Fine evaluation method for mineral composition of well section

Abstract

The invention provides an evaluation method for mineral composition of a well section to be calibrated. According to the method, technologies such as electron microscope scanning and electron probes are utilized, high-precision mineral recognition of rock debris and particles is achieved, and a data basis is provided for subsequent mineral composition analysis; meanwhile, on the basis of an image segmentation algorithm, a fine-grained rock debris sample mineral identification program is established, lithology identification, classification and mineral composition analysis based on mineral particle sizes are achieved, and a technical guarantee of fine understanding is provided for heterogeneous oil reservoirs; a multi-scale double-layer convolutional neural network of a sequence is also established, richer logging data details and global information are extracted through multi-scale feature enhancement, and accurate prediction of minerals through feature extraction of multiple pieces of logging data is realized in combination with the sequence convolutional neural network.

Description

Technical Field

[0001] The invention belongs to the technical field of petroleum logging reservoir evaluation, and relates to an evaluation method for the mineral composition of a well section to be calibrated, and in particular to a fine evaluation method for the mineral composition of the well section. Background Art

[0002] Accurate analysis of the mineral composition and lithological characteristics of rocks is essential in the exploration and production of oil and gas. This information helps geologists and engineers assess the potential value and development strategies of oil and gas reservoirs. Traditionally, core analysis and rock chip analysis are the main methods used to identify the characteristics of underground formations. Core analysis provides detailed stratigraphic information, but it is costly and the acquisition process is cumbersome. In contrast, rock chip analysis is more economical and practical, but it contains less information and is less accurate. In the field of rock chip analysis, common practices include observation using optical microscopes or electron microscopes, X-ray diffraction analysis, and the use of various scanning equipment. These devices can provide macroscopic and microscopic characteristics of rock samples, but when dealing with rock samples with strong vertical heterogeneity, these traditional techniques are often limited by large sampling intervals and strong sample mixing.

[0003] Although some technical solutions have been disclosed in existing research, using image analysis technology to perform macroscopic scanning and classification of rock cuttings, and using image recognition technology to identify different types of lithology. In addition, according to conventional logging curves, some empirical formulas after regression are used to predict the mineral composition at the current depth. However, these technical solutions still have the following shortcomings: the current logging rock cuttings identification method based on image processing is mainly aimed at natural light photos of rock cutting particles, and its resolution accuracy is limited. It can only identify lithology, and cannot directly identify mineral composition; because the rock cuttings sampling interval is 2 to 5m, and the logging is 0.125 meters per data point, its measurement accuracy is lower than that of the logging, and its samples are also mixed with rock cuttings from multiple depths, so it is impossible to locate the specific logging depth, so it is impossible to accurately calibrate the logging data for subsequent predictions; and when using empirical formulas to directly predict mineral composition through logging data, due to the fixed form of the formula and limited adjustable parameters, the calculation results have a large variance and cannot meet the accuracy requirements of use.

[0004] Therefore, how to find a more appropriate evaluation method to solve the shortcomings of existing technologies has become one of the urgent problems to be solved by many front-line researchers in the industry. Summary of the invention

[0005] In view of this, the technical problem to be solved by the present invention is to provide an evaluation method for the mineral composition of a well section to be calibrated, in particular, a fine evaluation method for the mineral composition of a well section. The present invention aims to introduce a mineral composition prediction method based on image segmentation, rock cuttings homing and sequence convolutional neural network, so as to achieve accurate prediction of different rock mineral compositions in rock cuttings using conventional logging data, thereby providing more accurate and detailed geological information for oil and gas exploration and evaluation.

[0006] The present invention provides a method for evaluating the mineral composition of a well section to be calibrated, comprising the following steps:

[0007] 1) The rock cuttings are cut, ground and polished to obtain a rock cutting material with a smooth surface, and then a conductive material is compounded on the surface of the rock cutting material to obtain a rock cutting sample;

[0008] 2) Scanning the rock cuttings sample obtained in the above steps with a SEM scanning electron microscope and / or an electron probe to obtain a colored mineral identification effect diagram of the rock cuttings sample;

[0009] 3) The mineral identification effect image obtained in the above steps is converted into a grayscale image, and then the Otsu threshold segmentation method is used to determine the optimal threshold, the rock particles and the gaps are separated, and then after fine processing, the edge of the segmented rock particles is detected to obtain the rock particle segmentation results marked with different colors;

