Shale oil sweet spot evaluation method affected by magmatic hydrothermal fluid based on well logging curve

Abstract

The application relates to the technical field of data analysis, in particular to a shale oil sweet spot evaluation method based on well logging curves and influenced by magmatic hydrothermal fluid, which comprises the following steps: collecting well logging and well logging data of shale oil wells, judging favorable reservoir sections through gas logging data; using a PSO particle swarm algorithm to optimize a BP neural network model, predicting TOC, and determining oil layer development sections; determining geology "sweet spot" development sections based on the change characteristics of sensitive well logging curves of each shale oil core well to be measured in a research area; using a multivariate statistical method to predict the content of each mineral based on whole rock diffraction data, using a brittle mineral content method and a rock mechanics parameter method to calculate a brittleness index, and determining engineering "sweet spots"; and combining the geology "sweet spot" and the engineering "sweet spot" to establish a shale oil "sweet spot" identification model influenced by magmatic hydrothermal fluid. The application aims to accurately evaluate shale oil "sweet spots" influenced by magmatic hydrothermal fluid.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, specifically to a method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves. Background Technology

[0002] In recent years, with the deepening of research on unconventional oil and gas resources, shale oil resources in source rock strata have begun further development. The Cretaceous dark mudstone and shale in the Gulong Depression in the northern Songliao Basin are thick, rich in organic matter, moderately mature, and have developed overpressure, making them a high-quality source rock with strong oil-generating capacity. This is conducive to the formation of industrially valuable shale oil and gas distribution within the mudstone and shale strata.

[0003] Previous in-depth studies on the identification markers and geological significance of hydrothermal fluid activity have revealed that the hydrothermal fluid activity area in the Gulong Depression in the northern Songliao Basin commonly contains characteristic minerals such as fluorite, barite, pyrite, chalcopyrite, chlorite, zeolite, anhydrite, and diaspore, accompanied by anomalous enrichment of elements such as F, Ba, S, Fe, Mg, Sr, Ce, and Eu. Simultaneously, the average temperature of the primary fluid inclusions of hydrothermal minerals is often higher than the normal formation temperature. Magmatic activity leads to accelerated maturation of total organic carbon (TOC), while also exhibiting strong dissolution capacity, high pressure, and migration capabilities, facilitating pore and fracture formation, and resulting in abundant hydrocarbon source rock resources. Therefore, a comprehensive evaluation of the "sweet spots" for shale oil development influenced by magmatic hydrothermal activity is urgently needed to improve the accuracy of shale oil well logging identification and sweet spot prediction, and to achieve a classified and graded evaluation of shale oil development sweet spots.

[0004] Currently, numerous methods exist for studying the "sweet spot" of shale oil, primarily involving determining "sweet spot" evaluation parameters to predict its outcome. However, the Gulong Shale in the Daqing Oilfield of the Songliao Basin, enriched with source rocks influenced by magmatic hydrothermal processes, presents challenges in selecting "sweet spot" evaluation parameters compared to matrix-type, interlayered, and fractured shale oils. Therefore, conventional well logging interpretation methods for shale oil "sweet spot" are not applicable, and sensitive parameters need to be reconsidered. Furthermore, the topology, weights, and thresholds of the BP neural network used for TOC prediction need optimization to adapt to the data characteristics extracted from sensitive parameters. Summary of the Invention

[0005] In view of the above, it is necessary to provide a method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves to solve the above problems.

[0006] One embodiment of this application provides a method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves, the method comprising:

[0007] Obtain the total hydrocarbon values ​​of the gas reservoirs in the study area, and classify the gas reservoirs in the study area based on the values ​​of the total hydrocarbon values;

[0008] Sensitive logging curves are obtained for each shale oil core well in each category of gas reservoir. Based on the shape characteristics of the overlapping AC and RT curves in the sensitive logging curves, and combined with the numerical relationship between the amplitude difference after overlap and TOC, the actual TOC is obtained. Feature parameters of all curves corresponding to each shale oil core well are extracted. Using the outer layer PSO algorithm, the number of hidden layers and the number of nodes in each hidden layer are determined based on the difference distribution between the TOC predicted by the neural network through feature parameters and the actual TOC. Under the network topology determined by the outer layer PSO algorithm each time, the weights and thresholds of the BP neural network are optimized using the inner layer PSO algorithm to obtain the BP neural network model optimized by the PSO algorithm. The TOC of each shale oil core well to be tested is predicted. Based on the predicted TOC and combined with the logging data of the well to be tested, the development section of the oil layer is determined.

[0009] Based on the variation characteristics of the AC and CNL curves in the sensitive logging curves of each shale oil core well in the study area, the geological "sweet spot" development zone is determined.

