A reservoir prediction method based on differential resonance

Abstract

The present invention relates to a reservoir prediction method based on differential resonance, and belongs to the technical field of petroleum exploration and development. The reservoir prediction method based on differential resonance of the present invention comprises the following steps: 1) smoothing and coarsening the wave impedance logging curves of known wells in the work area; 2) then, based on the wave impedance logging curves after coarsening and smoothing of the known wells, selecting a logging curve segment with differential resonance near the target layer; 3) determining the seismic layer and time window corresponding to the logging curve resonance segment based on the preferred logging curve segment that has differential resonance with the reservoir parameters; 4) extracting seismic waveform clustering attributes based on the Manhattan distance of the 90-degree phase seismic waveform according to the determined seismic layer and time window, and then combining the reservoir parameters of the known wells to perform interpolation prediction of the reservoir parameters of the target reservoir by mixing the weights of three differential factors. When the reservoir prediction method based on differential resonance of the present invention performs prediction, the prediction result has high certainty and reliability.

Description

Technical Field

[0001] The invention relates to a reservoir prediction method based on differential resonance, belonging to the technical field of petroleum exploration and development. Background Art

[0002] Previous researchers have continuously gained new insights into reservoir distribution through the application of drilling data, seismic attributes, seismic forward modeling, reservoir inversion, and other technologies, developing a diverse range of technical approaches. Currently, the mainstream research direction in reservoir prediction is to conduct research on increasingly thin and fine-scale reservoirs. By integrating information from different frequency bands of seismic and drilling data, the vertical resolution accuracy of thin interbedded reservoirs has been improved, resulting in a wealth of research results.

[0003] There are many examples of prior art using well-seismic combined prediction of reservoir distribution. For example, Chinese patent document CN110945385A discloses a method for identifying strata from seismic and well data using a formation knowledge base. This method proposes using machine learning to identify strata based on the formation knowledge base of seismic data and well logging data. Chinese patent document CN106291701A discloses a reservoir detection method that improves the reservoir detection accuracy of seismic exploration data by extracting weak seismic response characteristics of the reservoir through the establishment of a deep learning model for reservoir feature detection. Chinese patent document CN108254785A discloses a reservoir identification method. This method obtains multiple seismic component data by waveform decomposition of seismic data. Based on the reservoir reflection characteristics, the method uses reconstructed well logging curves to filter the multiple seismic component data to obtain seismic component data representing the reservoir, and then determines the reservoir based on this seismic component data. Chinese patent document CN108957532A discloses a reservoir prediction method. This method proposes phase-controlled modeling based on seismic data and then determines whether a suitable reservoir prediction result can be obtained based on the seismic records corresponding to the initial impedance model. Chinese patent document CN105388525A discloses a reservoir prediction method, which proposes obtaining three reconstructed data volumes through seismic attribute reconstruction and predicting reservoir types by performing color fusion processing on the three reconstructed data volumes.

[0004] Although existing reservoir prediction technologies have been explored and studied from multiple angles and fields, and new technological progress has been continuously made, it is still difficult to meet the prediction needs of thin reservoirs in fine exploration and development.

[0005] On the one hand, existing technologies often use seismic slice attributes to predict the distribution of thin reservoirs, attempting to slice out the distribution of each thin reservoir layer by layer from a single seismic phase axis. When the number of thin reservoirs within a single seismic phase axis is small, the ambiguity of the seismic slice attributes in reflecting the distribution of a single thin reservoir is relatively small. However, if a single seismic phase axis contains a large number of thin reservoirs, the seismic slices reflect the combined effect of multiple thin reservoirs, making it difficult to reasonably predict the distribution of each thin reservoir.

