Deep coal seam collapse pressure prediction method based on logging parameter uncertainty analysis
By performing basic and new uncertainty analysis on logging parameters of deep coal seam drilling, integrating uncertainty characteristics, and combining with rock mechanics models, the collapse pressure of deep coal seams can be accurately predicted. This solves the problem that the collapse pressure in deep coal seam drilling is not unique and improves the reliability of the prediction results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA PETROLEUM & CHEMICAL CORP
- Filing Date
- 2026-03-03
- Publication Date
- 2026-05-08
AI Technical Summary
In deep coal seam drilling, the uncertainty of existing logging parameters leads to non-unique prediction results of collapse pressure, which cannot meet the requirements of accurate assessment of wellbore stability in low-carbon coal mining.
By acquiring density, sonic transit time, and natural gamma logging data of the well section to be logged, we conduct base class and new class uncertainty analysis, integrate the uncertainty characteristics of logging parameters, and combine rock mechanics parameter models and improved methods to calculate the mean, standard deviation, and probability distribution range of collapse pressure.
It has achieved accuracy and reliability in predicting the collapse pressure of deep coal seams, avoided the interference of parameter fluctuations on the prediction results, and provided scientific support for drilling engineering.
Smart Images
Figure CN121993162A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of coalbed methane drilling engineering technology, and in particular to a method for predicting the collapse pressure of deep coal seams based on the uncertainty analysis of logging parameters. Background Technology
[0002] In low-carbon coal mining, wellbore stability in deep coal seams is crucial for ensuring safe and efficient mining operations, with collapse pressure prediction being a core component. Current technologies primarily rely on logging data such as density and sonic transit time to predict collapse pressure, methods used in conventional coal seam drilling. However, in deep coal seam scenarios, logging parameters are subject to uncertainty due to variations in testing methods and instruments, human factors, and underground environmental interference. This uncertainty is transmitted to calculations of rock mechanics parameters and geostress, leading to non-unique collapse pressure predictions. Traditional methods treat collapse pressure as a fixed value, failing to accurately reflect the actual situation and thus hindering the precise assessment and control of wellbore stability in deep coal seam drilling required for low-carbon coal mining. Summary of the Invention
[0003] This application solves the technical problem that the prediction results of deep coal seam collapse pressure are not unique due to the uncertainty of well logging parameters.
[0004] To address the aforementioned technical problems, this application proposes a method for predicting the collapse pressure of deep coal seams based on well logging parameter uncertainty analysis. The method includes: acquiring the well section to be logged; collecting data from the well section according to preset indicators based on preset measurement intervals to obtain a well logging parameter sequence; performing basic uncertainty analysis and new uncertainty analysis on the well logging parameter sequence to obtain basic uncertainty analysis results and new uncertainty analysis results; integrating the basic uncertainty analysis results and new uncertainty analysis results to obtain well logging parameter uncertainty characteristics; and gradually predicting the collapse pressure of deep coal seams based on the well logging parameter uncertainty characteristics to obtain the collapse pressure prediction result.
[0005] This application proposes one or more technical solutions, which have at least the following technical effects:
[0006] This application collects relevant parameter data at preset intervals for the well section to be logged. Through uncertainty analysis of the basic and new classes and result integration, the uncertainty characteristics of the parameters are obtained. The uncertainty characteristics of rock mechanics parameters, formation pressure, and horizontal stress are gradually derived. Combined with relevant criteria and improved methods, the mean, standard deviation, and probability distribution range of the collapse pressure are calculated. This allows for the accurate acquisition of the probability distribution of the collapse pressure in deep coal seams, making the prediction results of deep coal seam collapse pressure more accurate and reliable. This provides scientific support for drilling engineering and achieves the technical effect of avoiding the interference of parameter fluctuations on the prediction results and improving the reliability of the prediction results of deep coal seam collapse pressure. Attached Figure Description
[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0008] Figure 1 This is a flowchart illustrating the method for predicting the collapse pressure of deep coal seams based on the uncertainty analysis of well logging parameters provided in this application embodiment.
[0009] Figure 2 This is a flowchart illustrating the process of obtaining the uncertainty characteristics of logging parameters in the deep coal seam collapse pressure prediction method based on the uncertainty analysis of logging parameters provided in the embodiments of this application. Detailed Implementation
[0010] This application provides a method for predicting the collapse pressure of deep coal seams based on the uncertainty analysis of logging parameters, which solves the technical problem that the prediction results of deep coal seam collapse pressure are not unique due to the uncertainty of logging parameters.
[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0012] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.
[0013] like Figure 1 As shown, a method for predicting the collapse pressure of deep coal seams based on the uncertainty analysis of well logging parameters is described, wherein the method includes:
[0014] The well section to be tested is obtained, and data is collected from the well section to be tested according to preset indicators based on preset measurement intervals to obtain the well logging parameter sequence.
[0015] Specifically, the first step is to delineate the well sections to be measured. Based on the engineering requirements of deep coal seam drilling in low-carbon coal mining, well sections meeting the drilling operation assessment needs are selected as the measurement targets. Then, measurement intervals are set according to data acquisition specifications, ensuring the height of the well section to be measured is greater than 100m. Simultaneously, each 0.1m interval is designated as a data acquisition node, with at least one set of data collected at each node. This interval setting ensures high-density coverage of logging data, providing a sufficient sample base for subsequent analysis.