[0010] Based on whole-rock diffraction data, the content of each mineral is predicted using multivariate statistical methods. The brittleness index is calculated using the brittle mineral content method and the rock mechanics parameter method to determine the "sweet spot" of the project.

[0011] By combining geological and engineering "sweet spots," a model for identifying shale oil "sweet spots" affected by magmatic hydrothermal fluids is established.

[0012] Preferably, the process of obtaining the actual TOC is as follows:

[0013] Establish The explanation model, specifically the formula, is as follows:

[0014]

[0015] In the formula, , These represent the maximum and minimum values ​​on the AC curve when the AC and RT curves overlap, respectively. , These represent the maximum and minimum values ​​on the RT curve when the AC and RT curves overlap, respectively. Represents the logarithmic function with base 10; Indicates the resistivity value; Indicates the time difference of sound waves;

[0016] based on By establishing a linear relationship between the TOC measured in the core, an expression for the actual TOC can be obtained.

[0017] Preferably, the characteristic parameters of each curve specifically include: the mean, variance, extreme values, autocorrelation coefficient, spectral energy, dominant frequency component, mean slope, mean second derivative, and area enclosed by each curve and the coordinate axis for all points of each curve.

[0018] Preferably, the fitness function of the outer PSO algorithm is as follows: taking all sensitive logging curves of each shale oil coring well as a sample, taking a preset proportion of samples as training samples, and averaging the sum of squared differences between the actual TOC and the TOC predicted by the neural network for all training samples to obtain the fitness function.

[0019] Preferably, the fitness function of the inner PSO algorithm is consistent with the fitness function of the outer PSO algorithm.

[0020] Preferably, when optimizing weights based on the inner PSO algorithm, each dimension is randomized, and the initial range of the weight dimension of each dimension is adjusted: the minimum value generated by randomization of each weight dimension is used as the lower limit of each weight dimension;

[0021] The specific formula for the upper limit of each weight dimension is:

[0022]

[0023] In the formula, This represents the upper bound of the j-th weight dimension in the i-th iteration of the outer layer; This represents the maximum value generated by randomization in the j-th weight dimension. , Let represent the variance of the feature parameter corresponding to the j-th weight dimension, and represent the mean of the variances of the feature parameters corresponding to all weight dimensions, respectively. , These represent the number of nodes in the two layers connected by the j-th weight dimension.

[0024] Preferably, the BP neural network model optimized by the PSO algorithm is specifically as follows:

[0025] The range of the upper and lower limits of all dimensions obtained in the i-th iteration of the outer layer is taken as the optimization range of the inner layer under the global optimal solution obtained in the (i+1)-th iteration of the outer layer. This process is repeated until the maximum number of iterations of the inner layer PSO algorithm is reached, and the inner layer global optimal solution under the global optimal solution of the outer layer is obtained. Thus, the optimal topology, weights and threshold of the BP neural network are determined, and the BP neural network model optimized by the PSO algorithm is obtained.

[0026] Preferably, determining the development section of the oil layer includes: for each shale oil core well to be tested, if the TOC is greater than a preset value, it is judged to be a source rock rich in organic matter; otherwise, it is judged to be a source rock containing organic matter.

[0027] Preferably, the process of determining the geological "sweet spot" development segment specifically includes:

[0028] The area formed by the reverse intersection of the AC curve and the CNL curve; calculate the area difference between each depth layer and the adjacent previous depth layer;

[0029] The shale oil reservoir between the depth layer where the difference between the AC curve and the CNL curve is simultaneously greater than 0 and the adjacent upper depth layer is regarded as the geological "sweet spot" development segment.

[0030] Preferably, the determination of the engineering "sweet spot" is specifically the average value of the brittleness index obtained by the brittle mineral content method and the rock mechanical parameter method.

[0031] This application has at least the following beneficial effects:

[0032] This application considers the anomalous enrichment of elements such as F, Ba, S, Fe, Mg, and Sr in hydrothermal fluid activity zones, where temperatures are higher than normal formation temperatures, improving the physical properties of source rocks. This leads to differences in sensitive parameters and facies variations compared to conventional shale oil reservoirs. It provides a method for rapidly and accurately evaluating shale oil "sweet spots" influenced by magmatic hydrothermal fluids. This method has advantages such as high predictive effectiveness, low cost, and no limitations, providing technical support for the evaluation and development of shale oil "sweet spots."

