CN110502569A - A Visual Analysis Method for Standard Well Screening Based on Discrete Selection Model - Google Patents

Abstract

The invention discloses a visual analysis method for standard well screening based on a discrete selection model, comprising: maintaining the overall spatial distribution of standard wells by using an adaptive blue noise sampling model; Measure the similarity degree of well logging from the angle of analysis; integrate the prior knowledge of expert users, design a standard well screening method based on discrete selection model, maximize the utility between attribute similarity and standard wells; design an iterative interactive standard well screening method , which allows users to change standard wells based on prior knowledge, iteratively update discrete selection models, and optimize standard well screening processes and results. Based on the comprehensive consideration of logging spatial distribution and multi-dimensional attribute information, the present invention realizes the screening of supervised standard wells driven by visual analysis, and the obtained standard wells can represent the surrounding well logging to the greatest extent, which is beneficial to subsequent formation intelligent matching Increased accuracy and efficiency.

Description

A Visual Analysis Method for Standard Well Screening Based on Discrete Selection Model

technical field

The invention relates to a standard well screening visual analysis method based on a discrete selection model, belonging to the technical fields of petroleum exploration, graphics and visualization.

Background technique

Three-dimensional seismic wave data and one-dimensional well logging data are two common data types in the field of geological structure interpretation. A large number of research work focuses on the above two data types to carry out geological structure interpretation work. Seismic wave data is obtained after recording the subsurface waveform signal with the help of proprietary equipment and performing a series of preprocessing on it. It helps users to interpret geological structures through analysis methods such as time slicing, volume extraction, and so on. For example, Patel et al. designed a seismic wave data analysis method based on two-dimensional slices (Patel D, Giertsen C, Thurmond J, et al. The seismic analyzer: interpreting and illustrating 2D seismic data [J]. IEEE transactions on visualization & computer graphics. 2008, 14 (6): 1571-1578.), the user can pre-decompose the horizon structure of the seismic wave data, and use drawing algorithms such as deformation texture and texture transfer function to enhance the display of geological structure features, thereby improving the interpretation accuracy of geological structures and reducing geological Explain the workload of the staff. However, many factors in fact affect the generation and processing of seismic wave data, such as complex terrain conditions, limited equipment conditions and inevitable calculation errors, which bring great uncertainty to the interpretation of geological structures ( Zhou B, Hatherly P, Sun W. Enhancing the detection of small coal structures by seismic diffraction imaging[J]. International Journal of Coal Geology, 2017, 178(6): 1-12.).

Different from the seismic wave data obtained indirectly, the one-dimensional logging data is directly collected from the actual logging, and a variety of attributes closely related to the geological structure are recorded in detail. For example, the spontaneous potential property can record the electrochemical and electrokinetic interactions between the formation and the mud, which can be used to determine the permeability of rock formations and water-flooded layers (Li J, Liu D, Yao Y, et al. Evaluation of the reservoir permeability of anthracite coals by geophysical logging data[J]. International Journal of Coal Geology, 2011, 87(2):121-127.). The property of acoustic transit time is to determine the porosity of the rock formation by the propagation velocity of the acoustic wave in the formation, so as to identify the lithology (Tang X M, Zheng Y, Patterson D. Processing array acoustic-logging data to imagenear-borehole geologic structures [J]. Geophysics, 2007 , 72(2):87-97.). Stratigraphic correlation based on multi-attribute logging data is an important step in geological structure interpretation, and has received more and more attention in the field of geological exploration, such as dual-well matching algorithm based on cross-optimization algorithm, rule-based expert system algorithm, etc.