[0016] Next, data acquisition was carried out according to the preset indicators. For the designated well section to be tested, three core logging data types—density, sonic transit time, and natural gamma—were simultaneously collected at each set acquisition node, completing the initial data collection for the entire well section. After the acquisition was completed, the data cleaning process was initiated. Outlier identification and missing value handling methods were used to remove abnormal data caused by instrument errors, human operational deviations, and underground environmental interference, and to supplement or discard missing data to ensure the integrity and accuracy of the retained data.
[0017] Finally, the cleaned and effective data were organized in an orderly manner. According to the interval numbering order of the acquisition nodes, the density, sonic transit time and natural gamma data corresponding to each node were arranged in sequence to form a well logging parameter sequence arranged in numerical order, providing a structured data foundation for subsequent uncertainty analysis.
[0018] The base class uncertainty analysis and the new class uncertainty analysis are performed on the well logging parameter sequence to obtain the results of the base class uncertainty analysis and the new class uncertainty analysis.
[0019] Optionally, the logging parameter sequence is first decomposed into density sequence, sonic transit time sequence, and natural gamma logging data sequence. Base class and new class uncertainty analysis are first performed on the density sequence, and the analysis results are correspondingly included in the base class and new class uncertainty analysis results. Finally, the same two types of uncertainty analysis are performed on the sonic transit time sequence and the natural gamma logging data sequence, and the analysis results of each are supplemented into the corresponding total results to complete the collection of all analysis results.
[0020] The results of the base class uncertainty analysis and the new class uncertainty analysis are integrated to obtain the uncertainty characteristics of the logging parameters.
[0021] In one embodiment of this application, similarity identification is first performed on the two types of analysis results and an integrated similarity matrix is constructed. Based on the matrix, convolution interaction is performed on the two types of results to obtain the uncertainty analysis results of the interaction base class and the uncertainty analysis results of the interaction new class. Finally, the two interaction results are weighted according to preset weights to obtain the uncertainty features of logging parameters.
[0022] By combining the uncertainty characteristics of well logging parameters, the collapse pressure of deep coal seams is predicted step by step, and the collapse pressure prediction results are obtained.
[0023] Specifically, firstly, the uncertainty characteristics of rock mechanics parameters are determined by combining the rock mechanics parameter model. The mean and standard deviation of the uncertainty characteristics of the two types of parameters are then substituted into the formation pressure calculation model to obtain the uncertainty characteristics of formation pressure. Subsequently, the uncertainty characteristics of horizontal geostress are derived by combining the Huang model and the Rosenbluthe improved method, and its mean and standard deviation are obtained. Then, the mean, standard deviation, and probability distribution range of collapse pressure are obtained by combining the three types of uncertainty characteristics. Finally, these data are used as the collapse pressure prediction results.
[0024] Furthermore, the method provided in this application embodiment includes:
[0025] The preset parameters include density, sonic transit time, and natural gamma logging data.
[0026] Specifically, the core definitions of the three preset indicators are first clarified: density is a basic physical quantity that reflects the compactness of deep coal seam rocks; acoustic transit time is the time required for sound waves to travel a unit distance in coal seam rocks; and natural gamma logging data is a key parameter for recording the natural radioactivity intensity of deep coal seam strata. These three parameters together constitute the core basic data for predicting collapse pressure.
[0027] Next, for the designated well sections to be tested, data acquisition of three types of parameters was carried out simultaneously at each acquisition node using specialized logging equipment according to the preset measurement intervals. Density was obtained by detecting the degree of absorption of incident rays by the rock formation using density logging instruments; acoustic transit time was obtained by recording and calculating the time difference between sound wave emission and reception using acoustic logging equipment; and natural gamma logging data was directly recorded by capturing the intensity of gamma rays spontaneously released by the formation using a gamma logging instrument.
[0028] In actual data acquisition, an integrated logging device is deployed along the designated well section for a single operation. This type of device has continuous measurement capabilities and can automatically trigger data acquisition at preset 0.1m intervals as it moves along the well section. During movement, the device simultaneously activates the corresponding detection modules for density logging, sonic transit-time logging, and natural gamma logging, automatically recording the corresponding parameter data at each preset distance node. After acquisition is complete, the entire device is retrieved, and all recorded data is organized according to the acquisition node sequence to form a logging parameter sequence that meets the requirements.
[0029] Finally, the raw data of the three types of parameters were cleaned to remove outliers and missing data caused by instrument errors and environmental interference. The data were then sorted in the order of the acquisition interval number to form the corresponding density sequence, sonic transit time sequence and natural gamma logging data sequence. These were integrated to form a complete logging parameter sequence, providing accurate and structured basic data for subsequent uncertainty analysis of the base class and new class.
[0030] Furthermore, the method provided in this application embodiment includes:
[0031] The well logging parameter sequence is decomposed to obtain a density sequence, an acoustic transit time sequence, and a natural gamma ray logging data sequence. Basic class uncertainty analysis is performed on the density sequence to obtain the density basic class uncertainty analysis result. New class uncertainty analysis is performed on the density sequence to obtain the density new class uncertainty analysis result. The density basic class uncertainty analysis result and the density new class uncertainty analysis result are added to the basic class uncertainty analysis result and the new class uncertainty analysis result, respectively. Basic class uncertainty analysis and new class uncertainty analysis are performed on the acoustic transit time sequence and the natural gamma ray logging data sequence, and the analysis results are added to the basic class uncertainty analysis result and the new class uncertainty analysis result.