[0010] 4) Based on the rock particle segmentation results marked with different colors obtained in the above steps, the mineral information in the particles is counted by identifying the different colors in the rock particles to form the mineral composition statistics of a single rock particle;

[0011] 5) The cuttings obtained during the logging process of the target layer section are tested in the above steps. According to the logging lithology classification results and mineral prediction results, the corresponding position relationship of the logging cuttings in the corresponding depth section is matched. Based on the lithology matching and mineral distribution, the logging-logging cuttings particles are relocated to obtain the logging curve and the mineral content of the cuttings particles after relocation;

[0012] 6) The multi-scale volatility extraction curve of the logging curve of the target layer section is combined with the mapping relationship between the single-point logging value and the rock minerals as the input data of the neural network model, and a double-layer sequence convolutional neural network model is constructed. After training the model, the trained network model is obtained;

[0013] 7) For the subsequent uncalibrated well sections of the target layer, the well logging and mud recording data of the uncalibrated well sections are used, and the trained network model obtained in the above steps is used to predict the uncalibrated well sections based on the mineral composition.

[0014] Preferably, the rock cuttings are representative rock cuttings that are selected after screening and can reflect the characteristics of the entire rock mass;

[0015] The conductive material includes a conductive carbon material;

[0016] The compounding includes spraying.

[0017] Preferably, the mineral identification effect diagram of the colored rock cuttings sample is specifically a mineral identification effect diagram of the colored rock cuttings sample represented by different colors based on the differences in the composition and structure data of the rock cuttings;

[0018] The specific method for determining the optimal threshold includes calculating the optimal threshold by minimizing the inter-class variance.

[0019] Preferably, the fine processing includes fine processing using morphological operations of erosion and dilation.

[0020] Preferably, the erosion operation is used to remove noise points;

[0021] The expansion operation is used to fill the gaps between particles that are not completely separated.

[0022] Preferably, the fine processing is specifically to perform fine processing on noise points in the separation result and / or gaps between particles that are not completely separated.

[0023] The detection is specifically performed using a Canny contour detection method.

[0024] Preferably, the logging and mud logging rock cuttings particle homing is specifically to select rock cuttings particles in the scanned sample for homing at each logging measurement point, thereby forming high-precision logging and mud logging rock cuttings particle homing at a scale of 0.125 m.

[0025] Preferably, the well logging curve and mineral content of rock cuttings particles obtained in step 5) are used as the input and label of the sample set respectively, and as the sample library for training the double-layer sequential convolutional neural network model.

[0026] Preferably, the multi-scale volatility extraction is specifically:

[0027] The logging curves are resampled at different intervals, and the points within the sampling interval are processed by linear interpolation;

[0028] When constructing the sample, the input of each sample is the data of multiple sampling frequency logging curves within 2 meters above and below a depth point. The input of each sample is a l*n two-dimensional matrix;

[0029] Among them, l represents the logging section of 2 meters above and below the depth point as the center, and n represents multiple resampled logging curves including the original logging curve.

[0030] Preferably, the sample input is the content ratio of different minerals, and the total number of samples is the number of depth points at which the rock and mineral content has been experimentally tested;

[0031] The double-layer sequential convolutional neural network model includes three convolutional layers and one fully connected layer.

[0032] The present invention provides a method for evaluating the mineral composition of a well section to be calibrated. Compared with the prior art, the method for fine evaluation of the mineral composition of a well section designed by the present invention with specific steps is also a method for fine evaluation of minerals in metamorphic rock cuttings particles. The present invention first uses rock mineral scanning electron microscope data to obtain the accurate mineral composition of each particle of the cuttings sample, and then, based on the results of the mineral analysis of the cuttings sample, the rock mineral composition is obtained and accurately calibrated on the logging scale through cuttings classification and repositioning. Finally, based on a double-layer sequential convolutional neural network, accurate mineral composition prediction based on conventional logging data is achieved.

