Free oil porosity earthquake quantitative prediction method and system
Through high-resolution reservoir prediction technology and deep learning random forest algorithm, combined with seismic inversion data, a free oil porosity prediction model was established, which solved the problem that free oil porosity cannot be quantitatively predicted in the shale oil area, and achieved high precision The free oil porosity prediction supports the precise positioning and development of shale oil desserts.
Patent Information
- Application Number
- CN202310653144.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-02
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2043-06-02
AI Technical Summary
The existing technology cannot effectively and quantitatively predict the porosity of free oil in the shale oil area, resulting in difficulty in accurately identifying and characterizing shale oil desserts, affecting the drilling rate of high-quality desserts in shale oil horizontal wells and the optimization of segmentation and clustering schemes. .
Using high-resolution reservoir prediction technology and deep learning random forest algorithm, single-well free oil porosity data is calculated through nuclear magnetic well logging, combined with seismic inversion data, a machine learning prediction model for free oil porosity is established, and K-mean pooling is used. Class analysis and random forest algorithm make multiple predictions, and finally achieve quantitative prediction of free oil porosity through the KRR model.
Accurate prediction of the porosity of free oil in the shale oil area is achieved, the prediction ability of thin desserts is improved, and the prediction accuracy is 90%, which is more than 20% higher than conventional methods.
Smart Images

Figure CN119065008A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of oil and gas exploration, and particularly relates to a free oil porosity seismic quantitative prediction method and system. Background Art
[0002] my country is rich in shale oil resources, with recoverable resources of more than 13 billion tons and huge resource potential. Accelerating the exploration and development of shale oil resources is a realistic way to ensure national energy security. Shale oil can be divided into adsorbed oil and free oil according to the difference in its occurrence state. Free oil can be effectively developed through volume fracturing, while adsorbed oil needs to be transformed into free oil in situ to achieve effective exploitation because of its large adsorption energy. Therefore, the quantitative characterization and regional range prediction of free oil porosity are the core technologies for the precise positioning and evaluation of high-quality sweet spots in shale oil. However, the porosity of adsorbed oil and free oil is affected by many factors such as the pore size of shale reservoirs, crude oil components, mineral composition and thin interlayer development. Continuous quantitative characterization and large-scale prediction are difficult, which seriously affects the fine evaluation of shale oil sweet spots and restricts the improvement of the drilling rate of high-quality sweet spots in shale oil horizontal wells and the optimization of segmentation and clustering schemes.
[0003] At present, the existing technology based on the accurate calculation of free oil porosity in a single well based on experiments and nuclear magnetic logging technology has solved the problem of continuous characterization of free oil porosity, and has been applied to the Jimusar Shale Oil National Demonstration Area and the Mabei Fengcheng Formation Shale Oil with good results. However, it is difficult to effectively control the shale oil sweet spots in the region as a whole by relying solely on the data of a single well point. It is also necessary to accurately predict the free oil porosity in the region to achieve accurate identification and characterization of high-quality shale oil sweet spots.
[0004] However, there is no mature technology that can quantitatively predict free oil porosity. Commonly used seismic prediction techniques use some seismic elastic parameters to establish a relationship with free oil porosity, and indirectly predict free oil porosity with inverted elastic parameters. However, the premise for such methods to be effective is that there is a clear correspondence between seismic elastic parameters and free oil porosity, and such correspondence is often difficult to obtain in actual situations. Seismic elastic parameters are affected by various formation properties and reflect the comprehensive response of underground conditions, including the influence of free oil porosity, as well as the influence of other fluids, rock skeletons, etc. How to extract the relationship with free oil porosity from seismic elastic parameters is the key to quantitative prediction. Summary of the invention
[0005] In order to achieve accurate prediction of free oil porosity in shale oil sweet spots in the region and meet the current needs of fine-target exploration and development of shale oil, the present invention, based on an in-depth investigation of existing free oil prediction methods, proposes a free oil porosity seismic prediction method and system based on the characteristics of shale oil reservoirs by using high-resolution reservoir prediction technology and deep learning random forest algorithm.
[0006] The technical solution adopted by the present invention is as follows: a method for quantitative seismic prediction of free oil porosity, the method comprising the following steps:
[0007] 1) Use nuclear magnetic logging to calculate the free oil porosity data of the shale oil layer in a single well, and obtain high-quality near, medium and far offset partial stacked seismic data;
[0008] 2) Use the shale oil rock physics interpretation chart to clarify the relationship between seismic elastic parameters and free oil porosity, and perform targeted sample processing on the high-resolution seismic inversion elastic data obtained by pre-stack statistical inversion to form machine learning prediction feature value data for free oil porosity;
[0009] 3) Perform K-means clustering analysis on sensitive seismic attribute data and targeted seismic inversion elastic data to establish a random forest free oil content prediction model;
[0010] 4) Obtain the initial prediction data of free oil porosity according to the random forest free oil content prediction model, and then input it as the characteristic value into the random forest free oil porosity prediction model again to complete the second free oil porosity prediction and establish the KRR free oil porosity seismic prediction model;
[0011] 5) Inputting the machine learning prediction feature value data for free oil porosity and the sensitive seismic attribute data body into the KRR free oil porosity seismic prediction model to obtain the final prediction data body of free oil porosity;
[0012] 6) Verify the final prediction data of free oil porosity, and quantitatively characterize the high-quality sweet spots of shale oil within the work area by combining the seismic inversion elastic parameter body.