Therefore, a representative subset of standard wells was selected from the original large number of well logs (Ren Z H, Zhang WC, Zhu X M, et al. Method to Determine the Collar Section Depth in Standard WellLogging [J]. Petroleum Drilling Techniques, 2014 (6):68-72.), accurate and detailed stratigraphic comparison can quickly understand the geological conditions in the local area, and then effectively guide the stratigraphic matching of large-scale original logging, avoid a large number of repetitive manual marking processes, and to a certain extent It has improved the efficiency of geological exploration and development. However, the selection of standard wells relies on the prior knowledge of experts. In particular, the traditional standard well selection process relies heavily on manual marking. The screening process is not only complicated and time-consuming, but also often fails to consider the spatial distribution of well logs and the correlation of multiple attributes. This brings great uncertainty to the subsequent interpretation of geological structures, and it is difficult to improve the efficiency and accuracy of standard well selection. Therefore, in order to overcome the limitation of manual selection of standard wells, and comprehensively consider the spatial distribution of logging and the correlation of multi-dimensional attributes, a visual analysis method for supervised standard well screening based on discrete selection model is proposed.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a visual analysis method for standard well screening based on discrete selection model, so as to realize supervised standard well screening in the process of data visualization, reduce errors and uncertainties, and improve logging data Visualize Efficiency.

In order to achieve the above purpose, the technical solution adopted in the present invention is: a visual analysis method for standard well screening based on discrete selection model, which specifically includes the following steps:

(1) Based on the multi-dimensional logging data, the spatial distribution characteristics of the logging and the stratigraphic correlation of the multi-dimensional logging attributes were extracted respectively: on the premise of determining the standard well screening ratio, the adaptive blue noise sampling algorithm was used to obtain the spatial distribution of the wells. At the same time, a multi-scale formation matching model based on dynamic programming algorithm is designed to quantitatively measure the similarity of logging attributes from the perspective of different attributes.

(2) Obtain the standard well sample set according to the prior knowledge of experts, use the discrete selection model to maximize the utility between the similarity degree of logging attributes described in step 1 and the standard well screening, and traverse the adaptive blue noise sampling algorithm to obtain of Poisson disk, standard wells are screened according to the discrete selection model, and standard well screening information is obtained.

(3) According to the standard well screening information, guide the user to interactively designate or change standard wells through visual analysis technology, adaptively update the standard well sample set information, optimize discrete selection model parameters, and re-screen to meet the current needs of users. and standard well of experience.

Further, in step (1), the method for obtaining a Poisson disk satisfying local spatial distribution by using an adaptive blue noise sampling algorithm is specifically:

1) According to the spatial location of the logging, the blue noise sampling algorithm is used to obtain a Poisson disk that satisfies the local spatial distribution, and only one log is allowed to be sampled inside each Poisson disk;

2) Use the kernel density estimation algorithm to evaluate the spatial distribution of the log in the Poisson disk, calculate the spatial distribution density value of the log, and then adaptively update the size of the Poisson disk radius to maintain the Spatial distribution characteristics of well logging.

Further, in step (1), the method for designing a multi-scale formation matching model based on a dynamic programming algorithm is specifically:

1) According to the multi-dimensional logging data, use median filtering to smooth each logging attribute curve, nonlinearly process the multi-dimensional logging data, and normalize between 0 and 1 according to the discrete results;

2) According to the smooth logging curve, use the activity function to identify and divide the stratigraphic horizon boundary for logging;

3) The optimal horizon matching sequence between the two wells is solved by the dynamic programming algorithm, and the stratum thickness corresponding to the best matching sequence on each logging attribute is summed to obtain the quantitative attribute difference between the two wells.

Further, in step (2), the method of using the discrete selection model to maximize the utility between the similarity degree of the logging attributes described in step 1 and the standard well screening is specifically:

1) According to the Poisson disk that satisfies the local spatial distribution, use the prior knowledge of experts to obtain a standard well sample set that is manually marked;

2) Design a fixed utility function according to the multi-dimensional attribute similarity between the Poisson disk logs, use a multinomial regression model to simulate the random error distribution, and combine the two to obtain a total utility function, which is used to calculate the different logging selected by different users. The probability of being a standard well;

3) According to the standard well sample set, the discrete selection model is used to automatically calculate the corresponding parameters in the total utility function and traverse the Poisson disk obtained by the adaptive blue noise sampling algorithm, and then select the attribute combination in the local space. Standard wells to maximize user utility.