[0032] Optionally, based on the well logging parameter sequence formed by data cleaning and sorting according to the acquisition node number order, the parameter attributes corresponding to each node data in the sequence are first identified, and then extracted according to parameter type. The density data corresponding to all acquisition nodes are extracted one by one and integrated according to the node number order to form a density sequence. At the same time, the sonic transit time data and natural gamma logging data of each node are extracted separately, sorted and collected in order according to the acquisition node number order, and finally decomposed to obtain independent density sequence, sonic transit time sequence and natural gamma logging data sequence.
[0033] Then, the density sequence is subjected to noise reduction and screening for jump points and missing values to obtain a filtered density sequence. The filtered density sequence is then resampled at equal intervals to generate an updated density sequence. Finally, the probability distribution of the filtered density sequence is identified by the kernel density curve to obtain the density base class uncertainty analysis results. This step will be explained in detail in the following sections.
[0034] Next, a gradient trend segmentation operation is performed on the density sequence to obtain multiple density subsequences. The first density subsequence is selected from these density subsequences, and a new type of uncertainty analysis is carried out on this subsequence to generate the first density new type uncertainty sub-analysis result. Finally, the sub-analysis result is included in the density new type uncertainty analysis result. This step will be explained in detail in the following content.
[0035] Subsequently, the density base class uncertainty analysis results obtained for the density sequence are incorporated into the overall base class uncertainty analysis results, and the density new class uncertainty analysis results are incorporated into the overall new class uncertainty analysis results, thus completing the corresponding integration of the two types of density analysis results.
[0036] Finally, the process of performing base class and new class uncertainty analysis on the sonic transit time sequence and natural gamma logging data sequence is similar to the analysis of the density sequence described above. Base class uncertainty analysis involves denoising, equidistant resampling, and kernel density curve probability distribution identification to obtain corresponding results. New class uncertainty analysis involves gradient trend segmentation to obtain corresponding subsequences, extracting representative subsequences to complete the new class uncertainty analysis and obtain sub-analysis results. Then, the base class uncertainty analysis results for the sonic transit time sequence and natural gamma logging data sequence are incorporated into the overall base class uncertainty analysis results, and the new class uncertainty analysis results are incorporated into the overall new class uncertainty analysis results.
[0037] Furthermore, the method provided in this application embodiment includes:
[0038] The density sequence is filtered by skip point and missing value denoising to obtain a filtered density sequence; the filtered density sequence is resampled at equal intervals to obtain an updated density sequence; the probability distribution of the filtered density sequence is identified by kernel density curve to obtain the density base class uncertainty analysis results.
[0039] Specifically, firstly, the density sequence is denoised and screened for outliers and missing values. Outlier denoising uses the 3σ criterion: the mean and standard deviation of the density sequence are calculated, and data deviating from the mean by more than three times the standard deviation are identified as outliers and removed. Missing value processing uses linear interpolation: based on the valid density data before and after the missing data, the values at the missing positions are calculated and filled in through linear fitting. After these two processes, the filtered density sequence is obtained.
[0040] Next, the selected density sequence is resampled at equal intervals. The linear interpolation method is also used: first, the preset equal sampling interval is determined, and for the blank positions in the selected density sequence caused by denoising, the values at the corresponding positions are calculated and filled according to the effective density data on both sides of the blank area in a linear proportion. After filling, the updated density sequence is obtained.
[0041] Subsequently, the probability distribution of the selected density sequences was identified using kernel density curves. First, the variation ranges of density, sonic transit time, and natural gamma ray logging data were determined. Then, the number of data points was counted according to fixed intervals, with density calculated at 0.02... The time difference of sound waves is expressed in intervals of 5. Using 5 API intervals as the unit, natural gamma logging data was analyzed. Based on the number of data points in each interval, the probability distributions of the three types of logging data were initially fitted, thus obtaining the basic uncertainty characteristics of the logging parameters. Building upon this, with density sequences as the core, a Gaussian kernel function was selected to calculate the probability density of the data in different value intervals, generating kernel density curves.
[0042] Then observe the curve shape, record its approximate range and mode position, and at the same time superimpose a smooth density curve to help judge the true shape and tail shape of the curve. The bandwidth is first set to automatic or default value, and then the bandwidth size is finely adjusted according to the stability of the curve. Finally, the empirical distribution curve and distribution appearance of the density are obtained, including whether it is approximately symmetrical, whether it is heavy-tailed, whether it is bimodal, etc. This information together constitutes the uncertainty analysis results of the density base class.
[0043] Furthermore, the method provided in this application embodiment includes:
[0044] Gradient trend segmentation is performed on the density sequence to obtain multiple density subsequences; a first density subsequence is extracted from the multiple density subsequences and a new type of uncertainty analysis is performed to obtain the first density new type uncertainty sub-analysis result; the first density new type uncertainty sub-analysis result is added to the density new type uncertainty analysis result.
[0045] Specifically, firstly, an adjacency gradient identification operation is performed on the density sequence to generate the corresponding adjacency gradient sequence. Then, adjacency gradients exceeding a preset gradient threshold are selected from the adjacency gradient sequence, and multiple splitting points are determined accordingly. Based on these splitting points, the original density sequence is mapped and split, and finally multiple independent density subsequences are obtained. This step will be explained in detail in the following content.