[0013] Further, the step 2) is specifically as follows: constructing a rock physics interpretation map of shale oil reservoirs through rock physics analysis, and determining according to the map that the longitudinal wave impedance corresponding to the high-quality sweet spot with oil saturation>15% is greater than 9200g / cm 3 *m / s and less than 13000g / cm 3 *m / s, the ratio of P-wave to S-wave velocity is less than 1.85; the high-resolution P-wave impedance and high-resolution P-wave velocity ratio data obtained by pre-stack statistical inversion are processed, and the data corresponding to the non-sweet spots of mudstone and tight formations outside the range are eliminated. Then the processed data are subjected to a variety of mathematical transformations to form the eigenvalue data for machine learning prediction of free oil porosity.
[0014] Furthermore, in step 3), the K-means cluster analysis is performed by a K-means algorithm, and the K-means algorithm comprises the following steps:
[0015] ① Initialization: Randomly select K sample points as K initial cluster centers;
[0016] ② Cluster the samples: Calculate the distance from each sample to each cluster center, and classify each sample into the class with the closest cluster center. The distance calculation method of the cluster center is the Euclidean distance d, and the formula is as follows:
[0017]
[0018] Among them, x i Represents the horizontal coordinate value of the i-th point in the straight coordinate system, y i Represents the ordinate value of the i-th point in the straight coordinate system;
[0019] ③ Calculate new cluster centers: Calculate the mean of all samples in each class as the new cluster center;
[0020] ④If the cluster center does not change or reaches the end condition, stop the iteration, otherwise repeat ② and ③.
[0021] Furthermore, random forest is a classifier based on decision trees. The ID3 algorithm is used to calculate the data feature sample entropy to evaluate the decision tree branches. The calculation formula for the data feature sample entropy is as follows:
[0022]
[0023] Among them, H(X) is the data feature sample entropy, P i (X) is the probability function of the data feature, and n is the number of branch layers;
[0024] Each upper node is divided into multiple lower nodes, each node has data feature sample entropy, and the difference between the data feature sample entropy of the upper node and the weighted average of the data feature sample entropy of all the lower nodes is calculated as the data gain to optimize the decision tree;
[0025]
[0026]
[0027] Among them, k is the number of nodes, N(x j ) is the number of samples of the lower-level node, N is the total number of samples of the corresponding upper-level node, I(x j ) Data feature sample entropy; and They represent the entropy of the upper node and the lower node respectively, and Data Gain is the average difference between the entropy of the upper node and the lower node.
[0028] Furthermore, the step 1) includes: using nuclear magnetic resonance logging to calculate the free oil porosity data of the shale oil layer section of a single well, wherein a portion of the single wells are marked as learning wells and the remaining single wells are marked as verification wells, and the selection criteria for the learning wells are representative wells in the work area, which can reflect the geological characteristics of the target work area and are evenly distributed within the work area; obtaining high-quality near, medium and far offset partial stacking seismic data.
[0029] Furthermore, prestack statistical inversion is based on a Markov chain Monte Carlo algorithm that converges to the total within the actual time limit. During the inversion process, the variogram, probability distribution function, logging, seismic and lithofacies information are integrated, and a corresponding probability distribution model is established based on seismic data analysis. The variogram and probability distribution function are obtained through geological knowledge and well data analysis. Starting from the well point, the well point is made to comply with the original seismic information. The random simulation module is used in combination with the inversion software to generate inter-well wave impedance. The sub-wave extracted by deterministic inversion is convolved with the reflection coefficient converted from the inter-well wave impedance to generate a seismic synthetic record, and then the new sub-wave is extracted to generate a new synthetic record. This iteration is repeated until the matching rate of the synthetic seismic trace and the original seismic data reaches more than 85%, so that the correlation coefficient meets the pre-stack inversion requirements, and finally the inversion parameters are debugged to complete the inversion calculation.
[0030] Furthermore, the variogram is a distance function, which describes the change of the spatial distribution characteristics of a certain attribute with distance, and is expressed as the semi-variance of the increment of the regionalized variable Z(x) at two points x and x+h;
[0031]
[0032] Among them, r(x,h) is the experimental variogram, x represents the absolute position of a point in space, h represents the lag distance of the variogram, and h is the lag distance of the experimental variogram.
[0033] Furthermore, the three characteristic values of the sill value, range and nugget constant of the variogram reflect the spatial variation characteristics of the reservoir parameters; the range is the spatial distance when the variogram reaches a certain stable value, reflecting the average scale of the carrier of the regionalized variable in a certain direction or predicting the extension scale of the sand body in a certain direction.