[0033] Furthermore, a two-layer PSO is used to optimize the BP topology, weights, and thresholds. The optimization results of the outer topology guide the optimization of the inner weights and thresholds. The distribution characteristics of data features extracted from sensitive curves affected by magmatic hydrothermal fluids are also used for optimization to avoid asymmetric features, such as features with extremely large variances or values ​​biased to one end. These features tend to dominate the output of hidden layer neurons during weighted summation, leading to the suppression of other features and affecting the balance of feature extraction, thus ensuring accurate prediction of TOC. At the same time, in the optimization process, in order to address the problem that traditional methods of setting optimization boundaries based on standard deviation can produce overly broad or biased boundaries, this application uses the percentiles of the data to determine the upper and lower bounds of the search range, reducing the sensitivity to extreme values. Attached Figure Description

[0034] Figure 1 A flowchart of the method for evaluating sweet spots of shale oil affected by magmatic hydrothermal fluids based on well logging curves provided in this application;

[0035] Figure 2 A graph illustrating the correlation between the predicted and actual TOC values ​​provided in this application;

[0036] Figure 3 A schematic diagram of the five-parameter radar image of the oil and gas reservoir provided in this application;

[0037] Figure 4 A schematic diagram of the five-parameter radar image of the non-oil and gas reservoir provided in this application;

[0038] Figure 5 A diagram illustrating the interpretation results of the "sweet spot" logging of the ancient page 3HC provided for this application;

[0039] Figure 6 The image shows the interpretation results of the "sweet spot" logging of the Guye 8HC well provided for this application. Detailed Implementation

[0040] In the description of the embodiments in this application, the words "exemplary," "or," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary," "or," and "for example" is intended to present the relevant concepts in a specific manner.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application's specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0042] It should also be noted that the terms "first" and "second" in this application and its accompanying drawings are used to distinguish similar objects, rather than to describe a specific order or sequence. The methods disclosed in the embodiments of this application or the methods shown in the flowcharts include one or more steps for implementing the method. Without departing from the scope of protection of this application, the execution order of multiple steps can be interchanged, and some steps can also be deleted.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0044] This application proposes a method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal processes based on well logging curves, which is applied to the field of data analysis technology. (See attached document.) Figure 1 The method includes the following steps:

[0045] The first step is to obtain the total hydrocarbon values ​​of the gas reservoirs in the study area and classify the gas reservoirs in the study area based on the values ​​of the total hydrocarbon values.

[0046] Shale oil sweet spots typically refer to the optimal matching of hydrocarbon source lithology (total organic matter, maturity, etc.), mineral and rock framework (porosity, high-quality reservoir lithology, brittleness, etc.), and other related components (fractures, formation pressure, geostress, etc.) under different geological backgrounds.

[0047] First, logging curves and data from shale oil core wells affected by magmatic hydrothermal fluids in the study area were collected, including five conventional curves: natural gamma ray (GR), deep resistivity (RT), acoustic transit time (AC), neutron density (CNL), and density density (DEN), as well as gas logging data from whole-rock diffraction data and logging data.

[0048] Since the activity level of gas logging directly reflects the hydrocarbon content of a formation, and shale oil reservoirs are themselves source rocks, active gas logging is a primary characteristic of shale oil intervals. This application uses the total hydrocarbon content measured by gas logging to qualitatively determine favorable intervals in the reservoir. The specific determination process is as follows:

[0049] Since a high total hydrocarbon value usually reflects a high free gas content in the reservoir, reservoir classification can be quickly achieved based on the total hydrocarbon value, allowing for the selection of favorable development intervals. In this application, reservoirs are classified into Class I, Class II, and Class III based on the total hydrocarbon value and total hydrocarbon characteristics. Specifically, the measured total hydrocarbon values ​​at each depth in the study area are first obtained. The study area is then stratified according to the magnitude of the measured total hydrocarbon values. Reservoirs with a total hydrocarbon value greater than 12% are classified as Class I, indicating good gas content; reservoirs with a total hydrocarbon value in the range of 7%-12% are classified as Class II, indicating moderate gas content; and reservoirs with a total hydrocarbon value less than 7% are classified as Class III, indicating poor gas content.

[0050] The second step is to obtain the sensitive logging curves of each shale oil core well in each type of gas reservoir; based on the shape characteristics of the overlapping AC and RT curves in the sensitive logging curves, and combined with the numerical relationship between the amplitude difference after overlap and TOC, the actual TOC is obtained; the feature parameters of all curves corresponding to each shale oil core well are extracted, and the outer layer PSO algorithm is used to determine the number of hidden layers and the number of nodes in each hidden layer in the BP neural network based on the difference distribution between the TOC predicted by the neural network through the feature parameters and the actual TOC; under the network topology determined by the outer layer PSO algorithm each time, the inner layer PSO algorithm is used to optimize the weights and thresholds of the BP neural network to obtain the BP neural network model optimized by the PSO algorithm, and the TOC of each shale oil core well to be tested is predicted. Based on the predicted TOC and combined with the logging data of the well to be tested, the development section of the oil layer is determined.