[0006] On the other hand, in recent years, the inversion prediction method of thin interbedded reservoirs controlled by seismic waveform phase has been widely studied and applied. It attempts to use the high-frequency information of the logging curve, under the guidance of the seismic waveform phase, and adopt the waveform phase-controlled interpolation method within the seismic phase axis to depict the spatial three-dimensional distribution of each thin reservoir. However, since the thickness of thin reservoirs is often too thin (can be as thin as 1-5m), the thickness converted to the time domain is only about 0.5-3ms. The seismic frequency band is limited, and the vertical width of a seismic phase axis is usually around 10-30ms. Therefore, the thickness of thin reservoirs is usually much smaller than the vertical width of the seismic phase axis. In addition, seismic data are subject to complex influences such as signal-to-noise ratio and fidelity. As a result, the vertical accuracy of time-depth calibration of seismic synthetic records that relies on waveform similarity comparison is difficult to fully meet the needs of fine vertical calibration of thin interbedded reservoirs with a thickness of 1-3ms. The vertical calibration error of synthetic seismic records greatly affects the inter-well phase-controlled interpolation link in the waveform phase-controlled inversion process, resulting in reduced accuracy of the inversion results, which is difficult to fully meet the needs of fine prediction of thin interbedded reservoirs. In addition, in the three-dimensional phase-controlled inversion process, when performing phase-controlled prediction for a thin reservoir, the time window of the seismic waveform is usually a vertically symmetric time window centered on the thin layer. However, in fact, although the distribution law of the thin reservoir may have a certain degree of correlation with the seismic waveform difference within the vertical central symmetric time window, the micro-trend of the thin reservoir is not completely consistent with the macro-trend of the seismic waveform. From the perspective of sedimentary laws, the micro-distribution trend of the thin reservoir may be correlated with the seismic waveform within the central symmetric time window, or it may have a stronger correlation with the seismic waveform in the upper or lower deviated time window. Therefore, using only the waveform information within the vertically symmetric time window centered on the thin layer for thin reservoir prediction has certain accuracy limitations and sometimes cannot achieve the best reservoir prediction effect.

[0007] In summary, due to the limitations of the vertical resolution capability of seismic data, conducting thin reservoir prediction research under conditions that exceed the vertical resolution accuracy of seismic data often faces technical problems such as high uncertainty and low reliability of thin reservoir prediction results. Therefore, it is necessary to further deepen the research on thin reservoir prediction methods. Summary of the Invention

[0008] The purpose of the present invention is to provide a reservoir prediction method based on differential resonance, which can improve the accuracy of thin reservoir prediction when applied to thin reservoir prediction.

[0009] In order to achieve the above objectives, the technical solution adopted by the present invention is:

[0010] A reservoir prediction method based on differential resonance includes the following steps:

[0011] 1) Smoothing and coarsening the wave impedance logging curves of known wells in the work area;

[0012] 2) Then, based on the roughened and smoothed wave impedance logging curve of the known well, a logging curve segment that resonates differentially with the reservoir parameters is selected near the target layer;

[0013] 3) According to the preferred differential resonance logging curve segment, determine the seismic layer and time window corresponding to the logging curve resonance segment;

[0014] 4) Based on the seismic horizon and time window determined in step 3), the seismic waveform clustering attributes are extracted based on the Manhattan distance of the -90-degree phase seismic waveform. Then, combined with the reservoir information of known wells, the reservoir parameters of the target reservoir are interpolated and predicted by mixing the weights of the three difference factors;

[0015] In step 2), the method of selecting a well logging curve segment near the target layer that has a differential resonance with the reservoir parameters includes the following steps:

[0016] Taking the vertical depth of the target layer at the center of each known well as the reference depth, a series of depth windows are constructed by setting the vertical width value of the depth window and the upper and lower offset depth values ​​near the reference depth. Then, within each depth window, the Manhattan distance between the roughened well log curves of each two known wells is calculated; the absolute value of the reservoir parameter difference between each two known wells is calculated;

[0017] For each depth window, a crossplot is drawn with the absolute value of the reservoir parameter difference between wells as the horizontal axis and the Manhattan distance of the well logging curve between wells as the vertical axis. By fitting the scattered points in the crossplot with a linear formula, the slope of the Manhattan distance changing with the reservoir parameter difference between wells is obtained.

[0018] Compare the slope values ​​of all depth windows, and select the logging curve segment corresponding to the depth window with the largest slope change value as the preferred differential resonance logging curve segment;

[0019] In step 4), the three difference factors are plane distance, waveform difference, and waveform cumulative difference; in the interpolation prediction process, for each point to be predicted, the weights of the three difference factors of plane distance, waveform difference, and waveform cumulative difference are calculated respectively, and each difference factor is normalized to obtain the interpolation weight value of each known well participating in the prediction for each difference factor. Then, combined with geological knowledge, a weighted blend is performed to obtain the comprehensive weight of each known well participating in the prediction for the corresponding point to be predicted. Then, based on the comprehensive weight, the reservoir parameters of the corresponding prediction point are interpolated and predicted;

[0020] The plane distance is the distance between the point to be predicted and the known well points involved in the prediction, the waveform difference is the difference in waveform clustering results between the point to be predicted and the known well points involved in the prediction, and the waveform cumulative difference is the cumulative sum of the absolute values ​​of the differences in class values ​​of each two adjacent clustering points spanned by the line segment connecting the point to be predicted and the known well points involved in the prediction.