[0034] In addition, the present invention also provides a free oil porosity seismic quantitative prediction system, the system comprising:
[0035] The calculation module is used to calculate the free oil porosity data of the shale oil layer section of a single well using nuclear magnetic logging to obtain high-quality near, medium and far offset partial stacked seismic data;
[0036] The sample-targeted processing module is used to use the shale oil rock physics interpretation chart to clarify the relationship between seismic elastic parameters and free oil porosity, and to perform sample-targeted processing on the high-resolution seismic inversion elastic data volume obtained by prestack statistical inversion to form machine learning prediction feature value data for free oil porosity;
[0037] K-means clustering analysis module, which is used to perform K-means clustering analysis on sensitive seismic attribute data and targeted seismic inversion elastic data, and establish a random forest free oil content prediction model;
[0038] Establish a module for obtaining the initial prediction data of free oil porosity according to the random forest free oil content prediction model, and then input it as a feature value into the second random forest free oil porosity prediction model to complete the second free oil porosity prediction and establish the KRR free oil porosity seismic prediction model;
[0039] A free oil porosity final prediction data body module is used to input the free oil porosity machine learning prediction feature value data and the sensitive seismic attribute data body into the KRR free oil porosity seismic prediction model to obtain the free oil porosity final prediction data body;
[0040] The verification module is used to verify the final prediction data of free oil porosity, and quantitatively characterize the high-quality sweet spots of shale oil within the work area in combination with the seismic inversion elastic parameter body.
[0041] Furthermore, the sample-specific processing module includes a determination unit and a processing unit;
[0042] The determination unit is used to construct a rock physics interpretation map of shale oil reservoirs through rock physics analysis, and determine that the longitudinal wave impedance corresponding to the high-quality sweet spot with oil saturation>15% is greater than 9200g / cm according to the map. 3 *m / s and less than 13000g / cm 3 *m / s, the ratio of longitudinal and transverse wave velocities is less than 1.85;
[0043] The processing unit is used to process the high-resolution P-wave impedance and high-resolution P-wave velocity ratio data obtained by prestack statistical inversion, eliminate the data corresponding to the non-sweet spots of mudstone and tight formations outside the range, and then perform multiple mathematical transformations on the processed data to form machine learning prediction feature value data for free oil porosity.
[0044] The present invention establishes a method and system for seismic quantitative prediction of free oil porosity, which overcomes the problem that shale oil free oil porosity cannot be accurately predicted on a large scale. In the prediction process, a high-resolution seismic elastic parameter data body of pre-stack statistical inversion is used, and a set of machine learning prediction characteristic value data for free oil porosity is constructed based on rock physics analysis, which improves the prediction ability of free oil porosity in thin sweet spots. Finally, the KRR free oil porosity seismic prediction model is used to deeply mine the free oil porosity information in seismic elastic parameters and sensitive seismic attributes to achieve quantitative prediction of free oil porosity, with a prediction accuracy of 90%, which is more than 20% higher than the prediction accuracy of conventional methods.
[0045] Other features and advantages of the present invention will be described in the following description, and part of them will become obvious from the description, or be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the description and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0047] Figure 1 It is a flowchart of a method for quantitative seismic prediction of free oil porosity;
[0048] Figure 2 This is a schematic diagram of the variance analysis fitting;
[0049] Figure 3a It is the pre-stack statistical inversion result of the pre-stack statistical inversion of the longitudinal wave impedance profile; Figure 3b Prestack statistical inversion Prestack statistical inversion results of P-wave and S-wave velocity ratio profiles;
[0050] Figure 4 It is a rock physics plate for shale oil reservoirs;
[0051] Figure 5a It is the characteristic value data of the processed high-resolution longitudinal impedance profile; Figure 5b It is the characteristic value data of the high-resolution P-wave and S-wave velocity ratio profile after processing;
[0052] Figure 6 This is the K-means cluster analysis result diagram;
[0053] Figure 7 Construct a flow chart for the free oil porosity seismic prediction model;
[0054] Figure 8a Comparison between free oil porosity fitted by elastic parameter multivariate regression and free oil porosity calculated by nuclear magnetic logging; Figure 8b The free oil porosity predicted in the embodiment of the present invention is compared with the free oil porosity calculated by nuclear magnetic logging;
[0055] Fig. 9 It is a quantitative prediction profile of free oil porosity in shale oil layers in the example work area;
[0056] Fig.10 This is a comparison chart between the free oil porosity prediction results and the NMR logging interpretation results;
[0057] Fig.11a This is the plane distribution map of free oil porosity in the main layer No. 1 of the upper sweet spot; Fig.11b This is the plane distribution map of free oil porosity in the main layer No. 2 of the upper sweet spot; Fig.11c This is the plane distribution map of free oil porosity in the main layer No. 1 of the lower sweet spot; Fig.11d This is the plane distribution map of free oil porosity in the main layer No. 2 of the lower sweet spot;
[0058] Fig.12 Schematic diagram of a seismic quantitative prediction system for free oil porosity. DETAILED DESCRIPTION
[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0060] A method for quantitative seismic prediction of free oil porosity, such as Figure 1 As shown, the method comprises the following steps:
[0061] 1) Use nuclear magnetic logging to calculate the free oil porosity data of the shale oil layer in a single well, and obtain high-quality near, medium and far offset partial stacked seismic data;
[0062] 2) Use the shale oil rock physics interpretation chart to clarify the relationship between seismic elastic parameters and free oil porosity, and perform targeted sample processing on the high-resolution seismic inversion elastic data obtained by pre-stack statistical inversion to form machine learning prediction feature value data for free oil porosity;
[0063] 3) Perform K-means clustering analysis on sensitive seismic attribute data and targeted seismic inversion elastic data to establish a random forest free oil content prediction model;
[0064] 4) Obtain the initial prediction data of free oil porosity according to the random forest free oil content prediction model, and then input it as the characteristic value into the random forest free oil porosity prediction model again, complete the second free oil porosity prediction, and establish the KRR free oil porosity seismic prediction model; K refers to K-means cluster analysis, and R refers to the random forest machine learning algorithm. This method uses the free oil porosity prediction model obtained by K-means cluster analysis and two random forest machine learning algorithms, so it is named KRR free oil porosity seismic prediction model;
[0065] 5) Inputting the machine learning prediction feature value data for free oil porosity and the sensitive seismic attribute data body into the KRR free oil porosity seismic prediction model to obtain the final prediction data body of free oil porosity;
[0066] 6) Verify the final prediction data of free oil porosity, and quantitatively characterize the high-quality sweet spots of shale oil within the work area by combining the seismic inversion elastic parameter body.