[0051] Traditional backpropagation (BP) neural networks heavily rely on gradient descent-based optimization strategies, which inherently leads to several drawbacks during training: First, their performance is extremely sensitive to the initial weights and thresholds; improper initialization can result in slow convergence. Second, gradient descent is prone to getting trapped in local optima, failing to guarantee finding globally optimal network parameters and thus limiting the model's generalization ability. Therefore, this application proposes introducing particle swarm optimization (PSO) into the BP neural network model. By leveraging PSO's global search capability to optimize the network's topology, connection weights, and thresholds, the excellent global optimization ability of PSO is combined with the strong local optimization capability of BP neural networks. This improves the generalization ability and learning performance of the neural network, thereby enhancing its overall search efficiency, increasing the convergence speed of PSO, reducing computational workload, and ultimately improving the accuracy of TOC prediction.

[0052] Furthermore, magmatism leads to increased temperature and accelerated TOC maturation. Simultaneously, the reservoir's magmatic hydrothermal fluids contain abundant halogen elements such as F, Cl, and Br, exhibiting strong dissolving, high-pressure, and migration capabilities. This allows for pore and fracture formation, increasing porosity and resulting in higher AC. GR is significantly influenced by lithology. Additionally, the pores are filled with light oil, leading to lower reservoir density and higher CNL. Therefore, the natural gamma ray GR, sonic transit time AC, and neutron CNL curves are highly sensitive to magmatism. This application constructs a sample set based on these three sensitive logging curves and actual TOC data, and optimizes the BP neural network using the Particle Swarm Optimization (PSO) algorithm.

[0053] Based on the classification results of gas reservoirs in the S1 study area, three sensitive logging curves were obtained for each category of gas reservoir.

[0054] S2, for each category of gas reservoir, the acoustic transit time (AC) and resistivity curve (RT) are overlaid to establish the relationship between the overlap amplitude difference and the core measurement total equilibrium (TOC), which is then substituted into the system based on... From the TOC expression of the interpreting model, the fitting relation is obtained:

[0055]

[0056]

[0057] In the formula, , These are the coefficients obtained by fitting actual TOC data from the core samples; , These represent the maximum and minimum values ​​on the AC curve when the AC and RT curves overlap, respectively. , These represent the maximum and minimum values ​​on the RT curve when the AC and RT curves overlap, respectively. Represents the logarithmic function with base 10; Indicates the resistivity value; This represents the time difference of sound waves.

[0058] S3, construct the training set for training the BP neural network.

[0059] Because the study area contains a large number of source rocks influenced by magmatic hydrothermal fluids, the response characteristics of the three curves—natural gamma ray (GR), acoustic transit time (AC), and neutron flux density (CNL)—change significantly. Therefore, it is necessary to consider extracting sensitive parameters based on these three curves.

[0060] S31. Data is collected N times for each of the M shale oil core wells affected by magmatic hydrothermal fluids in the study area. For the three curves of each data collection, the dimensions of the three curves are unified by normalization to eliminate errors caused by the instrument system, such as depth alignment and environmental correction. In this embodiment, the maximum and minimum value normalization method is used.

[0061] S32, Outlier Removal: Noise points caused by interference sources (such as wellbore collapse or instrument drift) in the GR, AC, and CNL curves are detected and removed using statistical methods and anomaly detection methods. In this embodiment, the statistical method adopts the 3-standard deviation principle, and the anomaly detection adopts the LOF algorithm.

[0062] S33. Based on the denoised curve, extract B1 time-domain features, including mean, variance, extrema, autocorrelation coefficient, etc. in this embodiment; and B2 frequency-domain features, which are extracted by Fourier transform to the frequency domain and extract spectral energy, dominant frequency components, etc. Considering the dynamic characteristics of the natural gamma curve (GR), acoustic time difference (AC), and neutron frequency difference (CNL) curves, obtain B3 dynamic features. In this embodiment, calculate the mean slope of all points on the curve, the mean second derivative of all points, and the area enclosed by the curve and the coordinate axis.

[0063] S34, data from N acquisitions of M shale oil coring wells, yielding a total of For each sample, normalization is performed on each feature, such as Min-Max scaling or Z-Score normalization. In this embodiment, Min-Max scaling is used to avoid the model being biased towards large numerical features due to features of different scales. The sequence of feature values ​​after normalization of all features of each sample is used as the feature value sequence of each sample, and the actual TOC value corresponding to each sample is used as the sample label value.