[0046] Next, the mean of the first density subsequence is extracted as the starting point for analysis, and a corresponding starting point neighborhood is constructed based on the preset neighborhood bandwidth. Then, the starting point neighborhood is bidirectionally diffused along its edges to obtain the first diffusion neighborhood. The difference in data volume between two adjacent neighborhoods is compared. If the difference exceeds a preset threshold, diffusion continues until the difference meets the threshold requirement, and finally, the target diffusion neighborhood is determined. Probability distribution identification is carried out based on this target neighborhood, thereby obtaining the uncertainty sub-analysis results of the first density new class. This step will be explained in detail later.
[0047] Finally, the results of the new type of uncertainty sub-analysis of the first density subsequence, obtained after performing the new type of uncertainty analysis on the first density subsequence, are directly included in the results of the new type of uncertainty analysis of density, thus completing the integration and aggregation of the sub-analysis results.
[0048] Furthermore, the method provided in this application embodiment includes:
[0049] Adjacent gradient identification is performed on the density sequence to obtain an adjacent gradient sequence; adjacent gradients exceeding a preset gradient threshold in the adjacent gradient sequence are identified to obtain multiple segmentation points; the density sequence is mapped and segmented based on the multiple segmentation points to obtain multiple density subsequences.
[0050] Specifically, the first step is to use the first-order difference method to identify the adjacent gradient of the density sequence. The density data corresponding to two adjacent collection nodes in the density sequence are selected. The density value of the previous node is subtracted from the density value of the next node to calculate the density change gradient value between adjacent nodes. According to the numbering order of the collection nodes, all the calculated density change gradient values are arranged in sequence to form a complete adjacent gradient sequence, which is used to reflect the adjacent change characteristics of the density data.
[0051] Next, multiple split points are determined. First, the mean and standard deviation of the adjacent gradient sequence are calculated. The sum of the mean and 1.5 times the standard deviation is set as the preset gradient threshold. Then, the entire adjacent gradient sequence is traversed, and the relationship between each density change gradient value and the preset gradient threshold is compared one by one. The sequence position corresponding to the adjacent gradient with the value exceeding the threshold is marked as a split point. After the comparison of all gradient values is completed, all marked positions are collected to obtain multiple split points.
[0052] Then, the density sequence is mapped and segmented based on multiple segmentation points, and stability checks are carried out simultaneously. Specifically, stability checks can be performed using a line chart visualization method. The density trend graph with depth as the horizontal axis and density value as the vertical axis is quickly plotted. The chart is observed to determine whether there is a clear trend of density gradually increasing or decreasing with depth. If such a trend exists, the distribution is constructed separately according to the stratigraphic segment to avoid mixing density data from different stratigraphic layers to form a single distribution.
[0053] When observing charts to determine the obvious trend of density with depth, the key is to see if the overall trend of the drawn line graph shows a consistent and regular change. That is, does the curve continuously move in the positive direction of the horizontal axis, accompanied by a gradual increase or decrease in the density value on the vertical axis as the depth increases? This change needs to exclude the interference of small local fluctuations, forming a clearly discernible overall trend. If such a clear trend exists, it indicates that there are regular differences in the lithology, compaction degree, etc. of the strata at different depths within the well section, and the density data of different strata have their own unique distribution characteristics. At this time, the distribution should be constructed separately according to the strata. The depth where the density trend changes should be used as the boundary. Combined with geological data, the specific stratigraphic boundary point should be determined, and the originally continuous density sequence should be divided into multiple independent sub-data groups according to different strata. Each sub-data group contains only the density data within the same stratum.
[0054] After completing the stability check, based on the determined split points, the original density sequence is divided into multiple continuous sub-intervals according to the split point positions. The density data corresponding to each sub-interval is arranged in the original node number order to form multiple independent density sub-sequences.
[0055] By using the first-order difference method to calculate the adjacent gradient, the statistical method to determine the threshold for screening the splitting point, the visualization of the trend map for stability check, and the splitting point mapping for segmentation, the density sequence was accurately segmented and multiple density subsequences that conform to the stratigraphic characteristics were obtained.
[0056] Furthermore, the method provided in this application embodiment includes:
[0057] The mean of the first density subsequence is extracted as the starting point for the new type of uncertainty analysis. Based on the first density subsequence, a new type of uncertainty analysis starting point neighborhood is constructed according to a preset neighborhood bandwidth. The neighborhood of the new type of uncertainty analysis starting point is bidirectionally diffused along its edges according to the preset neighborhood bandwidth to obtain a first diffused new type of uncertainty analysis starting point neighborhood. The difference in neighborhood data volume between the first diffused new type of uncertainty analysis starting point neighborhood and the new type of uncertainty analysis starting point neighborhood is compared. When the difference in neighborhood data volume is greater than a preset difference threshold, the neighborhood of the first diffused new type of uncertainty analysis starting point neighborhood is bidirectionally diffused along its edges according to the preset neighborhood bandwidth until the difference in neighborhood data volume obtained from two adjacent diffusions is less than or equal to the preset difference threshold, thus obtaining a target diffused new type of uncertainty analysis starting point neighborhood. Probability distribution identification is performed based on the target diffused new type of uncertainty analysis starting point neighborhood to obtain the first density new type of uncertainty sub-analysis result.