[0067] Step 1) is specifically as follows: using nuclear magnetic resonance logging to calculate the free oil porosity data of the shale oil layer section of a single well, wherein a portion of the single wells are marked as learning wells and the remaining single wells are marked as verification wells, and the selection criteria for the learning wells are wells that are representative of the work area, can reflect the geological characteristics of the target work area and are evenly distributed within the work area; obtaining high-quality near, medium and far offset partial stacked seismic data; using high-quality seismic data to predict free oil porosity can fundamentally guarantee the accuracy of the prediction results.
[0068] Step 2) is as follows: Based on the relationship between free oil porosity and seismic elastic parameters analyzed on the well, the P-wave impedance and P-wave velocity ratio with good correlation with free oil porosity are selected as sensitive parameters of free oil porosity. Through rock physics analysis, a rock physics interpretation map of shale oil reservoirs is constructed (e.g. Figure 4 shown), according to Figure 4 The chart shown shows that the high-quality sweet spot with oil saturation >15% corresponds to a longitudinal wave impedance greater than 9200g / cm 3 *m / s and less than 13000g / cm 3 *m / s, and the ratio of P-wave velocity to S-wave velocity is less than 1.85. Based on the analysis results, the high-resolution P-wave impedance and high-resolution P-wave velocity ratio data obtained by pre-stack statistical inversion are processed to remove the data corresponding to the non-sweet spots of mudstone and dense formations outside the range. Figure 5a and Figure 5b They are the high-resolution P-wave impedance profile and high-resolution P-wave velocity ratio profile processed by the above method. The corresponding data of mudstone and tight formation are set to 0.0001, such as 5a and Figure 5bThen, each processed data is subjected to a variety of mathematical transformations (square, reciprocal, logarithm, etc.) to form a set of machine learning prediction feature value data for free oil porosity.
[0069] Table 1 shows the correlation between the characteristic value data of the present invention and the common elastic parameters and the free oil porosity. It can be seen from the data in Table 1 that, compared with the high-resolution longitudinal wave impedance and high-resolution longitudinal-to-segmental wave velocity ratio data obtained by inversion, the set of characteristic value data for free oil porosity machine learning prediction constructed by the present invention has a higher correlation with the free oil porosity, which improves the stability of the machine learning algorithm and thus improves the prediction accuracy of the free oil porosity.
[0070] Table 1 Correlation statistics between characteristic value data of the present invention and common elastic parameters and free oil porosity
[0071]
[0072] Prestack statistical inversion is based on the Markov chain Monte Carlo algorithm that converges to the whole within the actual time limit. It is an inversion method that organically combines random inversion technology and prestack simultaneous inversion technology. In the inversion process, the variogram, probability distribution function, well logging, seismic and lithofacies information are integrated, and the corresponding probability distribution model is established according to the actual situation of seismic data. The variogram and probability distribution function are obtained through geological knowledge and well data analysis. Starting from the well point, the well point is made to comply with the original seismic information. Combined with the corresponding inversion software, the random simulation module is used to generate the inter-well wave impedance, and the sub-wave extracted by deterministic inversion is convolved with the reflection coefficient converted from the inter-well wave impedance to generate a seismic synthetic record, and then the new sub-wave is extracted to generate a new synthetic record. This iteration is repeated until the matching rate of the synthetic seismic trace and the original seismic data reaches more than 85%, that is, the correlation coefficient meets the pre-stack inversion requirements, and finally the inversion parameters are debugged to complete the inversion calculation. Figure 3a and Figure 3b They are the high-resolution P-wave impedance profile and the high-resolution P-wave and S-wave velocity ratio profile obtained by prestack statistical inversion. Since the high frequency of the inversion results comes from the logging data and the lateral trend is controlled by seismic, thinner shale oil sweet spots can be continuously identified on the inversion profile.