[0033] The present invention provides a rock cutting mineral measurement and composition analysis step, in which the mineral composition of rock particles is obtained by electronic scanning of rock cutting samples according to the electronic feedback; the present invention provides a rock cutting particle segmentation, classification and logging scale calibration step, in which the rock cutting particles are segmented based on image segmentation technology, and the particles are classified according to the specific mineral composition. The mineral type predicted by the logging is calibrated, and the components of the mineral composition in the logging data are calibrated; further, the present invention also uses a multi-scale double-layer sequence convolution network to achieve accurate prediction of mineral composition, uses known results to establish a sample library, and uses a custom multi-scale double-layer sequence convolution network to achieve accurate prediction of mineral composition through conventional logging data.

[0034] The present invention utilizes technologies such as electron microscope scanning and electron probe to achieve high-precision mineral identification of rock cuttings and particles, providing a data basis for subsequent mineral composition analysis; at the same time, based on the image segmentation algorithm, a mineral identification program for fine-grained rock cuttings samples is established, which realizes lithology identification, classification and mineral composition analysis based on the mineral particle scale, providing technical support for detailed understanding of heterogeneous oil reservoirs; a sequential multi-scale two-layer convolutional neural network is also established, which extracts richer logging data details and global information through multi-scale feature enhancement, and then combines the sequential convolutional neural network to realize feature extraction of multiple logging data for accurate prediction of minerals. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Comparative diagram of the appearance of the rock cuttings samples prepared for the present invention;

[0036] Figure 2 This is a rendering of the mineral identification effect after scanning with the scanning electron microscope rockscan provided by the present invention;

[0037] Figure 3 The effect diagram of the rock particle segmentation result provided by the present invention;

[0038] Figure 4 The mineral composition analysis diagram of the rock cuttings particles provided by the present invention;

[0039] Figure 5 The multi-scale feature extraction effect diagram provided by the present invention;

[0040] Figure 6 A schematic diagram of the double-layer sequential convolutional neural network structure provided by the present invention;

[0041] Figure 7 The calibration and prediction effect of logging minerals provided by the present invention are demonstrated;

[0042] Figure 8 The effect diagram of the preparation of the metamorphic rock cuttings sample provided by the present invention;

[0043] Fig. 9 A schematic diagram of the calibration effect after the rock cuttings are segmented and reassembled provided by the present invention;

[0044] Fig.10 The content R of various minerals predicted by the logging data provided by the present invention 2 value;

[0045] Fig.11 The mineral content prediction effect (percentage) of the machine learning method provided by the present invention. DETAILED DESCRIPTION

[0046] In order to further understand the present invention, preferred embodiments of the present invention are described below in conjunction with examples, but it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention rather than limiting the patent claims of the present invention.

[0047] All noun expressions and abbreviations in the present invention are conventional noun expressions and abbreviations in the field. Each noun expression and abbreviation is clear and unambiguous in its relevant application field. Those skilled in the art can clearly, accurately and uniquely understand them based on the noun expressions and abbreviations.

[0048] The present invention provides a method for evaluating the mineral composition of a well section to be calibrated, comprising the following steps:

[0049] 1) The rock cuttings are cut, ground and polished to obtain a rock cutting material with a smooth surface, and then a conductive material is compounded on the surface of the rock cutting material to obtain a rock cutting sample;

[0050] 2) Scanning the rock cuttings sample obtained in the above steps with a SEM scanning electron microscope and / or an electron probe to obtain a colored mineral identification effect diagram of the rock cuttings sample;

[0051] 3) The mineral identification effect image obtained in the above steps is converted into a grayscale image, and then the Otsu threshold segmentation method is used to determine the optimal threshold, the rock particles and the gaps are separated, and then after fine processing, the edge of the segmented rock particles is detected to obtain the rock particle segmentation results marked with different colors;

[0052] 4) Based on the rock particle segmentation results marked with different colors obtained in the above steps, the mineral information in the particles is counted by identifying the different colors in the rock particles to form the mineral composition statistics of a single rock particle;

[0053] 5) The cuttings obtained during the logging process of the target layer section are tested in the above steps. According to the logging lithology classification results and mineral prediction results, the corresponding position relationship of the logging cuttings in the corresponding depth section is matched. Based on the lithology matching and mineral distribution, the logging-logging cuttings particles are relocated to obtain the logging curve and the mineral content of the cuttings particles after relocation;

[0054] 6) The multi-scale volatility extraction curve of the logging curve of the target layer section is combined with the mapping relationship between the single-point logging value and the rock minerals as the input data of the neural network model, and a double-layer sequence convolutional neural network model is constructed. After training the model, the trained network model is obtained;

[0055] 7) For the subsequent uncalibrated well sections of the target layer, the well logging and mud recording data of the uncalibrated well sections are used, and the trained network model obtained in the above steps is used to predict the uncalibrated well sections based on the mineral composition.