[0058] In one embodiment, the mean of the first density subsequence is first extracted, the sum of all data in the first density subsequence is calculated, and then the sum is divided by the total number of data in the subsequence. The calculated value is the mean of the first density subsequence. This mean is determined as the starting point for the new type of uncertainty analysis. The first density subsequence is usually selected from multiple density subsequences that have complete data, stable gradient trends, and can cover key formation features. This type of subsequence can reflect the core characteristics of the density data in the well section, and its analysis results can be used as a reference benchmark for subsequent analysis of other density subsequences.
[0059] Next, the specific value of the preset neighborhood bandwidth is defined. This bandwidth defines the maximum difference range between the data in the neighborhood and the starting point of the new type of uncertainty analysis. Then, all the data in the first density subsequence are traversed, and all the data whose difference from the starting point does not exceed the preset neighborhood bandwidth are selected. These selected data are integrated to form the neighborhood of the starting point of the new type of uncertainty analysis.
[0060] Then, according to the preset neighborhood bandwidth, the neighborhood of the starting point for the new type of uncertainty analysis is diffused bidirectionally at the neighborhood edge. Based on the already constructed starting neighborhood, it is expanded in both directions along the data sequence. All adjacent data outside the neighborhood edge and those whose difference with the edge data inside the neighborhood does not exceed the preset neighborhood bandwidth are included in the neighborhood range. After the first bidirectional diffusion is completed, the first diffused starting neighborhood of the new type of uncertainty analysis is formed.
[0061] Then, the difference in data volume is compared to determine whether to continue diffusion until the target diffusion new type uncertainty analysis starting neighborhood is obtained. First, the data volume of the first diffusion new type uncertainty analysis starting neighborhood and the original new type uncertainty analysis starting neighborhood are counted, and the numerical difference between the two is calculated. This difference is compared with a preset difference threshold. If the difference is greater than the preset difference threshold, the current first diffusion neighborhood is used as a basis, and bidirectional diffusion is performed again using the same preset neighborhood bandwidth. The above steps of calculating the difference and comparing the threshold are repeated until the difference in the neighborhood data volume formed by two adjacent diffusions is less than or equal to the preset difference threshold. The neighborhood at this time is the target diffusion new type uncertainty analysis starting neighborhood.
[0062] The preset difference threshold is determined by the basic data characteristics of the first density subsequence. First, the total amount of data and the data dispersion of the first density subsequence are statistically analyzed. 5%-10% of the total amount of data is used as the initial benchmark value. Then, it is fine-tuned by combining the standard deviation of the subsequence. If the data dispersion is high, the benchmark value is appropriately increased, and if the dispersion is low, it is correspondingly decreased. Finally, the preset difference threshold is determined to ensure that the threshold can avoid data mixing caused by excessive expansion of the neighborhood, while ensuring that the neighborhood covers sufficient representative data.
[0063] Finally, probability distribution identification is carried out based on the neighborhood of the starting point of the new uncertainty analysis of target diffusion. The kernel density estimation method is adopted, and all data in the neighborhood of target diffusion are used as samples. Similar to the steps of obtaining the uncertainty analysis results of density base class mentioned above, the Gaussian kernel function is selected to calculate the probability density of the data in different value intervals and generate kernel density curves. By observing the shape of the curve, the distribution characteristics of the data are determined, including whether the distribution is symmetrical, whether there are multiple peaks, tail characteristics, etc. These characteristic information together constitute the uncertainty sub-analysis results of the first density new class.
[0064] By employing the steps of determining the starting point of analysis using the arithmetic mean method, constructing the initial neighborhood using the screening method, expanding the neighborhood range using the bidirectional expansion method, determining the target neighborhood using the iterative comparison method, and identifying the probability distribution using the kernel density estimation method, the goal of accurately obtaining the analysis results of the new class of uncertain subsequences of the first density subsequence was achieved.
[0065] Furthermore, such as Figure 2 As shown, the method provided in this application embodiment includes:
[0066] Similarity identification is performed on the base class uncertainty analysis results and the new class uncertainty analysis results to construct an integrated similarity matrix; based on the integrated similarity matrix, the base class uncertainty analysis results and the new class uncertainty analysis results are respectively subjected to convolution interaction to obtain interactive base class uncertainty analysis results and interactive new class uncertainty analysis results; the interactive base class uncertainty analysis results and the interactive new class uncertainty analysis results are weighted according to preset weights to obtain the uncertainty features of logging parameters.
[0067] Optionally, the cosine similarity method is first used to identify the similarity between the uncertainty analysis results of the base class and the uncertainty analysis results of the new class. All features in both classes are extracted and converted into feature vectors. The similarity is then calculated for each corresponding feature vector of the base class and the new class, resulting in a set containing all feature similarities. Subsequently, the min-max normalization method is used to process this similarity set. By calculating the maximum and minimum values in the set, each similarity value is mapped to the 0-1 interval according to a formula. Finally, all normalized similarity values are sequentially filled into an initially empty matrix to form an integrated similarity matrix.
[0068] Next, based on the integrated similarity matrix, convolutional interactions are performed on the two types of analysis results. A one-dimensional convolution method is used. First, an appropriate one-dimensional convolution kernel is designed according to the feature dimensions of the two types of results. Using the integrated similarity matrix as the association basis, convolution operations are performed on the base class uncertainty analysis results and the new class uncertainty analysis results, respectively. During the convolution process, the convolution kernel slides along the feature sequence. By calculating the association value between local features and the similarity matrix, the cross-fusion of features from the two types of results is achieved. Finally, the two types of interactive results are output: the interactive base class uncertainty analysis result and the interactive new class uncertainty analysis result.