[0073] In order to make the random simulation meet the conditions for establishing the statistical correlation function between the spatial reservoir parameter points, the constructed variogram can be used in the spatial data field to describe the relationship between different data. The variogram r(x,h) is a distance function that describes the change of the spatial distribution characteristics of a certain attribute with distance, which can be expressed as the semi-variance of the increment of the regionalized variable Z(x) at points x and x+h:
[0074]
[0075] In formula (1), the ordinate r(x,h) is the experimental variogram, and the abscissa h is the lag distance of the experimental variogram. Figure 2 The figure shows the fitting diagram of variance analysis. Figure 2 There are three main characteristic values, namely the nugget constant, the sill value and the range, which can usually be obtained by fitting the experimental variogram using the theoretical model. Geostatistical inversion first establishes a geological model of the reservoir in the seismic time domain according to the geological conditions, in which the seismic horizon determines the plane, and the original wave impedance curve at the well location needs to be placed in the stratigraphic grid. Then, the inversion parameters are determined using seismic data and wells, thereby performing geostatistical inversion calculations. The three main characteristic values of the sill value, range and nugget constant of the variogram can be used to reflect the spatial variation characteristics of the reservoir parameters. The range is the spatial distance when the variogram reaches a certain stable value, which can reflect the average scale of the carrier of the regionalized variable in a certain direction, and can also represent the extension scale of the predicted sand body in a certain direction, thereby achieving the purpose of predicting the scale of the sand body.
[0076] What is adopted in the embodiment of the present invention is the random inversion technique, and the random inversion technique requires the input of the normally distributed data, so in addition to the need to determine the parameters required for conventional seismic inversion and the variogram (also called the variogram) of the longitudinal and transverse wave impedance and the inversion parameters determined by statistical analysis, the correlation coefficient between the density, and the parameter distribution histogram, it is also necessary to use a suitable transformation method to convert the non-normal actual data into normally distributed data. Wherein the variogram is usually characterized by parameters such as the minimum range, the maximum range azimuth, the vertical range, and the maximum range, and then the variogram is fitted by a statistical method to obtain the seismic inversion elastic data body such as the high-resolution longitudinal wave impedance and the longitudinal and transverse wave velocity ratio. The selection of the inversion parameter combination is based on the actual situation of the work area (as shown in Table 2), which can ensure that the thin sweet spot with a thickness of about 4m is accurately identified in the inversion result.
[0077] Table 2 Prestack statistical inversion parameters
[0078] Horizontal (unit: meter) Vertical (unit: milliseconds) Minimum range 600 2 Maximum range 1500 4
[0079] Step 3) is specifically as follows: K-means cluster analysis is performed on the seismic attributes extracted from the earthquake and their mathematical transformation bodies and a set of free oil porosity machine learning prediction feature value data constructed in step 2), the sample values are divided into K categories, and K random forest free oil porosity prediction models are established based on the K category sample values to predict the first multi-class free oil porosity prediction results, and the second random forest model is used to predict the final free oil porosity prediction results with high precision using the multi-class free oil porosity results and their mathematical transformations as samples, thereby obtaining the KRR free oil porosity seismic prediction model. Figure 7The flowchart of constructing the above-mentioned free oil porosity seismic prediction model is shown. First, the four types of seismic attributes obtained by K cluster analysis are respectively input as characteristic value data into the random forest free oil porosity prediction model to obtain the four free oil porosity prediction data bodies of the initial prediction. Then, the four free oil porosity prediction data bodies of the primary prediction are input as characteristic value data into the random forest free oil porosity prediction model to obtain the final free oil porosity prediction data body. The combination of two free oil porosity random forest machine learning enables the prediction model to fully mine the input characteristic value information and improve the prediction accuracy of free oil porosity.
[0080] The K-means algorithm adjusts the location of the cluster center and the affiliation of all sample points in the direction of minimizing the sum of the distances of all samples of the same class from the cluster center of that class through continuous iteration, until the best K cluster centers are found and all samples have found the best affiliation.
[0081] The K-means algorithm consists of four steps:
[0082] ① Initialization: Randomly select K sample points as K initial cluster centers;
[0083] ② Clustering samples: Calculate the distance from each sample to each cluster center, and classify each sample into the class with the closest cluster center. The distance calculation method used in the present invention is the Euclidean distance d, as shown in formula (2):
[0084]
[0085] Among them, x i Represents the horizontal coordinate value of the i-th point in the straight coordinate system, y i Represents the ordinate value of the i-th point in the straight coordinate system;
[0086] ③ Calculate new cluster centers: Calculate the mean of all samples in each class as the new cluster center;
[0087] ④If the cluster center no longer changes or reaches the cutoff condition, stop the iteration, otherwise repeat ② and ③.
[0088] Using K-means cluster analysis, seismic attributes and targeted high-resolution inversion elastic parameter volumes and their mathematical transformation data were classified. A total of four cluster centers were selected to divide the data into four categories. The classification results are as follows: Figure 6 As shown in Figure 2, the four selected cluster centers are as follows: Figure 6 As shown in the middle five-pointed star, the seismic attributes and the targeted high-resolution inversion elastic parameter data are divided into four categories according to the K-means clustering algorithm. The same type of data is represented by the same symbol, close to the cluster center, and has similar properties.