[0056] The present invention first cuts rock chips, then grinds and polishes them to obtain rock chip materials with smooth surfaces, and then compounds conductive materials on the surface of the rock chip materials to obtain rock chip samples.

[0057] In the present invention, the rock cuttings are preferably selected after screening to be representative rock cuttings that can reflect the characteristics of the entire rock mass.

[0058] In the present invention, the conductive material preferably includes a conductive carbon material.

[0059] In the present invention, the compounding preferably includes spraying.

[0060] In the present invention, the rock cuttings sample obtained in the above steps is scanned by a SEM scanning electron microscope and / or an electron probe to obtain a colored mineral identification effect diagram of the rock cuttings sample.

[0061] In the present invention, the mineral identification effect diagram of the colored rock cuttings sample is preferably formed by forming a mineral identification effect diagram of the colored rock cuttings sample represented by different colors based on the differences in the composition and structure data of the rock cuttings.

[0062] The present invention then converts the mineral identification effect image obtained in the above steps into a grayscale image, and then uses the Otsu threshold segmentation method to determine the optimal threshold, separates the rock particles and gaps, and then detects the edges of the segmented rock particles after fine processing to obtain the rock particle segmentation results marked with different colors.

[0063] In the present invention, the specific method for determining the optimal threshold preferably includes calculating the optimal threshold by minimizing the inter-class variance.

[0064] In the present invention, the fine processing preferably includes fine processing using morphological operation corrosion and dilation method.

[0065] In the present invention, the erosion operation is preferably used to remove noise points.

[0066] In the present invention, the expansion operation is preferably used to fill the gaps between particles that are not completely separated.

[0067] In the present invention, the fine processing is preferably to perform fine processing on noise points in the separation result and / or gaps between particles that are not completely separated.

[0068] In the present invention, the detection is preferably performed using a Canny contour detection method.

[0069] The present invention then obtains the segmentation results of rock particles marked with different colors based on the above steps, and counts the mineral information in the particles by identifying the different colors in the rock particles, thereby forming the mineral composition statistics information of a single rock particle.

[0070] The present invention then performs the above-mentioned detection steps on the rock cuttings obtained during the logging process of the target layer section, matches the corresponding position relationship of the logging rock cuttings in the corresponding depth section according to the lithology classification result and the mineral prediction result of the logging, and performs the logging-logging rock cutting particle relocation based on the lithology matching and mineral distribution to obtain the logging curve and the mineral content of the rock cutting particles after relocation.

[0071] In the present invention, the logging and mud logging rock cuttings particle homing is preferably to select rock cuttings particles in the scanned sample for homing at each logging measurement point, thereby forming high-precision logging and mud logging rock cuttings particle homing at a scale of 0.125 m.

[0072] In the present invention, the well logging curve and mineral content of rock cuttings particles obtained in step 5) are preferably used as the input and label of the sample set respectively, as the sample library for training the double-layer sequential convolutional neural network model.

[0073] The present invention further uses the logging curve of the target layer section for multi-scale volatility extraction, combines the mapping relationship between the single-point logging value and the rock mineral, as the input data of the neural network model, constructs a double-layer sequence convolutional neural network model, and then trains the model to obtain the trained network model.

[0074] In the present invention, the multi-scale volatility extraction is preferably:

[0075] The logging curves are resampled at different intervals, and the points within the sampling interval are processed by linear interpolation;

[0076] When constructing the sample, the input of each sample is the data of multiple sampling frequency logging curves within 2 meters above and below a depth point. The input of each sample is a l*n two-dimensional matrix;

[0077] Among them, l represents the logging section of 2 meters above and below the depth point as the center, and n represents multiple resampled logging curves including the original logging curve.

[0078] In the present invention, the sample input is preferably the content ratio of different minerals, and the total number of samples is the number of depth points at which the rock and mineral content has been experimentally tested.

[0079] In the present invention, the double-layer sequential convolutional neural network model preferably includes three convolutional layers and one fully connected layer.