[0064] Furthermore, the samples are divided into training and testing sets according to a ratio range of 6:4 to 9:1. In this embodiment, the samples are divided according to a ratio range of 7:3.

[0065] S4 determines the optimization target, including topology, weights, and thresholds.

[0066] S41, determine the object to be optimized in BP topology optimization.

[0067] The optimization of the topology of a BP neural network mainly involves determining the number of hidden layers and the number of nodes in each hidden layer. The number of nodes m in a hidden layer is usually related to the number of nodes in the connected input and output layers.

[0068] For any hidden layer, the number of hidden layer nodes satisfies a certain relationship with the number of connected input layer nodes and the number of output layer nodes. Taking a BP neural network with only one hidden layer as an example, we extract... The feature parameters are used as inputs to the BP neural network when predicting TOC, based on prior knowledge. ;in, This represents the number of hidden layer nodes. Since the prediction result only contains the TOC, the number of output nodes is... =1; This is a constant parameter, with a value range of 1-10. (This is from the example provided.) The value in the middle is 5; This indicates the rounding operation.

[0069] The main influencing factors for the number of hidden layers are the complexity of the sample set and the complexity of the problem to be solved. For example, in a binary classification problem, the two classes can be separated by a single dividing line, in which case one hidden layer is sufficient to represent a linear separating function. However, for more complex problems or sample sets, the number of hidden layers needs to be increased. In this application, when accurately predicting TOC using data features extracted from sensitive logging curves, the sample set has a high dimensionality, which is a multi-factor prediction problem. Therefore, a relatively high number of hidden layers should be set so that the BP neural network can handle the multi-dimensional features of the sensitive logging curve's natural gamma ray (GR), acoustic transit time (AC), and neutron flux density (CNL), ensuring the accuracy of TOC prediction.

[0070] S42, determine the objects to be optimized for the weights and thresholds of the BP neural network.

[0071] The weights and thresholds of the BP neural network are encoded into components representing particle positions, where the particle dimension equals the total number of weights and thresholds. The weights include connection weights between the input layer and hidden layers, and connection weights between the hidden layer and the output layer; the thresholds include hidden layer thresholds and output layer thresholds.

[0072] Taking a single hidden layer as an example: the number of nodes in the input layer of a BP neural network is... The number of output layer nodes is 1, and the number of hidden layer nodes is [missing information]. Therefore, the total number of weights is The total number of thresholds is m+1, and the particle dimension is... .

[0073] S5. The PSO algorithm is used to determine the optimal weights and thresholds of the BP neural network.

[0074] The goal of the outer PSO algorithm is to determine the topology of the BP neural network, i.e., the number of hidden layers and the number of nodes in each hidden layer. The goal of the inner PSO algorithm is to determine the weights and thresholds of the BP neural network. Since the topology of the BP neural network directly affects the number of weights and thresholds, in this application, the inner PSO algorithm is optimized simultaneously with the outer PSO algorithm.

[0075] Outer PSO algorithm: First, randomly generate X four-dimensional vectors, one dimension being the layer number and the other dimension being the number of nodes in each layer, represented as [layer number, number of nodes in hidden layer 1, number of nodes in hidden layer 2, number of nodes in hidden layer 3]. As an example, [2, 10, 8] represents two hidden layers, the first hidden layer has 10 nodes and the second hidden layer has 8 nodes. In this application, the optimization range is set to the number of hidden layers being 1-3 and the number of hidden layer nodes being 5-20.

[0076] Secondly, M initial particles are randomly generated, with M ranging from 20 to 50. The number of iterations is set to 50 to 200. The values ​​of the two learning factors range from 1.5 to 2, and the value of the inertia weight ranges from 0.4 to 0.9. In this embodiment, the values ​​are 30, 50, 1.5, 1.5, and 0.6, respectively, to construct the fitness function. Iterative optimization is performed within the optimization range.

[0077]

[0078] In the formula, Y is the number of training samples. , These are the actual TOC and the TOC predicted by the neural network for the i-th training sample, respectively.

[0079] Inner PSO algorithm: In this application, the optimal solution after each iteration of the outer PSO algorithm guides the inner PSO algorithm to optimize the weights and thresholds. That is, during the iteration process, the range of the constraint space during particle optimization is dynamically adjusted as the network topology changes.

[0080] It should be noted that during the optimization process of the outer PSO algorithm, when two examples with the largest fitness function values ​​appear in any iteration, the particle with more hidden layers is selected as the optimal solution. This is because, for the same data, the structure with more hidden layers and fewer nodes is easier to train than the structure with a single hidden layer and more nodes, and the data features extracted by the layer are richer.