[0089] In machine learning, random forest is a classifier based on decision trees. The key to its algorithm is to determine the splitting attributes of decision tree nodes. There are many node splitting algorithms. The ID3 algorithm is suitable for the classification of discrete data such as seismic data and well logging data. The decision tree branches are evaluated by calculating the entropy balance of data samples. The convergence of the decision tree is the process of stable change after the entropy is reduced to a certain value. The calculation formula of the data feature sample entropy is shown in (3):
[0090]
[0091] Among them, H(X) is the data feature sample entropy, P i (X) is the probability function of the data feature, and n is the number of branch layers;
[0092] Each upper node is divided into multiple lower nodes, each node has data feature sample entropy, and the difference between the data feature sample entropy of the upper node and the weighted average of the data feature sample entropy of all its lower nodes is calculated as the data gain to optimize the decision tree.
[0093]
[0094]
[0095] Among them, k is the number of nodes, N(x j ) is the number of samples of the lower-level node, and N is the total number of samples of the corresponding upper-level node; I(x j ) Data feature sample entropy; and They represent the entropy of the upper node and the lower node respectively, and Data Gain is the average difference between the entropy of the upper node and the lower node.
[0096] The present invention establishes a random forest machine learning model based on the above principle, and uses the free oil porosity calculated by nuclear magnetic logging of the learning well as a learning sample to train the random forest model. The four categories of data obtained by clustering are respectively input into the trained random forest free oil porosity prediction model as characteristic values to obtain four free oil porosity initial prediction data bodies, which are then input into the random forest free oil porosity prediction model as input characteristic values to complete the second free oil porosity prediction, and the KRR free oil porosity seismic prediction model established by the present invention is obtained.
[0097] Figure 8a The medium-dark gray curve (representing the predicted value) is the free oil porosity curve obtained by multivariate regression of the measured density, P-wave velocity, S-wave velocity, wave velocity ratio and other parameters of Well 1. There is a large error compared with the free oil porosity of the shale oil layer calculated by nuclear magnetic resonance logging (light gray curve, representing the measured value), and the agreement between the two is only 64%; Figure 8bThe middle dark grey curve (representing the predicted value) is the free oil porosity curve predicted by the random forest free oil porosity quantitative prediction method designed by the present invention, which has a high consistency with the free oil porosity of the shale oil layer calculated by nuclear magnetic resonance logging in Well 1 (light grey curve, representing the measured value), and the agreement is as high as 89%. This also proves that the prediction accuracy of the free oil porosity designed by the present invention is much higher than that of the general method.
[0098] The shale oil layer section in the example work area can be roughly divided into two sweet spots according to the actual exploration and development situation, namely the upper sweet spot and the lower sweet spot. Each sweet spot section is divided into multiple main layers according to the oil production capacity, which are the target layers for fine characterization of the sweet spot.
[0099] Fig. 9 In order to use the present invention to quantitatively predict the free oil porosity of the shale oil layer section in the example work area, four types of strata are divided according to the free oil porosity: Class I oil layer, Class II oil layer, Class III oil layer and non-oil layer. It can be seen from the profile that the present invention can accurately predict the free oil porosity of the thin sweet spot of about 4m. For example, the two thin sweet spots in the upper sweet spot of Well 1 and the two thin sweet spots in the lower sweet spot of Well 2 have a good correspondence between the predicted free oil porosity and the well section with good display on the well. According to the free oil porosity, four types of strata are divided, namely Class I oil layer, Class II oil layer, Class III oil layer and non-oil layer, realizing the classification evaluation of shale oil sweet spots.
[0100] In order to further verify the prediction accuracy of free oil porosity of the present invention, the trajectory of the horizontal well (well 3) was projected onto the free oil porosity prediction section of the present invention, and the nuclear magnetic resonance and conventional logging data of the well were combined for verification (such as Fig.10 As shown in Figure 1, the nuclear magnetic logging interpretation of well 4 shows that the drilling rate of high-quality sweet spots exceeds 90%, and its trajectory is located in the layer with high free oil porosity on the free oil porosity profile predicted by the present invention. The well exits the layer near the target point and does not encounter the target layer, which corresponds to the area with low free oil porosity. The prediction result is consistent with the actual drilling result, which proves that the free oil porosity prediction accuracy of the present invention is high. Fig.11a This is the plane distribution map of free oil porosity in the main layer No. 1 of the upper sweet spot; Fig.11b This is the plane distribution map of free oil porosity in the main layer No. 2 of the upper sweet spot; Fig.11c This is the plane distribution map of free oil porosity in the main layer No. 1 of the lower sweet spot; Fig.11d This is the plane distribution map of free oil porosity in the main layer of the lower sweet spot No. 2. It can be seen that the distribution patterns of free oil porosity in each main oil-producing section are different, which means that the distribution positions of high-quality sweet spots are different and the drilling positions are also different.
