: To obtain the probability of oil and gas enrichment that characterizes the reliability of oil and gas resources, as well as geological condition parameters and exploration evidence parameters that quantify the probability of oil and gas enrichment. A first dataset is constructed based on geological condition parameters. Random errors are added to the geological condition parameters in the first dataset to form a first low-resolution dataset. A second dataset is constructed based on the first low-resolution dataset, exploration evidence parameters, and oil and gas enrichment probabilities. Random errors are added to the exploration evidence parameters in the second dataset to form a second low-resolution dataset. A first error parameter recognition model is trained using a first low-resolution dataset, and a second error parameter recognition model is trained using a second low-resolution dataset. Based on the second dataset and the second low-resolution dataset, we trained an error-free prediction model for enriched probability, an error prediction model for enriched probability product terms, and an error prediction model for enriched probability exploration evidence, respectively. These prediction models and the error identification model were dynamically adapted according to the error labels. Using the source rock conditions, reservoir conditions, caprock conditions, trap conditions, filling conditions, and preservation conditions of the first error-free sample in the first low-resolution dataset as input, and the product term of the first error-free sample as output, a neural network CNN is trained to obtain an error-free prediction model for the product term. Using the source rock conditions, reservoir conditions, caprock conditions, trap conditions, filling conditions, and preservation conditions of low-error samples in the first low-resolution dataset as input, and the product term output by the error-free prediction model of the product term as output, the neural network CNN is retrained to obtain the low-error prediction model of the product term. Using the source rock conditions, reservoir conditions, caprock conditions, trap conditions, filling conditions, and preservation conditions of high-error samples in the first low-resolution dataset as input, and the product term output by the low-error prediction model of the product term as output, the neural network CNN is retrained to obtain the high-error prediction model of the product term. Training loss of a product-term error-free prediction model ,in, The product term is the output of the error-free prediction model for the product term. The product term of the first error-free sample. This is for calculating the mean square error; Training loss of low-error prediction models for product terms ,in, The product term is the output of the low-error prediction model for the product term. For mean square error calculation. For KL divergence calculation; Training loss of high-error prediction models with product terms ,in, This is the product term output by the high-error prediction model for the product term.

2. The method for quantitative characterization of oil and gas resource reliability according to claim 1, characterized in that: The geological condition parameters include source rock conditions, reservoir conditions, caprock conditions, trap conditions, charging conditions, and preservation conditions; The parameters of source rock conditions, reservoir conditions, caprock conditions, trap conditions, charging conditions, preservation conditions, and exploration evidence are assigned by expert experience. The probability of hydrocarbon enrichment is obtained by weighted summation of the product of source rock conditions, reservoir conditions, caprock conditions, trap conditions, charging conditions, and preservation conditions with exploration evidence parameters.

3. The method for quantitative characterization of oil and gas resource reliability according to claim 2, characterized in that: The source rock conditions, reservoir conditions, caprock conditions, trap conditions, charging conditions, preservation conditions, and the product terms are combined to form the first error-free sample, and all the first error-free samples are summarized to form the first dataset; Random noise is added to any one or two of the first error-free samples to obtain low-error samples. Random noise is added to any three or four items in the first error-free sample to obtain a high-error sample. All low-error samples and all high-error samples are added to the first dataset to form the first low-resolution dataset.

4. The method for quantitative characterization of oil and gas resource reliability according to claim 3, characterized in that: Methods for training an error parameter identification model on a first low-resolution dataset include: Using data samples from the first low-resolution dataset as input and the first error label of the data samples as output, a classification model is trained to obtain a first error parameter recognition model. The first error label includes: no error, low error, and high error. The training loss of the first error parameter identification model ,in, The first error label is identified by the model outputting the first error parameter. This represents the true value of the first error label. This is for cross-entropy calculation.

5. The method for quantitative characterization of oil and gas resource reliability according to claim 4, characterized in that: The product terms of the first error-free samples in the first low-resolution dataset, the exploration evidence parameters, and the oil and gas enrichment probabilities are combined to form the second error-free samples, thus constituting the second dataset. The product term output by the low-error prediction model, the product term output by the high-error prediction model, the exploration evidence parameters, and the oil and gas enrichment probability are combined into a product term error sample. Noise is randomly added to the exploration evidence parameters of the second error-free sample to obtain the exploration evidence error sample; Add product term error samples and exploration evidence error samples to the second dataset to form a second low-resolution dataset.

6. The method for quantitative characterization of oil and gas resource reliability according to claim 5, characterized in that: Methods for training error parameter identification models on a second low-resolution dataset include: Using data samples from the second low-resolution dataset as input and the second error label of the data samples as output, a classification model is trained to obtain a second error parameter identification model. The second error label includes the second error-free error, the product term error, and the exploration evidence error. The training loss of the second error parameter identification model ,in, The second error label is identified by the model output for the second error parameter. This represents the true value of the second error label. This is for cross-entropy calculation.

7. The method for quantitative characterization of oil and gas resource reliability according to claim 6, characterized in that: Using the product term of the second error-free sample in the second low-resolution dataset and the exploration evidence parameters as input, and the oil and gas enrichment probability of the second error-free sample as output, a neural network CNN is trained to obtain an enrichment probability error-free prediction model. Using the product terms of the product term error samples in the second low-resolution dataset and the exploration evidence parameters as inputs, and the enriched probability output by the enriched probability error-free prediction model as output, the neural network CNN is retrained to obtain the enriched probability product term error prediction model. Using the product term of the exploration evidence error samples in the second low-resolution dataset and the exploration evidence parameters after random noise addition as input, and the enriched probability output by the enriched probability error-free prediction model as output, the neural network CNN is retrained to obtain the enriched probability exploration evidence error prediction model. The training loss of the enrichment probability error-free prediction model ,in, Let P be the enrichment probability output by the error-free prediction model, and let P be the oil and gas enrichment probability of the second error-free sample. This is for calculating the mean square error; The training loss of the enriched probability product term error prediction model ,in, The enrichment probability is the output of the enrichment probability product term error prediction model. For KL divergence calculation; The training loss of the enriched probability exploration evidence error prediction model ,in, This represents the enrichment probability output by the enrichment probability exploration evidence error prediction model.

8. A quantitative characterization system for the reliability of oil and gas resources, characterized in that, The system, which is applied to the quantitative characterization method for oil and gas resource reliability according to any one of claims 1-7, comprises: The data collection unit is used to acquire the probability of oil and gas enrichment, which characterizes the reliability of oil and gas resources, as well as geological condition parameters and exploration evidence parameters that quantify the probability of oil and gas enrichment. The data preprocessing unit is used to construct a first dataset based on geological condition parameters, add random errors to the geological condition parameters in the first dataset to form a first low-resolution dataset, and construct a second dataset based on the first low-resolution dataset, exploration evidence parameters and oil and gas enrichment probability, and add random errors to the exploration evidence parameters in the second dataset to form a second low-resolution dataset. The machine learning unit is used to train a first error parameter identification model using a first low-resolution dataset, train a second error parameter identification model using a second low-resolution dataset, and train an enriched probability error-free prediction model, an enriched probability product term error prediction model, and an enriched probability exploration evidence error prediction model based on the second dataset and the first low-resolution dataset, respectively. These prediction models and error identification models are dynamically adapted according to the error labels.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the method as described in any one of claims 1-7.

Images