[0021] The reservoir prediction method based on differential resonance of the present invention can, when predicting thin reservoirs, start from the geological information of a single thin reservoir that has been drilled, extract seismic attribute information based on seismic horizons, and perform planar prediction of reservoir parameters rather than three-dimensional prediction. This method can avoid the adverse effects of insufficient time-depth calibration accuracy of thin reservoirs on reservoir parameter prediction. In addition, when interpolating and predicting reservoir parameters, the three factors of planar distance, seismic waveform difference, and cumulative seismic waveform difference are used together to participate in weighted constraints, which can further improve the certainty and reliability of thin reservoir prediction results.

[0022] It is understandable that waveform cluster attribute maps are typically composed of discrete data points with a certain plane grid density, resulting in the presence of plane grid elements. The line segments connecting the predicted points and the known well points involved in the prediction will pass through a series of grid elements in the waveform cluster attribute plane map, where a series of "adjacent cluster points" will exist. The central vertical depth refers to the vertical depth of the reservoir thickness center point of each well, which has different depth values ​​for different wells. The upper and lower offsets of the depth window are based on this vertical depth; that is, for each well, the vertical depth offset at the depth of the reservoir thickness center point is 0.

[0023] Furthermore, the reservoir parameters include reservoir thickness, reservoir velocity, reservoir density, reservoir wave impedance, reservoir permeability, or reservoir porosity. In other words, the differential resonance-based reservoir prediction method of the present invention can predict reservoir parameters such as reservoir thickness, reservoir velocity, reservoir density, reservoir wave impedance, reservoir permeability, or reservoir porosity.

[0024] Furthermore, before smoothing and coarsening the wave impedance logging curves of the known wells in the work area, the logging data of the work area is first obtained to calculate the target layer depths and reservoir parameters of the known wells.

[0025] Furthermore, in step 1), the smoothing and coarsening process includes the following steps: obtaining -90-degree phase seismic data of the work area, calculating the vertical two-way time width of the seismic phase axis near the target layer, drawing a probability distribution diagram, and determining the peak vertical two-way time corresponding to the peak position point of the vertical two-way time width distribution; then using the peak vertical two-way time to perform time-to-depth conversion to obtain the corresponding depth domain vertical thickness, and then performing mean smoothing processing on the logging wave impedance curve with a depth window of 0.2 to 0.3 times the depth domain vertical thickness.

[0026] Furthermore, in step 2), when calculating the Manhattan distance between the smoothed and roughened logging curves of each two known wells, the Manhattan distance formula is as follows:

[0027]

[0028] M is the Manhattan distance between the smoothed and coarsened well logging curves of two known wells, A is the value of the smoothed and coarsened well logging curve of a well at the i-th sampling point, B is the value of the smoothed and coarsened well logging curve of the other well at the i-th sampling point, N is the number of vertical depth sampling points in the well logging curve layer within the depth window, and i is a natural number of vertical depth sampling points traversed within the window.

[0029] Further, in step 3), according to the preferred differential resonance logging curve segment, the time-depth calibration and -90 degree phase seismic horizon are analyzed to determine the seismic horizon and time window corresponding to the logging curve resonance segment. Specifically, when determining the seismic horizon and time window corresponding to the logging curve resonance segment, the width of the depth window with the largest slope change value and the vertical offset of the depth in the window relative to the reference depth are converted into two-way time, and the average velocity of the formation near the target layer segment is used when converting the two-way time; then, on the -90 degree phase seismic data, the seismic profile is flattened along the seismic horizon closest to the target reservoir calibration position, and the average timeline position of the vertical center point of the target sand body of each known well after time-depth calibration is approximated to the calibration position of the reference depth, and the time offset between the calibration position and the seismic interpretation horizon is observed; based on the calibration value time offset of the depth window center of the preferred differential resonance logging curve resonance segment relative to the reference depth, and the time offset of the calibration position of the reference depth relative to the seismic interpretation horizon, the seismic time window corresponding to the logging curve resonance segment is determined.