[0069] Finally, the two types of results after interaction are weighted according to preset weights. Specifically, empirical weights can be used to set these weights, taking into account the priority of the two types of results in collapse pressure prediction. Fixed weight values are assigned to the interaction base class uncertainty analysis results and the interaction new class uncertainty analysis results, and the sum of the weight values of the two types of results is 1. The interaction base class uncertainty analysis results are multiplied by their corresponding weights, and the interaction new class uncertainty analysis results are multiplied by their corresponding weights. The two products are then added together to obtain the well logging parameter uncertainty characteristics. Specifically, this is a comprehensive information body containing three types of parameters: density, sonic transit time, and natural gamma. It covers the probability distribution form, mean, and standard deviation of the base class analysis, as well as key information such as the special fluctuation patterns of the subsequences and the uncertainty of abnormal intervals in the new class analysis, forming a structured and standardized feature set.
[0070] By constructing an integrated similarity matrix using cosine similarity and normalization methods, realizing feature convolution interaction using one-dimensional convolution, and completing weighted calculation using empirical weighting, the results of two types of uncertainty analysis are accurately integrated, and comprehensive and reliable characteristics of well logging parameter uncertainty are obtained.
[0071] Furthermore, the method provided in this application embodiment includes:
[0072] Based on the uncertainty characteristics of well logging parameters, the uncertainty characteristics of rock mechanics parameters are determined by combining them with a rock mechanics parameter model. The mean and standard deviation of the uncertainty characteristics of well logging parameters and rock mechanics parameters are substituted into the formation pressure calculation model to obtain the uncertainty characteristics of formation pressure. Based on the uncertainty characteristics of well logging parameters, rock mechanics parameters, and formation pressure, the uncertainty characteristics of horizontal stress are derived using the Huang model and the Rosenbluthe improved method, obtaining its mean and standard deviation. By combining the uncertainty characteristics of rock mechanics parameters, formation pressure, and horizontal stress, the mean, standard deviation, and probability distribution range of collapse pressure are obtained. The mean, standard deviation, and probability distribution range of collapse pressure are used as the prediction results for collapse pressure.
[0073] In this embodiment, the Huang model is a triaxial strain geostress calculation model that considers the combined effects of overlying strata pressure and horizontal tectonic stress, introduces a tectonic stress coefficient, and is used to derive horizontal geostress with uncertainties including mean and standard deviation. The Rosenbluthe improved method is a method that calculates the mean and standard deviation of uncertain parameters, such as rock mechanics parameters and formation pressure, by inputting the mean and standard deviation of these parameters, to achieve uncertainty propagation analysis.
[0074] Optionally, firstly, based on the uncertainty characteristics of well logging parameters combined with a rock mechanics parameter model, the uncertainty characteristics of the rock mechanics parameters are determined. Using the Taylor series expansion method and the Rosenbluthe improved method, well-known to those skilled in the art, the mean and standard deviation of density, sonic transit time, and natural gamma are extracted from the uncertainty characteristics of the well logging parameters. These parameters are then substituted into a rock mechanics parameter model encompassing elastic modulus, Poisson's ratio, cohesion, and internal friction angle. The model is linearized using the Taylor series expansion method to eliminate nonlinear interference between parameters. Then, the Rosenbluthe improved method is used to calculate the mean and standard deviation of each rock mechanics parameter, ultimately determining the uncertainty characteristics of the rock mechanics parameters including these four parameters.
[0075] Next, the mean and standard deviation of the uncertainty characteristics of well logging parameters and rock mechanics parameters are substituted into the formation pressure calculation model to obtain the formation pressure uncertainty characteristics. The Eaton method, commonly used in oil and gas drilling, is selected as the formation pressure calculation model. First, the required well logging parameters (such as sonic transit time) and rock mechanics parameters (such as elastic modulus) are defined in the model. Then, the mean and standard deviation of these two types of parameters are substituted into the model one by one. Following the calculation logic of the Eaton method, and combining basic data such as formation depth and hydrostatic pressure, the mean and standard deviation of the formation pressure are derived, thus constituting the formation pressure uncertainty characteristics.
[0076] Then, combining the Huang model and the Rosenbluthe improved method, the uncertainty characteristics of horizontal geostress were derived, and its mean and standard deviation were obtained. Using the uncertainty characteristics of well logging parameters, rock mechanics parameters, and formation pressure as input data, the Huang model, well-known to those skilled in the art, was used as the core calculation model. This model comprehensively considers key parameters such as density, Poisson's ratio, formation pressure, and tectonic stress coefficient. After substituting the mean and standard deviation of the above parameters into the Huang model, the Rosenbluthe improved method was used to handle the transmission process of parameter uncertainty. The mean and standard deviation of the maximum and minimum horizontal principal stresses were calculated, forming the uncertainty characteristics of horizontal geostress.
[0077] Subsequently, by integrating three types of uncertainties—rock mechanics parameters, formation pressure, and horizontal stress—the mean, standard deviation, and probability distribution interval of the collapse pressure were obtained. The Mohr-Coulomb criterion was adopted as the core basis for the collapse pressure calculation, incorporating the mean and standard deviation of rock mechanics parameters such as cohesion and internal friction angle, formation pressure, and horizontal stress into the criterion model. Referring to the normal distribution law obtained by fitting well logging data such as density and sonic transit time, and considering the uncertainty of each parameter, the mean and standard deviation of the collapse pressure were calculated through superposition. Then, its probability distribution interval was determined based on statistical laws.