Further, the method of guiding the user to interactively specify or change the standard well through the visual analysis technology described in step (3) is specifically:

1) Design a variety of visualization schemes to collaboratively display logging data information: map the actual spatial location of logging to the map, and retain the visualization results of its spatial distribution characteristics; The same attribute data of different logs in the interior are counted in the specified attribute space, and the standard wells and non-standard wells are distinguished by tone mapping technology; the difference between the logging curves is quantified by variance, and the standard wells in the local space are obtained by statistics. The total attribute difference histogram of , preserving the contrast of attribute features.

2) According to the visualization results, design a standard well screening visualization scheme, count the feature matching results on each attribute of each local space, and obtain a multi-dimensional attribute horizon matching correlation matrix; set a screening ring according to the horizon matching results, which is intuitive on the map Contrast and enhance the display of property differences between logs in a local space.

3) According to the comparison of the attribute differences, support user interactive analysis and replacement of standard wells, and then update samples, optimize discrete selection model parameters according to the new sample set, and recalculate and screen standard wells that meet user needs.

Compared with the prior art, the present invention has the beneficial effects that: in the standard well screening process, not only the spatial distribution of well logging and the multi-dimensional attribute correlation of logging data are comprehensively considered, but also the prior knowledge of experts in the field of quantitative evaluation is quantified. The use of visual analysis technology to achieve supervised standard well screening effectively reduces the workload of manual marking and improves the accuracy and efficiency of standard well screening. To sum up, the parameter adjustment process in the standard well selection model in the method of the present invention is simple and easy to implement, the feature information in the logging data can be displayed quickly and intuitively through a variety of collaborative visual designs, and a convenient multi-functional The visual analysis system supports the dynamic update and optimization of standard wells, which can significantly improve the analysis efficiency in the process of standard well selection, improve the accuracy of standard well selection, and can effectively meet the real-time application needs of users.

Description of drawings

Fig. 1 is the schematic flow chart of the method of the present invention;

Figure 2 is a schematic diagram of the blue noise sampling model, wherein (a) is the initial schematic diagram, (b), (c) are example diagrams of the standard well selection process;

3 is a schematic diagram of a multi-scale formation matching model based on a dynamic programming algorithm;

Figure 4 is a system interface diagram (geographic view), wherein (a) is an information view, (b) is a sampling panel, (c) is a selection optimization panel, (d) is a geographic view, and (e) is a multi-dimensional attribute view, (f) is the stratigraphic matching view, (g) is the stratigraphic correlation matrix view, and (h) is the total attribute difference view;

Figure 5 is a graph of the property evaluation results of the selected standard wells at a given sampling rate, in which (a) is the map projection result of the selected standard wells, (b) is the logging property curve drawing results of the standard wells, (c) is the multi-dimensional attribute similarity performance of standard wells, (d) is the total attribute difference result of standard wells, (e) is the currently calculated user preference coefficient result, and (f) is the actual formation correlation degree between wells.

Figure 6 is a schematic diagram of the comparative analysis of the standard wells after updating the samples, in which (a) is the user preference coefficient result before the sample is updated, (b) is the stratigraphic correlation result on the multi-dimensional attributes of the standard wells before the sample is updated, (c) is the recalculated user preference coefficient result after the sample is updated, (d) is the stratigraphic correlation result on the multi-dimensional attributes of the standard well after the sample update.

Detailed ways

The method for visual analysis of standard wells based on the discrete selection model of the present invention will be further described below with reference to the accompanying drawings.