[0030] It can be understood that the "stratum near the target layer section" refers to the stratum within a certain vertical depth range including the target layer. The stratum thickness range must at least include the target layer and the upper and lower adjacent surrounding rocks. Depending on the stability of the stratum velocity, the thickness of the upper and lower strata can also be appropriately increased to obtain a relatively stable and reliable average stratum velocity; the obtained "average stratum velocity" can be understood as an approximation to the average stratum velocity of the target layer and the upper and lower surrounding rocks.

[0031] Furthermore, the normalized calculation method for the planar distance factor first calculates the distance span between the maximum and minimum planar distances between the predicted point and each participating well. The planar distance between the participating wells and the predicted point is then divided by the distance span. The resulting value is squared and then reciprocated (in special cases, consideration must be given to avoiding a 0 denominator for the reciprocal, which can be approximated by using a very small positive number in the denominator). This yields the interpolation weight for each participating well. The total interpolation weight is then summed for each participating well to obtain the total interpolation weight. The interpolation weight for each participating well is then divided by the total interpolation weight to obtain the normalized interpolation weight for the planar distance factor. The weight normalization calculation method for the waveform difference factor and the waveform cumulative difference factor is similar to the above-mentioned planar distance weight normalization method. The planar distance value is replaced with the waveform difference value and the waveform cumulative difference value derived from the waveform clustering results, respectively.

[0032] Furthermore, based on the interpreted seismic horizons, the waveform clustering attributes based on the Manhattan distance are extracted from the seismic data using the optimized time window. The formula for calculating the Manhattan distance of the seismic waveform within the time window is as follows:

[0033]

[0034] The meanings of the parameters in the above formula are as follows:

[0035] M is the Manhattan distance of the earthquake waveform, A i is the value of the waveform of a seismic trace at the i-th sampling point, B i is the value of the waveform of another seismic channel at the i-th sampling point, N is the total number of time sampling points of the seismic waveform in the time window, and i is a natural number of time sampling points in the traversal window.

[0036] Furthermore, in step 4), according to the interpreted seismic horizons, based on the Manhattan distance of the seismic waveforms in the time window, an unsupervised K-Means clustering algorithm is used to cluster the seismic waveforms in the time window optimally obtained in the study area. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of a reservoir prediction method based on differential resonance in an embodiment of the present invention;

[0038] Figure 2 is a planar distribution map of the work area boundary and known wells in an embodiment of the present invention;

[0039] Figure 3 is a probability distribution diagram of the vertical width of the earthquake event in an embodiment of the present invention;

[0040] Figure 4 This is a formation velocity analysis diagram in an embodiment of the present invention;

[0041] Figure 5 3. This is a comparison diagram of the wave impedance logging curve before and after coarsening processing in an embodiment of the present invention;

[0042] Figure 6 is a schematic diagram of the resonance gradient of the Manhattan distance difference relative to the reservoir thickness difference in an embodiment of the present invention;

[0043] Figure 7 Schematic diagram of a preferred differential resonance logging curve segment in an embodiment of the present invention;

[0044] Figure 8 Schematic diagram of analyzing sensitive earthquake layers and timely windows in an embodiment of the present invention;

[0045] Figure 9 is a seismic waveform clustering attribute graph extracted in an embodiment of the present invention;

[0046] Figure 10 is a reservoir thickness distribution map predicted by a three-factor weighted mixed prediction in an embodiment of the present invention;

[0047] Figure 11 This is a reservoir thickness distribution map predicted using only plane distance weights in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention is further described below in conjunction with a certain research area.

[0049] The study area is approximately 26 km 2 , belonging to the fan delta sand body depositional system; the overall development pattern of the sand body is that it advances from the southeast provenance area to the northwest and gradually tapers out, and the reservoir-cap configuration is favorable; the northwest-trending pinch-out tongue-shaped sand body and the northeast-trending structural line are configured to form an up-dip pinch-out lithologic trap; the oil source conditions in this area are excellent, and the drilled wells have good oil and gas shows, making it a favorable oil and gas exploration area.

[0050] Eleven wells have been drilled in the study area (see Figure 2Based on reservoir data from existing wells, the target sand body thickness ranges from approximately 1 to 5 meters. Three wells are located in the target sand body pinch-out zone (sand body thickness is 0 meters). Two wells have poor sand body properties, with well logging interpretations indicating dry layers. The technical challenge in this area is how to use a combined well and seismic analysis method to accurately and reliably predict the distribution of the target sand body, providing a basis for identifying lithologic traps.