[0080] Finally, the present invention uses the logging and mud recording data of the uncalibrated well section for subsequent uncalibrated well sections of the target layer, and adopts the trained network model obtained in the above steps to predict the uncalibrated well section based on mineral composition.

[0081] The present invention realizes the analysis technology of the mineral composition of rock cuttings particles by using a scanning electron microscope, realizes the technology of automatic segmentation, classification and fine placement of rock cuttings particles based on the Canny algorithm, and also establishes a sequential multi-scale two-layer convolutional neural network, ultimately realizing the accurate prediction of the mineral composition through conventional logging data.

[0082] The present invention is to complete and refine the overall evaluation method, better realize the stability of the prediction evaluation, and further improve the accuracy of the prediction evaluation. The detailed evaluation method of the mineral composition of the above-mentioned well section may specifically include the following contents:

[0083] Step 1: Rock cuttings collection and sample preparation

[0084] First, the rock cuttings samples are screened to select representative rock cuttings that can accurately reflect the overall rock mass characteristics. Subsequently, precision rock cutting equipment is used to cut the larger rock samples into small fragments suitable for scanning electron microscopy (SEM) analysis. Thereafter, fine grinding and polishing steps are performed to ensure that the sample surface reaches the required smoothness and flatness for high-resolution electron microscopy scanning. In view of the inherently non-conductive nature of rock, the present invention uses carbon spraying technology to make the sample surface conductive to avoid the accumulation of electron beams during the scanning process, which leads to a decrease in image quality. The sample preparation effect is as follows: Figure 1 As displayed. Figure 1 The comparison diagram of the appearance of the rock cuttings samples prepared by the present invention. The left diagram is the appearance of the rock cuttings sample before spraying, and the right diagram is the appearance of the rock cuttings sample before spraying.

[0085] Step 2: Mineral Scanning and Identification

[0086] Place the prepared rock cuttings sample in the sample tray or sample chamber of the scanning electron microscope. Make sure the sample is stably fixed and the scanning area is aligned with the scanning head of the device. For specific analysis requirements, it may be necessary to adjust the angle or position of the sample to obtain the best scanning angle. The device will automatically scan the selected sample area line by line, using X-rays or other detection technologies to collect composition and structural data of the rock cuttings. The entire scanning process may take from a few minutes to a few hours, depending on the set scanning parameters and sample size. The final scan result is as follows: Figure 1 Different colors represent different mineral types. Figure 2 This is a diagram showing the effect of mineral identification after scanning with the scanning electron microscope rockscan provided by the present invention.

[0087] Step 3: Rock particle segmentation

[0088] First, the color image is converted into a grayscale image because the Otsus threshold segmentation method used in the present invention is based on grayscale. Next, the present invention applies the Otsus threshold segmentation method to automatically determine the optimal threshold value to separate rock particles from the background (i.e., white gaps). Otsu's method automatically calculates the optimal threshold value by minimizing the inter-class variance. Then, for some noise in the segmentation result or small gaps between particles that are not completely separated, morphological operations erosion and expansion are used for additional processing. The erosion operation can help remove small noise points, while the expansion operation can help fill small gaps between particles. Finally, the Canny contour detection method is used to detect the edges of the segmented rock particles.

[0089] See also Figure 3 , Figure 3 This is an effect diagram of the rock particle segmentation result provided by the present invention.

[0090] Step 4: Lithology and mineral analysis and evaluation

[0091] The present invention establishes a corresponding query dictionary for different colors of minerals, as shown in Table 1, and counts the mineral information in the particles by identifying the different colors in the particles, thereby forming statistics on the mineral composition of each rock particle, such as Figure 4 shown. Figure 4 This is an analysis diagram of the mineral composition of rock cuttings particles provided by the present invention.

[0092] Table 1 Dictionary of rock chip scanning colors and minerals

[0093]

[0094]

[0095] Step 5: Lithology calibration and sample library establishment

[0096] The present invention classifies the rock cuttings particles into lithological types according to the mineral compositions of the rock cuttings particles, combines particles of the same lithological type, and performs detailed statistics of the mineral composition for each lithology.