Figure 1 is a flowchart of a visual analysis method for standard well screening based on discrete selection model of the present invention, and the method is specifically:

(1) Based on the multi-dimensional logging data, the spatial distribution characteristics of the logging and the stratigraphic correlation of the multi-dimensional logging attributes were extracted respectively: on the premise of determining the standard well screening ratio, the adaptive blue noise sampling algorithm was used to obtain the spatial distribution of the wells. Poisson disk; based on the actual complex geological conditions, a multi-scale formation matching model based on the dynamic programming algorithm is designed to quantitatively measure the similarity of logging attributes from the perspective of different attributes.

Figure 2 is a schematic diagram of a blue noise sampling model, and the described method for obtaining a Poisson disk that satisfies local spatial distribution by using an adaptive blue noise sampling algorithm is specifically:

1) According to the spatial location of the logging, the blue noise sampling algorithm is used to obtain a Poisson disk that satisfies the local spatial distribution, and only one log is allowed to be sampled inside each Poisson disk;

2) Use the kernel density estimation algorithm to evaluate the spatial distribution of the log in the Poisson disk, calculate the spatial distribution density value of the log, and then adaptively update the size of the Poisson disk radius to maintain the Spatial distribution characteristics of well logging.

First, w _i represents a log, Represents a sampled Poisson disk with _wi as the center and ri =r/f( _{wi )} as the radius, the parameter r represents the sampling rate defined by the user interaction, and the function f(w) represents the spatial distribution of logging. Estimate the calculated kernel density function, and the calculation process is as formula (1);

In formula (1), K _h represents a Gaussian kernel function with bandwidth h, and f( _wi ) represents the kernel density estimation value of logging _wi . In this method, it is stipulated that only one well is allowed to be selected as a standard well for each sampling disk; meanwhile, the following definitions are made: detection ring represents an annular area with _wi as the center, _{ri as the inner diameter and d i = 2ri} as the outer diameter, Q represents the manually selected sampling well queue, SP _Q represents all logging wells in the sampling tray corresponding to the sampling wells in the Q queue collection.

Second, mark all logs as "active". One of the wells is randomly selected as a sample and added to queue Q. During iterative sampling, the wells in SP _Q were changed to "inactive". As shown in Fig. 2(a), randomly select a sample well w ₀ from the queue Q to determine its sampling disk and detection ring As shown in Figure 2(b), from An "active" well w ₂ is randomly selected as a possible sampling well for detection, if covers any sampling wells marked in queue Q, as shown in Figure ₂ (c), the "active" well w2 will be changed to "inactive" and will be Re-select another "active" well w ₃ to continue detection until a log that meets the conditions is found, mark it as a sampling well and add it to the Q queue. If there are _no valid sample wells around w0, it is removed from Q. Repeat the sampling process until all logs become inactive.

The method for designing a multi-scale formation matching model based on a dynamic programming algorithm is specifically:

1) According to the multi-dimensional logging data, median filtering is used to smooth each logging attribute curve to obtain the corresponding smooth logging curve, and the multi-dimensional logging data is processed nonlinearly and normalized to 0 and 0 according to the discrete results. between 1;

2) According to the smooth logging curve, use the activity function to identify and divide the stratigraphic horizon boundary for logging;

3) The optimal horizon matching sequence between the two wells is solved by the dynamic programming algorithm, and the stratum thickness corresponding to the best matching sequence on each logging attribute is summed to obtain the quantitative attribute difference between the two wells.

The specific process of the method for designing the multi-scale formation matching model based on the dynamic programming algorithm is shown in Figure 3. The first step is to smooth the original logging attribute curve, and the second step is to use the activity function to smooth each line. The logging curve is used to identify the horizon, and the activity function equation is shown in formula (2):

where E _i represents the active value at depth i. y _j represents the logging attribute value in the depth range [iL, i+L], and L is the half-window length. If the active value at well depth i is greater than a given threshold, the formation is identified using the corresponding depth i as a boundary. The third step is to find out the optimal matching path between the target wells by using the two-well matching method based on dynamic programming according to the formation horizons of the target wells, so as to obtain the best matching sequence between the target wells. Assuming that a well A with m layers and a well B with n layers are given, the best matching path between them can be found according to the global consumption function, as shown in formula (3):