[0101] In addition, the present invention also provides a free oil porosity seismic quantitative prediction system, such as Fig.12 As shown, the system comprises:
[0102] The calculation module is used to calculate the free oil porosity data of the shale oil layer section of a single well using nuclear magnetic logging to obtain high-quality near, medium and far offset partial stacked seismic data;
[0103] The sample-targeted processing module is used to use the shale oil rock physics interpretation chart to clarify the relationship between seismic elastic parameters and free oil porosity, and to perform sample-targeted processing on the high-resolution seismic inversion elastic data volume obtained by prestack statistical inversion to form machine learning prediction feature value data for free oil porosity;
[0104] K-means clustering analysis module, which is used to perform K-means clustering analysis on sensitive seismic attribute data and targeted seismic inversion elastic data, and establish a random forest free oil content prediction model;
[0105] Establish a module for obtaining the initial prediction data of free oil porosity according to the random forest free oil content prediction model, and then input it as a feature value into the second random forest free oil porosity prediction model to complete the second free oil porosity prediction and establish the KRR free oil porosity seismic prediction model;
[0106] A free oil porosity final prediction data body module is used to input the free oil porosity machine learning prediction feature value data and the sensitive seismic attribute data body into the KRR free oil porosity seismic prediction model to obtain the free oil porosity final prediction data body;
[0107] The verification module is used to verify the final prediction data of free oil porosity, and quantitatively characterize the high-quality sweet spots of shale oil within the work area in combination with the seismic inversion elastic parameter body.
[0108] Specifically, the sample targeted processing module includes a determination unit and a processing unit;
[0109] The determination unit is used to construct a rock physics interpretation map of shale oil reservoirs through rock physics analysis, and determine that the longitudinal wave impedance corresponding to the high-quality sweet spot with oil saturation>15% is greater than 9200g / cm according to the map. 3 *m / s and less than 13000g / cm 3 *m / s, the ratio of longitudinal and transverse wave velocities is less than 1.85;
[0110] The processing unit is used to process the high-resolution P-wave impedance and high-resolution P-wave velocity ratio data obtained by prestack statistical inversion, eliminate the data corresponding to the non-sweet spots of mudstone and tight formations outside the range, and then perform multiple mathematical transformations on the processed data to form a set of machine learning prediction feature value data for free oil porosity.
[0111] In summary, the actual application results in the example work area show that the present invention can accurately predict the free oil porosity of unconventional shale reservoirs, thereby achieving accurate positioning of high-quality sweet spots for shale oil, and providing important technical support for the well location, well network deployment, and well trajectory design optimization of exploration wells, evaluation wells, and development horizontal well groups.
[0112] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for quantitative prediction of free oil porosity by seismic analysis, the method comprising the following steps: 1) Use nuclear magnetic logging to calculate the free oil porosity data of the shale oil layer in a single well and obtain high-quality near, medium and far offset partial stacked seismic data; 2) Use the shale oil rock physics interpretation chart to clarify the relationship between seismic elastic parameters and free oil porosity, and perform targeted sample processing on the high-resolution seismic inversion elastic data obtained by pre-stack statistical inversion to form machine learning prediction feature value data for free oil porosity; 3) Perform K-means clustering analysis on sensitive seismic attribute data and targeted seismic inversion elastic data to establish a random forest free oil content prediction model; 4) Obtain the initial prediction data of free oil porosity according to the random forest free oil content prediction model, and then input it as the characteristic value into the random forest free oil porosity prediction model again to complete the second free oil porosity prediction and establish the KRR free oil porosity seismic prediction model; 5) Inputting the machine learning prediction feature value data for free oil porosity and the sensitive seismic attribute data body into the KRR free oil porosity seismic prediction model to obtain the final prediction data body of free oil porosity; 6) Verify the final prediction data of free oil porosity, and quantitatively characterize the high-quality sweet spots of shale oil within the work area by combining the seismic inversion elastic parameter body.
2. The method according to claim 1, wherein: The step 2) is specifically as follows: The rock physics interpretation chart of shale oil reservoirs was constructed through rock physics analysis. According to the chart, it was determined that the longitudinal wave impedance corresponding to the high-quality sweet spot with oil saturation >15% was greater than 9200g / cm 3 *m / s and less than 13000g / cm 3 *m / s, the ratio of P-wave to S-wave velocity is less than 1.85; the high-resolution P-wave impedance and high-resolution P-wave velocity ratio data obtained by pre-stack statistical inversion are processed, and the data corresponding to the non-sweet spots of mudstone and tight formations outside the range are eliminated. Then the processed data are subjected to a variety of mathematical transformations to form the eigenvalue data for machine learning prediction of free oil porosity.
3. The method according to claim 1, wherein: In step 3), the K-means cluster analysis is performed by a K-means algorithm, which includes the following steps: ① Initialization: Randomly select K sample points as K initial cluster centers; ② Cluster the samples: Calculate the distance from each sample to each cluster center, and classify each sample into the class with the closest cluster center. The distance calculation method of the cluster center is the Euclidean distance d, and the formula is as follows: Among them, x i Represents the horizontal coordinate value of the i-th point in the straight coordinate system, y i Represents the ordinate value of the i-th point in the straight coordinate system; ③ Calculate new cluster centers: Calculate the mean of all samples in each class as the new cluster center; ④If the cluster center no longer changes or reaches the cutoff condition, stop the iteration, otherwise repeat ② and ③.