[0097] Use logging and mud logging data to conduct vertical fine lithology division and rough mineral prediction for the target layer section. According to the logging lithology division results and mineral prediction results, match the corresponding position relationship of the logging rock cutting sample particles in the current depth section, and carry out high-precision logging-mud logging rock cutting particle relocation at a scale of 0.125m based on lithology matching and roughly the same mineral distribution. For each logging measurement point, select the rock cutting particles in the scanned sample for relocation. The relocated logging curve and rock cutting particle mineral content are used as the input and label of the sample set to form a sample library for the subsequent training of the multi-scale fluctuation model.

[0098] Step 6: Multiscale fluctuation model establishment, training and well mineral calculation

[0099] Since in the process of logging curve identification, in addition to the specific values ​​of various logging sequences, there is also the observation and analysis of logging trend information, so the method established this time, in addition to the mapping relationship between single-point logging values ​​and rock minerals, also needs to add the extraction of local multi-scale volatility in the input data. The role of multi-scaling is, on the one hand, to reduce the interference of abnormal point information through scale upgrading, and on the other hand, it allows the neural network model to pay more attention to trend information and ignore unnecessary detail information through scale upgrading. For a specific data logging curve, the present invention will perform multi-scale feature extraction on each logging data, that is, through multi-scaling, each logging curve is formed into multiple logging curves of different scales, which are used together as the input data of the subsequent neural network model, so as to retain both local feature details and global trend information.

[0100] See also Figure 5 , Figure 5 This is a diagram of the multi-scale feature extraction effect provided by the present invention.

[0101] In order to achieve this function, this paper first resamples the logging curve at different intervals. In order to retain the necessary information, linear interpolation is used for the points within the sampling interval. At this time, the low sampling frequency will retain more trend information, and the lower the sampling frequency, the higher the proportion of trend information and local information in the logging curve.

[0102] When re-sampling, the input of each sample is the data of multiple sampling frequency logging curves within 2m above a depth point. Therefore, the input of each sample is a l*n two-dimensional matrix, where l represents the logging section 2m above and below the depth point, and n represents multiple re-sampled logging curves including the original logging curve. The input of the sample is the content ratio of different minerals. The total number of samples is the number of depth points that have been experimentally tested for rock and mineral content.

[0103] Construct a two-layer sequence convolutional neural network suitable for multi-feature fluctuation sequences, see Figure 6 . Figure 6 Schematic diagram of the double-layer sequential convolutional neural network structure provided by the present invention.

[0104] The double-layer sequential convolutional neural network provided by the present invention includes three convolutional layers and one fully connected layer. The model is then trained using samples. Based on the trained network model, the logging data of the well section without rock and mineral experiments is used to automatically deduce and predict the mineral composition of the uncalibrated well section. Figure 7 . Figure 7 The present invention provides a demonstration of the logging mineral calibration and prediction effect.

[0105] The present invention realizes the acquisition of the change of mineral content along the well, and provides an important data basis for the subsequent brittleness evaluation.

[0106] The above content of the present invention provides a method for fine evaluation of the mineral composition of a well section. The present invention first uses rock mineral scanning electron microscope data to obtain the accurate mineral composition of each particle of the cuttings sample, and then, based on the results of the rock cuttings sample mineral analysis, through the classification and repositioning of the cuttings, the rock mineral composition acquisition and accurate calibration on the logging scale are realized, and finally, based on the double-layer sequence convolutional neural network, the accurate mineral composition prediction based on conventional logging data is realized.

[0107] The present invention provides a rock cutting mineral measurement and composition analysis step, in which the mineral composition of rock particles is obtained by electronic scanning of rock cutting samples according to the electronic feedback; the present invention provides a rock cutting particle segmentation, classification and logging scale calibration step, in which the rock cutting particles are segmented based on image segmentation technology, and the particles are classified according to the specific mineral composition. The mineral type predicted by the logging is calibrated, and the components of the mineral composition in the logging data are calibrated; further, the present invention also uses a multi-scale double-layer sequence convolution network to achieve accurate prediction of mineral composition, uses known results to establish a sample library, and uses a custom multi-scale double-layer sequence convolution network to achieve accurate prediction of mineral composition through conventional logging data.

[0108] The present invention utilizes technologies such as electron microscope scanning and electron probe to achieve high-precision mineral identification of rock cuttings and particles, providing a data basis for subsequent mineral composition analysis; at the same time, based on the image segmentation algorithm, a mineral identification program for fine-grained rock cuttings samples is established, which realizes lithology identification, classification and mineral composition analysis based on the mineral particle scale, providing technical support for detailed understanding of heterogeneous oil reservoirs; a sequential multi-scale two-layer convolutional neural network is also established, which extracts richer logging data details and global information through multi-scale feature enhancement, and then combines the sequential convolutional neural network to realize feature extraction of multiple logging data for accurate prediction of minerals.