[0078] Finally, the mean, standard deviation, and probability distribution range of the collapse pressure are directly used as the collapse pressure prediction results, fully presenting the uncertainty of the collapse pressure in deep coal seams and providing intuitive reference data for drilling engineering.
[0079] By gradually deriving the uncertainty characteristics of various parameters, the system achieves the effect of accurately obtaining the prediction results of deep coal seam collapse pressure and adapting to the complex geological conditions of deep coal seams.
[0080] Furthermore, the method provided in this application embodiment includes:
[0081] Uncertainty characteristics of rock mechanics parameters include elastic modulus, Poisson's ratio, cohesion, and internal friction angle; uncertainty characteristics of formation pressure include the mean and standard deviation of formation pressure.
[0082] Optionally, the elastic modulus is a mechanical parameter that measures the ability of coal and rock to resist elastic deformation. It reflects the ease with which coal and rock deform under stress. Its uncertainty directly affects the calculation accuracy of rock mechanics models and can provide a key basis for judging the deformation trend of coal and rock under stress. Poisson's ratio is the ratio of transverse strain to longitudinal strain of coal and rock under axial stress. It is dimensionless and mainly used to characterize the correlation between transverse and longitudinal deformation of coal and rock. In the derivation of horizontal geostress in the Huang model, it can accurately reflect the Poisson effect of coal and rock, ensuring the rationality of the horizontal principal stress calculation results, and thus affecting the accuracy of collapse pressure prediction. Cohesion is... The inherent cohesion between coal and rock particles is one of the core parameters for resisting shear failure. Its magnitude determines the collapse resistance of coal and rock without external pressure. When calculating collapse pressure based on the Mohr-Coulomb criterion, the uncertainty of cohesion directly affects the determination of the lower limit of collapse pressure and is a key indicator for quantifying the collapse resistance of coal and rock. The internal friction angle is the angle between the shear surface and the direction of the force when coal and rock undergo shear failure. It reflects the magnitude of the frictional resistance between coal and rock particles. Together with cohesion, it constitutes the core element of the shear strength of coal and rock and is used to measure the ease of relative sliding between coal and rock particles in the derivation of collapse pressure.
[0083] The mean value of formation pressure is an average value of formation pressure obtained through statistical analysis. It represents the baseline level of formation pressure in the well section to be predicted, providing a stable basic parameter for the calculation of collapse pressure. It is the core benchmark value for subsequent calculations of rock mechanics parameters, horizontal stress, and the correlation between collapse pressure. The standard deviation of formation pressure is a statistical parameter that measures the dispersion of formation pressure data. It is used to quantify the fluctuation range of formation pressure in different well sections and at different depths. Its value directly reflects the degree of uncertainty of formation pressure and can provide data support for subsequent uncertainty transmission analysis, ensuring that the final collapse pressure prediction interval can accurately cover the actual range of formation pressure variation.
[0084] In summary, the deep coal seam collapse pressure prediction method based on well logging parameter uncertainty analysis provided in this application has the following technical effects:
[0085] This application obtains the logging parameter sequence of the well section to be tested, and integrates the uncertainty characteristics of the logging parameters through basic and new class uncertainty analysis. Combined with rock mechanics parameter models, Huang's model, etc., the prediction is gradually derived to obtain the mean, standard deviation and probability distribution range of the collapse pressure, thereby achieving accurate prediction of the collapse pressure of deep coal seams. This makes the prediction results more in line with the actual needs of engineering, and achieves the technical effect of avoiding the interference of parameter fluctuations on the prediction results and improving the reliability of the prediction results of the collapse pressure of deep coal seams.
[0086] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
[0087] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.
Claims
1. A method for predicting collapse pressure in deep coal seams based on uncertainty analysis of well logging parameters, characterized in that, The method includes: The well section to be tested is obtained, and data is collected from the well section to be tested according to preset indicators based on preset measurement intervals to obtain the well logging parameter sequence; Perform basic class uncertainty analysis and new class uncertainty analysis on the well logging parameter sequence to obtain the results of basic class uncertainty analysis and new class uncertainty analysis; The results of the base class uncertainty analysis and the new class uncertainty analysis are integrated to obtain the uncertainty characteristics of the logging parameters; By combining the uncertainty characteristics of well logging parameters, the collapse pressure of deep coal seams is predicted step by step, and the collapse pressure prediction results are obtained.
2. The method for predicting deep coal seam collapse pressure based on well logging parameter uncertainty analysis as described in claim 1, characterized in that, The preset parameters include density, sonic transit time, and natural gamma logging data.
3. The method for predicting deep coal seam collapse pressure based on well logging parameter uncertainty analysis as described in claim 1, characterized in that, Perform basic class uncertainty analysis and new class uncertainty analysis on the well logging parameter sequence to obtain the results of the basic class uncertainty analysis and the new class uncertainty analysis, including: The logging parameter sequence is decomposed to obtain the density sequence, sonic transit time sequence, and natural gamma logging data sequence; Perform base class uncertainty analysis on the density sequence to obtain the density base class uncertainty analysis results; Perform a new type of uncertainty analysis on the density sequence to obtain the results of the new type of density uncertainty analysis; The uncertainty analysis results of the density base class and the uncertainty analysis results of the density new class are added to the uncertainty analysis results of the base class and the uncertainty analysis results of the new class, respectively. The sonic transit time sequence and the natural gamma logging data sequence are subjected to base class uncertainty analysis and new class uncertainty analysis, and the analysis results are added to the base class uncertainty analysis results and the new class uncertainty analysis results.