[0051] Example

[0052] The reservoir prediction method based on differential resonance in this embodiment has a flow chart as shown in FIG. Figure 1 As shown, the following steps are included:

[0053] 1) Obtain logging data from the work area and calculate target layer depths and reservoir parameters of known wells.

[0054] This embodiment takes the prediction of reservoir thickness parameters as an example, collects and obtains logging data of the work area, and counts the target layer depth, reservoir thickness parameters, and reservoir interpretation information of known wells. The data are shown in Table 1 below.

[0055] Table 1 Statistical results

[0056]

[0057]

[0058] 2) Obtain -90-degree phase seismic data from the work area, calculate the vertical two-way time width of the seismic event, and use this data to smooth and coarsen the wave impedance logging curve. Specifically:

[0059] Obtain the -90-degree phase seismic data of the work area, calculate the vertical two-way time width of the seismic phase axis near the target layer, and draw a probability distribution diagram, as shown in the attached figure. Figure 3 As shown in the figure, it can be seen that the two-way time width corresponding to the peak value (also approximately the average value) of the vertical two-way time width distribution of the seismic phase axis is about 16.8ms; according to the well logging acoustic wave curve, the average formation velocity near the target layer in the work area is about 4200m / s (see Appendix Figure 4 The time-depth conversion of 16.8 ms using this velocity (the formula is 4200×16.8÷1000÷2) corresponds to a vertical thickness of about 35 m in the depth domain. Representing a seismic event waveform requires approximately four control points, and 35 m divided by 4 is approximately 9 m. Therefore, the logging impedance curve was smoothed and coarsened with a depth window of 9 m, making the vertical variation scale of the well curve roughly similar to that of the seismic waveform (see Appendix). Figure 5 ).

[0060] 3) Based on the smoothed and coarsened wave impedance logging curve, the gradient analysis method of the Manhattan distance of the well curve relative to the plane distance between wells is used to select the logging curve layer segment that has differential resonance with the reservoir parameters near the target layer.

[0061] Manhattan distance is an effective statistical measure of similarity / difference. Manhattan distance uses two waveforms of equal length and N samples within the waveforms and calculates the absolute value of the sample difference for all corresponding samples. Therefore, Manhattan distance is given by the following formula:

[0062]

[0063] In the process of optimizing the differential resonance logging curve segment in this embodiment, the meanings of the parameters in the above formula are as follows:

[0064] M is the Manhattan distance between the logging curves of two known wells, A is the logging curve of one well, B is the logging curve of another well, N is the number of vertical depth sampling points in the logging curve layer within the depth window, and i is a natural number of vertical depth sampling points traversed within the window.

[0065] Taking the central vertical depth of the target layer (i.e., the sum of the reservoir top surface depth and the bottom surface depth divided by 2) at the depth of each well as the reference depth, a series of depth windows can be constructed by setting the vertical width value and the upper and lower offset depth values ​​of the depth window near the reference depth; then, within each depth window, the Manhattan distance value between the logging curves of each two wells can be calculated using the above Manhattan distance formula. In addition, the absolute value of the reservoir thickness difference between each two wells can also be calculated.

[0066] For each depth window, the absolute value of the reservoir thickness difference between the wells is used as the horizontal coordinate, and the Manhattan distance of the well logging curve between the wells is used as the vertical coordinate to draw the intersection diagram (see Appendix). Figure 6 ), by fitting the scattered points in the crossplot with a linear formula, we can obtain the slope of the Manhattan distance changing with the reservoir thickness difference between wells (or called the gradient of change); when the gradient of the Manhattan distance of the logging curve changing with the reservoir thickness difference is large, it means that the resonance between the logging curve difference information and the reservoir difference information within the depth window is relatively strong, so that the logging curve layer segment with strong difference resonance can be selected near the target layer. Figure 7 As shown in FIG, by plotting the change gradient values ​​obtained in different depth windows, the depth window number corresponding to the highest point of the change gradient value can be selected; in this embodiment, the point with the highest change gradient value is located on the first vertical dotted line from the left. According to the window number query, it is known that the corresponding depth window width is 2m, and the vertical offset of the middle depth in the window relative to the middle depth of the target sand body is positive 10m (i.e., 10m downward). The logging curve segment corresponding to this window is the preferred differential resonance logging curve segment.

[0067] 4) According to the preferred differential resonance logging curve segment, analyze the time-depth calibration and -90-degree phase seismic horizon to determine the seismic horizon and time window corresponding to the logging curve resonance segment.