[0081] Specifically, due to the influence of a large amount of magmatic hydrothermal fluid, the data fluctuations at different locations in the three sensitive logging curves are not stable. The extracted data features of B1, B2, and B3 are obviously asymmetric data. In order to adapt to the extracted data features, the weights and thresholds will also show a certain skewness in distribution. When setting the optimization boundary based on the standard deviation in the traditional way, the boundary will be too broad or biased. In this application, the percentile of the data is used to determine the upper and lower bounds of the search range to reduce the sensitivity to extreme values.

[0082] First, for any iteration of the outer PSO algorithm, taking the i-th iteration as an example, we obtain the number of hidden layers corresponding to the optimal solution of the i-th iteration and the number of nodes in each hidden layer. This determines the number of weights and thresholds, and yields the dimension of the particles in the inner PSO algorithm, denoted as . .

[0083] Secondly, the initial range for randomization of each dimension is [-1, 1], and the fitness function is set. This is used for iterative optimization in the inner PSO algorithm. Since the purpose of the BP neural network is to predict TOC, this application sets the fitness function... ;

[0084] Subsequently, during the optimization process of the inner PSO algorithm, the output layer threshold determines the regression baseline, reducing output bias, while the hidden layer threshold controls the activation sensitivity of neurons to avoid premature saturation. The weights affect the hidden layer's extraction of feature parameters and the final network output; therefore, the priority of the weights should be higher than the thresholds. Thus, for any weight dimension, taking the j-th corresponding weight dimension as an example, the initial range of the weights is adjusted to be inversely proportional to the variance of the input features. This is because input features with large variance tend to dominate the output of hidden layer neurons during weighted summation, leading to the suppression of other features and affecting the balance of feature extraction. The variance of the input features corresponding to the weights of the j-th dimension is calculated; the larger the variance, the smaller the corresponding weight should be.

[0085] Subsequently, the initial data for each dimension in the optimization space of the inner PSO algorithm are sorted, and an upper bound and a lower bound are set for each dimension. The upper and lower bounds of the j-th dimension are represented as follows: , Let the upper bound of the j-th weight dimension in the i-th iteration of the outer layer be expressed as... :

[0086]

[0087] In the formula, This represents the maximum value generated by randomization in the j-th weight dimension. , respectively represent the distribution variance of the characteristic parameters corresponding to the j-th weight dimension, and the mean value of the distribution variances of the characteristic parameters corresponding to all weight dimensions; and respectively represent the number of nodes in the two layers connected by the j-th weight dimension.

[0088] Among them, the role of using the number of nodes connecting the two layers as the denominator is to prevent the upper limit value from being too low. This is because the more nodes there are in the layer where the weight acts, the greater the impact on the final predicted TOC result. Therefore, the upper limit value should not be too low.

[0089] Finally, take the upper limit and lower limit (the minimum value in the random initialization) of the j-th weight dimension as the range for optimizing the j-th weight dimension, and perform the same processing for the remaining weight dimensions; for the threshold dimension, in order to maintain the training efficiency, no additional adjustment is made, and the maximum and minimum values in the result of the random initialization are used as the upper limit and lower limit respectively. Take the ranges determined by the upper and lower limits of all dimensions as the optimization range determined by the i-th iteration of the outer PSO algorithm.

[0090] Furthermore, obtain the global optimal solution at the (i + 1)-th iteration of the outer PSO algorithm, determine the dimension of the particles in the inner PSO algorithm, and perform optimization within the optimization range determined by the i-th iteration of the outer PSO algorithm, and so on until the maximum number of iterations of the inner PSO algorithm is reached, to obtain the inner global optimal solution under the outer global optimal solution, thereby determining the optimal topological structure, weights, and thresholds of the BP neural network. Among them, the PSO algorithm and the training of the neural network are both well-known technologies, and the specific process will not be elaborated here.

[0091] S6. Make a depth mark every 10 meters for the wells to be measured in the study area. According to the above sample processing process for the three sensitive curves, extract the data characteristics of the sensitive curves and input them into the trained BP neural network to predict the TOC value of the wells to be measured. When TOC > 2%, it is a rich organic matter source rock; when 0 < TOC < 2%, it is an organic matter-containing source rock. The correlation explanation diagram between the predicted TOC value and the actual value is as Figure 2 shown Figure 2 in the rightmost two columns in the figure. One column is the depth, and the other column is the measured TOC and the predicted TOC. The depth value interval is 10 meters. The visualization result in the rightmost column shows that the correlation between the measured data and the predicted data is as high as 82.6%.

[0092] The third step: Based on the variation characteristics of the AC curve and the CNL curve in the sensitive logging curves of each core well of the shale oil to be measured in the study area, determine the development section of the geological "sweet spot".