Among them, C(A _i , B _j ) represents the sum of the differences on the matching paths from (A ₁ , B ₁ ) to (A _i , B _j ); d(A _i , B _j ) is the two formations (well A The sum of all feature differences between layer i and well B (layer j), used to measure the similarity of A _i and B _j . The difference between the missing formations in the two wells is expressed as g(A _i ) or g(B _j ). The problem of minimum value of C(A _i , B _j ) is solved by dynamic programming method, and the optimal sequence of stratigraphic comparison is obtained. Finally, the formation thicknesses of the two target wells corresponding to the best matching sequence are summed to obtain the multi-dimensional attribute difference between the target wells.

(2) In the traditional geological interpretation process, standard wells are always selected manually according to the prior knowledge of domain experts, and the spatial distribution and multi-attribute correlation of well logging cannot be considered, which is prone to large errors and uncertainties. Affect subsequent interpretation of geological formations. In the method of the present invention, a discrete selection model is introduced, and a standard well supervision selection visualization framework based on the discrete selection model is proposed, which can consider the spatial distribution of well logging and the correlation of various attributes: construct utility based on user preference and utility maximization theory It sets and calculates the personal preference coefficient for multi-dimensional attribute data, and uses the discrete choice model to correlate the multi-dimensional attribute logging matching results with the local area distribution of logging.

Obtain the standard well sample set according to the prior knowledge of experts, use the discrete selection model to maximize the utility between the similarity degree of logging attributes described in step 1 and the standard well screening, and traverse the Poisson obtained by the adaptive blue noise sampling algorithm disk, screen standard wells according to the discrete selection model, and obtain standard well screening information.

The method of using the discrete selection model to maximize the utility between the similarity degree of the logging attributes described in step 1 and the standard well screening is specifically:

1) According to the Poisson disk that satisfies the local spatial distribution, use the prior knowledge of experts to obtain a standard well sample set that is manually marked;

2) Design a fixed utility function according to the multi-dimensional attribute similarity between the Poisson disk logs, use a multinomial regression model to simulate the random error distribution, and combine the two to obtain a total utility function, which is used to calculate the different logging selected by different users. The probability of being a standard well;

3) According to the standard well sample set, the discrete selection model is used to automatically calculate the corresponding parameters in the total utility function and traverse the Poisson disk, and then select standard wells whose attribute combination in the local space maximizes the user's utility.

First, in order to better ensure the spatial distribution uniformity of standard wells, each logging well is regarded as an alternative, and the logging set in each Poisson disk is regarded as a relatively independent standard well alternative. Invite domain experts to manually identify standard wells for some logging sets based on their prior knowledge, and obtain the actually selected standard well sample set.

Then, based on the discrete choice model, the utility function is constructed by using user preference and utility maximization theory, and the total utility formed by different users choosing different logging wells as standard wells is simulated and calculated, as shown in formula (4):

U _qi =V _qi +ε _qi (4)

Among them, U _qi represents the total utility that the selection of alternative “i” may bring to user “q”, V _qi represents the expected utility value of the alternative solution, and ε _qi represents the random residual, which is used to quantify the difference between the average utility value and the actual utility value. Bias in utility value.

Since the logging data is multidimensional attribute data, we assume that the fixed utility function V is a multivariate linear function, as shown in Equation 5:

V _qi = β ^T x _qi (5)

Among them, β=[β ₁ ,β ₂ ,...,β _n ] ^T represents the user's preference coefficient corresponding to n data attributes. x=x _qi1 , x _qi2 ,...,x _qin represents the total similarity degree of the alternatives on each attribute, which can be calculated according to step 2. Then, the multinomial Logit model is used to simulate the density distribution of the random residual ε, as shown in Equation 6:

Then, formula (5) and formula (6) are substituted into the total utility formula (4), and the probability of user “q” choosing plan “i” in a given plan set is simulated and calculated, that is, the utility of alternative plan “i” is greater than The probability of the utility of all other alternatives in a given set of alternatives is given by Equation 7:

Finally, by substituting the obtained standard well sample set into formula (7), the user preference coefficient "β" can be automatically calculated through regression and updated into the total utility function, helping users to select attribute combinations in the target data set to maximize their Utility standard well. Following the above process, a discrete selection model is used to maximize the utility between log attribute similarity and standard wells, enabling supervised standard well screening.