[0109] In order to further illustrate the present invention, a method for evaluating the mineral composition of a well section to be calibrated provided by the present invention is described in detail below in combination with examples. However, it should be understood that these examples are implemented on the premise of the technical solution of the present invention, and detailed implementation methods and specific operating processes are given only to further illustrate the features and advantages of the present invention, rather than to limit the claims of the present invention, and the protection scope of the present invention is not limited to the following examples.

[0110] Application Examples

[0111] In the study of a metamorphic rock well project, in order to gain a deeper understanding of the reservoir characteristics and physical properties of the key well section (3400m to 4400m), the present invention used comprehensive logging technology to conduct a detailed investigation of the area. This technology includes the measurement of eight logging curves such as natural potential, acoustic wave, density, etc., and collects rock cuttings samples at a frequency of once every 2 meters for further analysis.

[0112] Nevertheless, it has proven challenging to concisely identify reservoir lithology through traditional rock cuttings logging methods. Therefore, the present invention uses technologies such as electronic probes to achieve high-definition scanning of rock cuttings samples and precise mineral composition analysis, thereby facilitating more accurate lithology identification. The first step is to collect rock cuttings samples and complete the production of samples of every 2m through a series of processes such as grinding, cementation, and carbon plating. Figure 8 . Figure 8 This is a rendering of the effect of preparing the metamorphic rock cuttings sample provided by the present invention.

[0113] Considering the time cost of high-resolution scanning, the present invention initially prepared and scanned only the cuttings samples within the depth range of 3400-3940m, and based on this, the mineral composition at each depth was recorded in detail. In view of the significant vertical heterogeneity of the well section, the present invention carefully segmented and classified the cuttings particles in the sample, and calibrated them according to the well logging and core data on the well section, ensuring the accurate correspondence between the lithology and mineral composition and the corresponding depth position.

[0114] After the sample is prepared, it is sent to the mineral analysis equipment for scanning, and the mineral composition of the cuttings sample is obtained at each depth. Since the entire well section has strong vertical heterogeneity, the cuttings particles in the sample are first segmented and combined according to different cuttings lithology, such as Fig. 9 As shown, the corresponding specific depth position of the sampling interval well section is matched. Fig. 9 This is a schematic diagram of the calibration effect after the rock cuttings segmentation and reorganization provided by the present invention.

[0115] In order to optimize the time cost and experimental budget, the present invention further utilizes the experimental data and logging information obtained, adopts machine learning methods to transform the logging data into lithology and mineral types, and predicts the lithology of the next well section. Finally, in order to verify the prediction accuracy of the model, the cuttings of the subsequent 500m well section were continuously Rockscan scanned and compared with the prediction results.

[0116] See also Fig.10 , Fig.10 The content R of various minerals predicted by the logging data provided by the present invention 2 value.

[0117] Fig.10 It can be seen that the neural network model has obtained an R of more than 0.8 through learning the mineral scanning results of the upper well section. 2 The value is the prediction result of the main mineral types with higher accuracy, especially for the brittle minerals closely related to the fracturing process, see Fig.10 . Fig.11 The mineral content prediction effect (percentage) of the machine learning method provided by the present invention.

[0118] Depend on Fig.11 It can be seen that the predicted results are highly consistent with the actual scanning results, proving the effectiveness and reliability of the adopted method. Due to space limitations, this paper only shows the prediction effects of related brittle minerals to demonstrate the results of this study and its application potential.

[0119] The above is a detailed introduction to the method for fine evaluation of the mineral composition of a well section provided by the present invention. The principle and implementation method of the present invention are explained in this article using specific examples. The description of the above embodiments is only used to help understand the method and its core idea of ​​the present invention, including the best mode, and also enables any technician in the field to practice the present invention, including making and using any device or system, and implementing any combined method. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, the present invention can also be improved and modified in several ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention. The scope of patent protection of the present invention is defined by the claims and may include other embodiments that can be thought of by those skilled in the art. If these other embodiments have structural elements that are not different from the literal expression of the claims, or if they include equivalent structural elements that are not substantially different from the literal expression of the claims, then these other embodiments should also be included in the scope of the claims.