[0093] For each well to be measured in the study area, obtain the logging response characteristic radar chart. Among them, the schematic diagram of the five-parameter radar chart of the oil and gas reservoir is as Figure 3 As shown in the diagram, the five-parameter radar image of a non-oil and gas reservoir is as follows: Figure 4 As shown, the five parameters are: natural gamma GR, density DEN, neutron CNL, acoustic transit time AC, and deep lateral resistivity LLD. Figure 3 , Figure 4 The wells to be logged include: Chao 21, Guye 1, Guye 2HC, Guye 3HC, Guye 8HC, Guye 9HC, and Ying X58. This application selects sensitive logging curves that can clearly distinguish between mature and immature source rocks: the AC curve and the CNL curve. By observing the intersection of these two opposing curves, influenced by magmatic hydrothermal fluids, the dissolution porosity increases, leading to a larger AC. The main filling material in the pores is light oil, resulting in a lower oil layer density and a higher CNL. Therefore, when both AC and CNL increase simultaneously, it can be identified as a geological "sweet spot" development zone. The larger the area filled by AC and CNL, the greater the probability that the shale oil reservoir is in a developmental stage. The area filled by all curves is calculated separately, and the difference in area between each depth layer under each curve and the adjacent upper depth layer is calculated separately. A difference greater than 0 indicates growth; therefore, the shale oil reservoir between the depth where the difference between the two curves is simultaneously greater than 0 and the adjacent upper depth layer is considered a geological "sweet spot" development zone, and the greater the area growth, the more developed the shale oil reservoir.

[0094] The fourth step is to predict the content of each mineral based on whole-rock diffraction data using multivariate statistical methods, and to calculate the brittleness index using the brittle mineral content method and rock mechanics parameter method to determine the "sweet spot" of the project.

[0095] Specifically, based on the measured data of each mineral content obtained from whole-rock diffraction data, the mineral content is predicted using principal component analysis, linear dimensionality reduction, and multivariate statistics. Taking the Guye 8HC well as an example, the mineral content of the shale oil reservoir is modeled using principal component analysis.

[0096] Carbonate rocks:

[0097] Clay:

[0098] Pyrite:

[0099] Baiyunyan:

[0100] In the formula, Indicates carbonate rock content, RD represents the natural gamma value, and deep two-sided resistivity value. This represents the logarithmic function with the natural constant as base. CNL represents density, and CNL represents neutron. Indicates clay content, Indicates pyrite content, This indicates the dolomite content.

[0101] The brittleness index was calculated using the brittle mineral content method and the rock mechanics parameter method to determine the "sweet spot" of the project.

[0102] Mineral content method:

[0103] Rock mechanics parameter method:

[0104] Brittleness index:

[0105] In the formula: The brittleness index is obtained from the mineral brittleness content method. Indicates quartz content, Esta represents the brittleness index obtained by the rock mechanics parameter method, Esta is Young's modulus, Esta_min is the minimum Young's modulus, Esta_max is the maximum Young's modulus, Vsta is Poisson's ratio, Vsta_min is the minimum Poisson's ratio, and Vsta_max is the maximum Poisson's ratio.

[0106] The fifth step: combine geological and engineering "sweet spots" to establish a model for identifying shale oil "sweet spots" affected by magmatic hydrothermal fluids.

[0107] The above steps were applied to the two cored wells, Guye 3HC and Guye 8HC, in the study area. The development zones of Guye 3HC are 2310-2370m, 2400-2420m, 2430-2450m, and 2460-2480m, with the fractured zone at 2460-2480m. Currently, Guye 3HC is producing 26.8 cubic meters of oil and 16298 cubic meters of gas per day during testing. The development zones of Guye 8HC are 2316-2360m, 2375-2394m, 2412-2424m, 2436-2450m, and 2484-2496m, with the fractured zone at 2395-2418m. The interpretation results of the "sweet spot" logging for 3HC are shown in the diagram below. Figure 5 As shown in the diagram, the interpretation results of the "sweet spot" logging at Guye 8HC are as follows. Figure 6 As shown.