(3) According to the standard well screening information, guide the user to interactively specify or change standard wells through visual analysis technology, adaptively update the standard well sample set information, optimize the discrete selection model parameters, and then re-screen to meet the current needs of users and standard well of experience.

The method for guiding users to interactively specify or change standard wells through visual analysis technology is specifically:

1) Design a variety of visualization schemes to collaboratively display logging data information: in the main view of the system shown in Figure 4, the actual spatial position of logging is mapped to the map, and the visualization results of its spatial distribution characteristics are preserved; logging according to different attributes Data, design attribute curve graph, count the same attribute data of different logging in the local space in the specified attribute space, use tone mapping technology to distinguish the standard well from non-standard well; use variance to quantify the difference between the logging curves Statistically obtain the total attribute difference histogram of standard wells in the local space, and retain the contrast of attribute features;

2) According to the visualization results, design a standard well screening visualization scheme: count the feature matching results on each attribute of each local space to obtain a multi-dimensional attribute horizon matching correlation matrix; set a screening ring according to the horizon matching results, which is intuitive on the map Contrast and enhance the display of property differences between logs in local space;

3) According to the comparison of the attribute differences, support user interactive analysis and replacement of standard wells, and then update samples, optimize discrete selection model parameters according to the new sample set, recalculate and screen standard wells that meet user needs, the specific process is shown in Figure 6 shown.

Figure 4 shows the complete system interface diagram. Among them, Figure 4(a) is an information view, which displays basic data set information, such as the total number of well logs in the data set, the total number of standard wells currently screened, the number of the currently selected standard wells, and the number of currently divided local areas; Figure 4 (b) is the sampling panel, which is used to adjust the sampling rate and determine the sampling wells; Figure 4(c) is the selection optimization panel, which provides the update function of the sample set, and calculates and displays the current user attribute preference coefficients; Figure 4(d) is a geographic view, and logging is regarded as a dot, which is projected on the map according to its latitude and longitude values, and the spatial distribution of logging is visually displayed on the map; Figure 4(e) is a multi-dimensional attribute view, which is described differently according to the original logging data. Curve changes of logging attributes; Fig. 4(f) is a schematic diagram of formation matching, which visually shows the formation matching between any two wells; Fig. 4(g) is a view of formation correlation matrix, which is used to present the relationship between multiple wells in a local area. Stratigraphic matching results, each matrix represents a logging attribute, and each row or column represents a log in the selected local area, and the color mapping technique is used in the matrix cells to represent the attribute similarity between a pair of logs , the cells on the diagonal from the upper left corner to the lower right corner represent the total similarity between each log and other wells in the local range; Fig. 4(h) is the histogram of the total attribute difference of standard wells, through The variance of each attribute on the logging curve is calculated, and different tone mapping schemes are used to display the total attribute difference between standard wells and non-standard wells.