Claims

1. A method for evaluating the mineral composition of a well section to be calibrated, It is characterized in that The following steps are involved: 1) The rock cuttings are cut, ground and polished to obtain a rock cutting material with a smooth surface, and then a conductive material is compounded on the surface of the rock cutting material to obtain a rock cutting sample; 2) Scanning the rock cuttings sample obtained in the above steps with a SEM scanning electron microscope and / or an electron probe to obtain a colored mineral identification effect diagram of the rock cuttings sample; 3) The mineral identification effect image obtained in the above steps is converted into a grayscale image, and then the Otsu threshold segmentation method is used to determine the optimal threshold, the rock particles and the gaps are separated, and then after fine processing, the edge of the segmented rock particles is detected to obtain the rock particle segmentation results marked with different colors; 4) Based on the rock particle segmentation results marked with different colors obtained in the above steps, the mineral information in the particles is counted by identifying the different colors in the rock particles to form the mineral composition statistics of a single rock particle; 5) The cuttings obtained during the logging process of the target layer section are tested in the above steps. According to the logging lithology classification results and mineral prediction results, the corresponding position relationship of the logging cuttings in the corresponding depth section is matched. Based on the lithology matching and mineral distribution, the logging-logging cuttings particles are relocated to obtain the logging curve and the mineral content of the cuttings particles after relocation; 6) The multi-scale volatility extraction curve of the logging curve of the target layer section is combined with the mapping relationship between the single-point logging value and the rock minerals as the input data of the neural network model, and a double-layer sequence convolutional neural network model is constructed. After training the model, the trained network model is obtained; 7) For the subsequent uncalibrated well sections of the target layer, the well logging and mud recording data of the uncalibrated well sections are used, and the trained network model obtained in the above steps is used to predict the uncalibrated well sections based on the mineral composition.

2. The evaluation method according to claim 1, It is characterized in that The rock cuttings are specifically representative rock cuttings that are selected after screening and can reflect the characteristics of the entire rock mass; The conductive material includes a conductive carbon material; The compounding includes spraying.

3. The evaluation method according to claim 1, It is characterized in that The mineral identification effect diagram of the colored rock cuttings sample is specifically a mineral identification effect diagram of the colored rock cuttings sample represented by different colors based on the differences in the composition and structure data of the rock cuttings; The specific method for determining the optimal threshold includes calculating the optimal threshold by minimizing the inter-class variance.

4. The evaluation method according to claim 1, It is characterized in that The detailed processing includes performing detailed processing by using morphological operation corrosion and dilation method.

5. The evaluation method according to claim 4, It is characterized in that The erosion operation is used to remove noise points; The expansion operation is used to fill the gaps between particles that are not completely separated.

6. The evaluation method according to claim 1, It is characterized in that The detailed processing specifically includes detailed processing of noise points in the separation result and / or gaps between particles that are not completely separated. The detection is specifically performed using a Canny contour detection method.

7. The evaluation method according to claim 1, It is characterized in that The logging and mud logging rock cuttings particle homing is specifically to select rock cuttings particles in the scanned sample for homing at each logging measurement point, thereby forming a high-precision logging and mud logging rock cuttings particle homing at a scale of 0.125 m.

8. The analysis and evaluation method according to claim 1, It is characterized in that The well logging curve and rock cutting particle mineral content obtained in step 5) are used as the input and label of the sample set respectively, and as the sample library for training the double-layer sequence convolutional neural network model.

9. The analysis and evaluation method according to claim 1, It is characterized in that The multi-scale volatility extraction is specifically as follows: The logging curves are resampled at different intervals, and the points within the sampling interval are processed by linear interpolation; When constructing the sample, the input of each sample is the data of multiple sampling frequency logging curves within 2 meters above and below a depth point. The input of each sample is a l*n two-dimensional matrix; Among them, l represents the logging section of 2 meters above and below the depth point as the center, and n represents multiple resampled logging curves including the original logging curve.

10. The analysis and evaluation method according to claim 9, It is characterized in that The sample input is the content ratio of different minerals, and the total number of samples is the number of depth points where the rock and mineral content has been experimentally tested; The double-layer sequential convolutional neural network model includes three convolutional layers and one fully connected layer.

Images