[0108] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0109] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves, characterized in that, The method includes the following steps: Obtain the total hydrocarbon values ​​of the gas reservoirs in the study area, and classify the gas reservoirs in the study area based on the values ​​of the total hydrocarbon values; Sensitive logging curves are obtained for each shale oil core well in each category of gas reservoir. Based on the shape characteristics of the overlapping AC and RT curves in the sensitive logging curves, and combined with the numerical relationship between the amplitude difference after overlap and TOC, the actual TOC is obtained. Feature parameters of all curves corresponding to each shale oil core well are extracted. Using the outer layer PSO algorithm, the number of hidden layers and the number of nodes in each hidden layer are determined based on the difference distribution between the TOC predicted by the neural network through feature parameters and the actual TOC. Under the network topology determined by the outer layer PSO algorithm each time, the weights and thresholds of the BP neural network are optimized using the inner layer PSO algorithm to obtain the BP neural network model optimized by the PSO algorithm. The TOC of each shale oil core well to be tested is predicted. Based on the predicted TOC and combined with the logging data of the well to be tested, the development section of the oil layer is determined. Based on the variation characteristics of the AC and CNL curves in the sensitive logging curves of each shale oil core well in the study area, the geological "sweet spot" development zone is determined. Based on whole-rock diffraction data, the content of each mineral is predicted using multivariate statistical methods. The brittleness index is calculated using the brittle mineral content method and the rock mechanics parameter method to determine the "sweet spot" of the project. By combining geological and engineering "sweet spots," a model for identifying shale oil "sweet spots" affected by magmatic hydrothermal fluids is established.

2. The method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves as described in claim 1, characterized in that, The process of obtaining the actual TOC is as follows: Establish The explanation model, specifically the formula, is as follows: ； In the formula, , These represent the maximum and minimum values ​​on the AC curve when the AC and RT curves overlap, respectively. , These represent the maximum and minimum values ​​on the RT curve when the AC and RT curves overlap, respectively. Represents the logarithmic function with base 10; Indicates the resistivity value; Indicates the time difference of sound waves; based on By establishing a linear relationship between the TOC measured in the core, an expression for the actual TOC can be obtained.

3. The method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves as described in claim 1, characterized in that, The specific characteristic parameters of each curve include: the mean, variance, extreme values, autocorrelation coefficient, spectral energy, dominant frequency component, mean slope, mean second derivative, and the area enclosed by each curve and the coordinate axis.

4. The method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves as described in claim 1, characterized in that, The fitness function of the outer PSO algorithm is as follows: take all the sensitive logging curves of each shale oil coring well as a sample, take a preset proportion of samples as training samples, and average the sum of squared differences between the actual TOC and the TOC predicted by the neural network for all training samples to obtain the fitness function.

5. The method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves as described in claim 1, characterized in that, The fitness function of the inner PSO algorithm is the same as that of the outer PSO algorithm.

6. The method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves as described in claim 1, characterized in that, When optimizing weights based on the inner PSO algorithm, each dimension is randomized, and the initial range of the weight dimension for each dimension is adjusted: the minimum value generated by randomization of each weight dimension is used as the lower limit of each weight dimension. The specific formula for the upper limit of each weight dimension is: ； In the formula, This represents the upper bound of the j-th weight dimension in the i-th iteration of the outer layer; This represents the maximum value generated by randomization in the j-th weight dimension. , Let represent the variance of the feature parameter corresponding to the j-th weight dimension, and represent the mean of the variances of the feature parameters corresponding to all weight dimensions, respectively. , These represent the number of nodes in the two layers connected by the j-th weight dimension.

7. The method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves as described in claim 1, characterized in that, The BP neural network model optimized by the PSO algorithm is specifically as follows: The range of the upper and lower limits of all dimensions obtained in the i-th iteration of the outer layer is taken as the optimization range of the inner layer under the global optimal solution obtained in the (i+1)-th iteration of the outer layer. This process is repeated until the maximum number of iterations of the inner layer PSO algorithm is reached, and the inner layer global optimal solution under the global optimal solution of the outer layer is obtained. Thus, the optimal topology, weights and threshold of the BP neural network are determined, and the BP neural network model optimized by the PSO algorithm is obtained.

8. The method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves as described in claim 1, characterized in that, The determination of the development section of the oil layer includes: for each shale oil core well to be tested, if the TOC is greater than a preset value, it is judged to be a source rock rich in organic matter; otherwise, it is judged to be a source rock containing organic matter.

9. The method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves as described in claim 1, characterized in that, The process of determining the geological "sweet spot" development zone is as follows: The area formed by the reverse intersection of the AC curve and the CNL curve; calculate the area difference between each depth layer and the adjacent previous depth layer; Shale oil reservoirs between depth layers where the difference between the AC curve and the CNL curve is simultaneously greater than 0 and the adjacent upper depth layer are considered geological "sweet spot" development zones.

10. The method for evaluating the sweet spot of shale oil affected by magmatic hydrothermal fluids based on well logging curves as described in claim 1, characterized in that, The "sweet spot" of the project is specifically the average value of the brittleness index obtained by the brittle mineral content method and the rock mechanics parameter method.

Images