Figure 5 shows the property evaluation results of the selected standard wells at a given sampling rate, which can help users evaluate the accuracy of the selected standard wells in the method described in this paper. Among them, Fig. 5(a) shows the standard well set obtained by formula (1) under the current user preference coefficient after the user determines the sampling rate according to their own needs, and the projection result on the map remains the same as the original logging spatial distribution. It shows that the standard wells selected by the method of this paper can satisfy the local spatial distribution characteristics of well logging; Fig. 5(b) is the plotting result of the logging attributes of standard wells and non-standard wells in a given local space. Due to the horizon shift caused by complex geological conditions, it is difficult to fit well logging properties with similar geographical locations; Fig. 5(c) shows the multi-dimensional logging properties of standard wells and non-standard wells in a given local space. Difference performance results, each fan-shaped area represents a logging attribute, and each layer of rings represents a non-standard well. Through the distribution of the annular histogram, the similarity between standard wells and non-standard wells can be enhanced; Figure 5 (d) is the sum of the differences between standard wells and non-standard wells in each attribute dimension in a given local space. It is obtained by calculating the variance of logging attribute data, and the similarity of standard wells in different logging attributes can be quickly perceived through the histogram. Figure 5(e) is a display of the user preference coefficient results calculated by the discrete selection model designed by the method of the present invention shown in formulas (4)-(7), and different color mapping schemes are used to enhance the positive and negative values of the display coefficients. properties, to quickly understand the current user's preference for different attributes; Figure 5(f) is the actual formation matching result calculated by the formation matching model designed by the method of the present invention shown in formula (2) and formula (3). In a given local area, the relationship between any two wells is displayed in the form of a matrix diagram, and the actual stratigraphic correlation degree between the two wells is presented by using tone mapping technology, which can display the multi-dimensional logging in a given local space in detail and intuitively. similarity in properties.

Figure 6 shows the comparison of multi-dimensional attribute differences between standard wells and non-standard wells before and after updating the user preference coefficient, that is, it shows the attribute difference between standard wells and non-standard wells in a given local space before and after updating the user preference coefficient. And the preference performance of standard wells on multi-dimensional logging attributes. Through the visual analysis system shown in Figure 4, users can obtain the standard well screening information shown in Figure 5, so that they can interactively change the selection of standard wells according to their own needs, and incorporate them into standard wells as new samples sample set. Then, by clicking the "Update" button in the selection panel, re-correct and generate a new user preference coefficient to meet the changes of the user's real-time application requirements. As shown in Fig. 6(a), the user originally paid more attention to the attributes COND and AC, and less attention to the attribute SP, and later paid great attention to the attribute SP according to the change of their own needs. After the user interactively changes the sample set, the new preference coefficient calculated as shown in Figure 6(c) is in line with the user's demand preference, indicating that the method of the present invention can effectively reflect the user's real-time demand preference for logging attributes, and the standard is updated accordingly. Well screening results. At the same time, Fig. 6(b) and Fig. 6(d) show the variation of the standard well selection results provided by the method of the present invention in the formation correlation matrix. It is not difficult to find that the similarity performance of the selected standard wells in each attribute It shows strong consistency with the positive and negative preferences of users on each attribute. 6, it can be seen that the result image obtained by the method of the present invention can intuitively and quickly realize the difference analysis between standard wells and non-standard wells, and support users to update or update the standard well sample set through visual analysis technology. Optimized to quickly acquire a new set of standard wells, efficiently meet real-time changes in user application requirements, and effectively improve the efficiency and accuracy of standard well selection;

Compared with the traditional visual process of manual selection of standard wells, the biggest advantage of the method of the present invention is that it proposes a visual analysis method for standard well screening based on discrete selection model, which comprehensively considers the spatial distribution characteristics of well logging data, multi-dimensional attribute correlation and domain experts. By building a convenient multi-functional visual analysis system to realize supervised standard well screening, the efficiency and accuracy of standard well selection are effectively improved.