4. The method for predicting deep coal seam collapse pressure based on well logging parameter uncertainty analysis as described in claim 3, characterized in that, Perform base class uncertainty analysis on the density sequence to obtain the density base class uncertainty analysis results, including: The density sequence is filtered by skip point and missing value denoising to obtain a filtered density sequence; The selected density sequence is resampled at equal intervals to obtain an updated density sequence; The probability distribution of the selected density sequence is identified by kernel density curves, and the results of density base class uncertainty analysis are obtained.
5. The method for predicting deep coal seam collapse pressure based on well logging parameter uncertainty analysis as described in claim 3, characterized in that, Perform a new type of uncertainty analysis on the density sequence to obtain the results of the new type of density uncertainty analysis, including: The density sequence is divided by gradient trend segmentation to obtain multiple density subsequences; A first density subsequence is extracted from the plurality of density subsequences and a new type of uncertainty analysis is performed to obtain the first density new type of uncertainty subsequence analysis results; Add the results of the first density new class uncertainty sub-analysis to the results of the density new class uncertainty analysis.
6. The method for predicting deep coal seam collapse pressure based on well logging parameter uncertainty analysis as described in claim 5, characterized in that, The density sequence is divided by gradient trend segmentation to obtain multiple density subsequences, including: Adjacency gradient identification is performed on the density sequence to obtain the adjacency gradient sequence; Identify adjacent gradients in the adjacent gradient sequence that exceed a preset gradient threshold to obtain multiple split points; The density sequence is mapped and segmented based on the multiple segmentation points to obtain multiple density subsequences.
7. The method for predicting deep coal seam collapse pressure based on well logging parameter uncertainty analysis as described in claim 5, characterized in that, A first density subsequence is extracted from the plurality of density subsequences and subjected to a new type of uncertainty analysis to obtain the first density new type of uncertainty sub-analysis results, including: The mean of the first density subsequence is extracted as the starting point for the new type of uncertainty analysis; Based on the first density subsequence, a new type of uncertainty analysis starting point neighborhood is constructed according to the preset neighborhood bandwidth; According to the preset neighborhood bandwidth, the neighborhood of the new type of uncertainty analysis starting point is diffused bidirectionally at the neighborhood edge to obtain the first diffused new type of uncertainty analysis starting point neighborhood. Compare the difference in neighborhood data volume between the starting neighborhood of the first diffusion new type uncertainty analysis and the starting neighborhood of the new type uncertainty analysis. When the difference in neighborhood data volume is greater than a preset difference threshold, continue to perform bidirectional diffusion at the neighborhood edge of the first diffusion new type uncertainty analysis starting neighborhood according to the preset neighborhood bandwidth until the difference in neighborhood data volume obtained from two adjacent diffusions is less than or equal to the preset difference threshold, and obtain the target diffusion new type uncertainty analysis starting neighborhood. Based on the neighborhood of the starting point of the new type of uncertainty analysis of the target diffusion, the probability distribution is identified to obtain the first density new type of uncertainty sub-analysis results.
8. The method for predicting deep coal seam collapse pressure based on well logging parameter uncertainty analysis as described in claim 1, characterized in that, The results of the base class uncertainty analysis and the new class uncertainty analysis are integrated to obtain the uncertainty characteristics of the logging parameters, including: Similarity identification is performed on the uncertainty analysis results of the base class and the uncertainty analysis results of the new class, and an integrated similarity matrix is constructed; Based on the integrated similarity matrix, convolution interaction is performed on the base class uncertainty analysis results and the new class uncertainty analysis results to obtain interactive base class uncertainty analysis results and interactive new class uncertainty analysis results. The uncertainty analysis results of the interactive base class and the uncertainty analysis results of the interactive new class are weighted according to preset weights to obtain the uncertainty characteristics of the logging parameters.
9. The method for predicting deep coal seam collapse pressure based on well logging parameter uncertainty analysis as described in claim 1, characterized in that, By combining the uncertainty characteristics of well logging parameters, the collapse pressure of deep coal seams is predicted step by step, and the prediction results are obtained, including: Based on the uncertainty characteristics of well logging parameters, the uncertainty characteristics of rock mechanics parameters are determined by combining the rock mechanics parameter model; Substituting the mean and standard deviation of the uncertainty characteristics of well logging parameters and rock mechanics parameters into the formation pressure calculation model, the uncertainty characteristics of formation pressure are obtained. Based on the uncertainty characteristics of well logging parameters, rock mechanics parameters, and formation pressure, the uncertainty characteristics of horizontal geostress are derived by combining the Huang model and the Rosenbluthe improved method, and its mean and standard deviation are obtained. By combining the uncertainties of rock mechanics parameters, formation pressure, and horizontal stress, the mean, standard deviation, and probability distribution range of collapse pressure are obtained. The mean, standard deviation, and probability distribution interval of the collapse pressure are used as the prediction results for the collapse pressure.
10. The method for predicting deep coal seam collapse pressure based on well logging parameter uncertainty analysis as described in claim 9, characterized in that, Uncertainty characteristics of rock mechanics parameters include elastic modulus, Poisson's ratio, cohesion, and internal friction angle; Uncertainty characteristics of formation pressure include the mean and standard deviation of formation pressure.