4. The method according to claim 3, wherein: Random forest is a classifier based on decision trees. The ID3 algorithm is used to calculate the data feature sample entropy to evaluate the decision tree branches. The calculation formula for the data feature sample entropy is as follows: Among them, H(X) is the data feature sample entropy, P i (X) is the probability function of the data feature, and n is the number of branch layers; Each upper node is divided into multiple lower nodes, each node has data feature sample entropy, and the difference between the data feature sample entropy of the upper node and the weighted average of the data feature sample entropy of all the lower nodes is calculated as the data gain to optimize the decision tree; Among them, k is the number of nodes, N(x j ) is the number of samples of the lower-level node, N is the total number of samples of the corresponding upper-level node, I(x j ) Data feature sample entropy; and They represent the entropy of the upper node and the lower node respectively, and Data Gain is the average difference between the entropy of the upper node and the lower node.
5. The method according to claim 1, wherein: The step 1) comprises: using nuclear magnetic resonance logging to calculate the free oil porosity data of the shale oil layer section of a single well, wherein a part of the single wells are marked as learning wells and the remaining single wells are marked as verification wells, and the selection criteria of the learning wells are wells that are representative in the work area, can reflect the geological characteristics of the target work area and are evenly distributed within the work area; obtaining high-quality near, medium and far offset partial stacked seismic data.
6. The method according to claim 2, wherein: Prestack statistical inversion is based on the Markov chain Monte Carlo algorithm that converges to the whole within the actual time limit. In the inversion process, the variogram, probability distribution function, well logging, seismic and lithofacies information are integrated, and the corresponding probability distribution model is established according to the seismic data analysis. The variogram and probability distribution function are obtained through geological understanding and well data analysis. Starting from the well point, the well point is made to comply with the original seismic information. The random simulation module is combined with the inversion software to generate the inter-well wave impedance. The sub-wave extracted by deterministic inversion is convolved with the reflection coefficient converted from the inter-well wave impedance to generate a seismic synthetic record, and then the new sub-wave is extracted to generate a new synthetic record. This iteration is repeated until the matching rate of the synthetic seismic trace and the original seismic data reaches more than 85%, so that the correlation coefficient meets the pre-stack inversion requirements, and finally the inversion parameters are debugged to complete the inversion calculation.
7. The method according to claim 6, wherein: The variogram is a distance function that describes the change in the spatial distribution characteristics of a certain attribute with distance, and is expressed as the semi-variance of the increment of the regionalized variable Z(x) at two points x and x+h; Among them, r(x,h) is the experimental variogram, x represents the absolute position of a point in space, h represents the lag distance of the variogram, and h is the lag distance of the experimental variogram.
8. The method according to claim 7, wherein: The three characteristic values of the sill value, range and nugget constant of the variogram reflect the spatial variation characteristics of the reservoir parameters; The range is the spatial distance when the variogram reaches a certain stable value, reflecting the average scale of the carrier of the regionalized variable in a certain direction or predicting the extension scale of the sand body in a certain direction.
9. A free oil porosity seismic quantitative prediction system, the system comprising: The calculation module is used to calculate the free oil porosity data of the shale oil layer section of a single well using nuclear magnetic logging, and obtain high-quality near, medium and far offset partial stacked seismic data; The sample-targeted processing module is used to use the shale oil rock physics interpretation chart to clarify the relationship between seismic elastic parameters and free oil porosity, and to perform sample-targeted processing on the high-resolution seismic inversion elastic data volume obtained by prestack statistical inversion to form machine learning prediction feature value data for free oil porosity; K-means clustering analysis module, which is used to perform K-means clustering analysis on sensitive seismic attribute data and targeted seismic inversion elastic data, and establish a random forest free oil content prediction model; Establish a module for obtaining the initial prediction data of free oil porosity according to the random forest free oil content prediction model, and then input it as a feature value into the second random forest free oil porosity prediction model to complete the second free oil porosity prediction and establish the KRR free oil porosity seismic prediction model; A free oil porosity final prediction data body module is used to input the free oil porosity machine learning prediction feature value data and the sensitive seismic attribute data body into the KRR free oil porosity seismic prediction model to obtain the free oil porosity final prediction data body; The verification module is used to verify the final prediction data of free oil porosity, and quantitatively characterize the high-quality sweet spots of shale oil within the work area in combination with the seismic inversion elastic parameter body.
10. The system according to claim 9, wherein: The sample targeted processing module includes a determination unit and a processing unit; The determination unit is used to construct a rock physics interpretation map of shale oil reservoirs through rock physics analysis, and determine that the longitudinal wave impedance corresponding to the high-quality sweet spot with oil saturation>15% is greater than 9200g / cm according to the map. 3 *m / s and less than 13000g / cm 3 *m / s, the ratio of longitudinal and transverse wave velocities is less than 1.85; The processing unit is used to process the high-resolution P-wave impedance and high-resolution P-wave velocity ratio data obtained by prestack statistical inversion, eliminate the data corresponding to the non-sweet spots of mudstone and tight formations outside the range, and then perform multiple mathematical transformations on the processed data to form machine learning prediction feature value data for free oil porosity.
Citation Information
Patent Citations
Method for quantificationally predicting sandstone reservoir fluid saturation by combining well and seism
CN101887132A
Compact glutenite gas reservoir quantificational prediction method
CN105044770A
Shale gas reservoir free gas content earthquake predicating method and system thereof
CN107797139A
Shale oil analysis method and device capable of continuously characterizing contents of adsorbed oil and free oil
CN110687612A
Shale gas reservoir logging porosity rapid prediction method and system
CN110717249A