Claims (5)

1. a standard well screening visual analysis method based on discrete selection model, is characterized in that, specifically comprises the steps: (1) Based on the multi-dimensional logging data, the spatial distribution characteristics of the logging and the stratigraphic correlation of the multi-dimensional logging attributes were extracted respectively: on the premise of determining the standard well screening ratio, the adaptive blue noise sampling algorithm was used to obtain the spatial distribution of the wells. At the same time, a multi-scale formation matching model based on dynamic programming algorithm is designed to quantitatively measure the similarity of logging attributes from the perspective of different attributes. (2) Obtain the standard well sample set according to the prior knowledge of experts, use the discrete selection model to maximize the utility between the similarity degree of logging attributes described in step 1 and the standard well screening, and traverse the adaptive blue noise sampling algorithm to obtain of Poisson disk, standard wells are screened according to the discrete selection model, and standard well screening information is obtained. (3) According to the standard well screening information, guide the user to interactively designate or change standard wells through visual analysis technology, adaptively update the standard well sample set information, optimize discrete selection model parameters, and re-screen to meet the current needs of users. and standard well of experience. 2. the standard well screening visual analysis method according to claim 1, is characterized in that, in step (1), described utilizing adaptive blue noise sampling algorithm to obtain the method that satisfies the Poisson disk of local spatial distribution, is specifically: 1) According to the spatial location of the logging, the blue noise sampling algorithm is used to obtain a Poisson disk that satisfies the local spatial distribution, and only one log is allowed to be sampled inside each Poisson disk; 2) Use the kernel density estimation algorithm to evaluate the spatial distribution of the log in the Poisson disk, calculate the spatial distribution density value of the log, and then adaptively update the size of the Poisson disk radius to maintain the Spatial distribution characteristics of well logging. 3. The standard well screening visual analysis method according to claim 1, wherein in step (1), the method for designing a multi-scale formation matching model based on a dynamic programming algorithm is specifically: 1) According to the multi-dimensional logging data, use median filtering to smooth each logging attribute curve, nonlinearly process the multi-dimensional logging data, and normalize between 0 and 1 according to the discrete results; 2) According to the smooth logging curve, use the activity function to identify and divide the stratigraphic horizon boundary for logging; 3) The optimal horizon matching sequence between the two wells is solved by the dynamic programming algorithm, and the stratum thickness corresponding to the best matching sequence on each logging attribute is summed to obtain the quantitative attribute difference between the two wells. 4. The visual analysis method for standard well screening according to claim 1, wherein in step (2), the use of a discrete selection model to maximize the similarity between the logging attributes described in step 1 and the standard well screening is performed. The method of utility between them is as follows: 1) According to the Poisson disk that satisfies the local spatial distribution, use the prior knowledge of experts to obtain a standard well sample set that is manually marked; 2) Design a fixed utility function according to the multi-dimensional attribute similarity between the Poisson disk logs, use a multinomial regression model to simulate the random error distribution, and combine the two to obtain a total utility function, which is used to calculate the different logging selected by different users. The probability of being a standard well; 3) According to the standard well sample set, the discrete selection model is used to automatically calculate the corresponding parameters in the total utility function and traverse the Poisson disk obtained by the adaptive blue noise sampling algorithm, and then select the attribute combination in the local space. Standard wells to maximize user utility. 5. The visual analysis method for standard well screening according to claim 1, wherein the method for guiding a user to interactively designate or change standard wells through the visual analysis technology described in step (3) is specifically: 1) Design a variety of visualization schemes to collaboratively display logging data information: map the actual spatial location of logging to the map, and retain the visualization results of its spatial distribution characteristics; The same attribute data of different logs in the interior are counted in the specified attribute space, and the standard wells and non-standard wells are distinguished by tone mapping technology; the difference between the logging curves is quantified by variance, and the standard wells in the local space are obtained by statistics. The total attribute difference histogram of , preserving the contrast of attribute features. 2) According to the visualization results, design a standard well screening visualization scheme, count the feature matching results on each attribute of each local space, and obtain a multi-dimensional attribute horizon matching correlation matrix; set a screening ring according to the horizon matching results, which is intuitive on the map Contrast and enhance the display of property differences between logs in a local space. 3) According to the comparison of the attribute differences, support user interactive analysis and replacement of standard wells, and then update samples, optimize discrete selection model parameters according to the new sample set, and recalculate and screen standard wells that meet user needs.