[0068] The average formation velocity near the target layer is about 4200 m / s. Based on this, the depth window selected in the previous step is converted. The selected depth window width (2 m) is converted to a two-way time of about 1 ms. The vertical offset depth value of the deep in the selected depth window relative to the middle depth of the sand body (positive 10 m) is converted to a two-way time of positive 5 ms. That is, the deep in the window needs to be deviated downward by 5 ms relative to the middle depth of the target sand body.

[0069] Based on the -90-degree phase seismic data, a seismic layer closest to the target reservoir calibration position is selected to perform layer flattening on the seismic profile; on the layer flattening profile of the -90-degree phase seismic, the relative position of the time-depth calibration result of the target reservoir and the seismic interpretation layer is observed and analyzed (see Appendix). Figure 8 ), find the average line of the vertical center point of the target sand body of each known well ( Figure 8 The dotted horizontal line indicates that it is located roughly at the trough earthquake layer on the seismic profile ( Figure 8 The position about 2ms above the solid horizontal line).

[0070] According to the above information, if a time window with a two-way time width of 1 ms and a time window center offset of 3 ms relative to the interpreted layer (i.e., 3 ms downward) is used based on the interpreted trough seismic layer, it can better correspond to the well depth section of the differential resonance. This time window is the seismic time window optimized based on scale differential resonance.

[0071] 5) According to the preferred seismic horizon and time window, waveform clustering attributes are extracted based on the Manhattan distance of the -90-degree phase seismic waveform.

[0072] The phase of seismic data used in seismic tectonic interpretation is usually 0 degrees. According to the principle of seismic reflection, the seismic waveform reflects the "stratum reflection interface" and has no direct correspondence with the "stratum segment". When interpreting the stratum distribution, it is necessary to go through the indirect analysis process of the "stratum reflection interface", which is very complicated. At the same time, due to the superposition of the reflection waves of the top and bottom interfaces of the "stratum segment", the vertical distribution range of the reflection waveform is relatively wide, which affects the vertical resolution accuracy of the stratum and is not conducive to the fine characterization of thin interbedded strata.

[0073] By selecting -90-degree phase seismic data, the seismic waveform of the formation becomes symmetrically distributed with the formation as the center, which can establish a direct correspondence between the seismic data and the "stratum segment", and is conducive to the direct corresponding interpretation of the "stratum segment" using seismic data; at the same time, after the phase shift processing process, the longitudinal waveform width of the seismic waveform of the "stratum segment" can be narrowed, which is conducive to enhancing the longitudinal resolution accuracy of thin interbeds.

[0074] According to the interpreted seismic horizons, waveform clustering attributes based on Manhattan distance are extracted from seismic data using the optimized time window.

[0075] The formula for calculating the Manhattan distance of earthquake waveforms within the time window is as follows:

[0076]

[0077] In the process of calculating the Manhattan distance of the earthquake waveform within the time window of this embodiment, the meanings of the parameters in the above formula are as follows:

[0078] M is the Manhattan distance of the earthquake waveform, A i is the value of the waveform of a seismic trace at the i-th sampling point, B i is the value of the waveform of another seismic channel at the i-th sampling point, N is the total number of time sampling points of the seismic waveform in the time window, and i is a natural number of time sampling points in the traversal window.

[0079] According to the interpreted seismic horizons, based on the Manhattan distance of the seismic waveforms in the time window, the unsupervised K-Means clustering algorithm is used to cluster the seismic waveforms in the time window obtained by the optimal selection of the study area. The output classes are roughly sorted by similarity to obtain the seismic waveform clustering attribute map, as shown in the attached figure. Figure 9 shown.

[0080] 6) Based on the extracted seismic waveform clustering attributes and combined with the reservoir thickness information of known wells, the reservoir thickness of the target reservoir is interpolated and predicted:

[0081] a) When using known well reservoir thicknesses to interpolate and predict reservoir thickness at unknown points, the weights of the three difference factors (plane distance, waveform difference, and waveform cumulative difference) are calculated and normalized separately.

[0082] The plane distance in this step refers to the plane distance between the predicted point and the known well points, which can be calculated based on the horizontal and vertical coordinates of the two points. According to geological sedimentary laws, if the plane distance between two points is close, the stratigraphic sedimentary results at the two points will be more likely to be similar. Therefore, the plane distance factor can affect the weight of the known well points in the interpolation prediction process.

[0083] The waveform difference in this step refers to the difference between the seismic waveform clustering results of the predicted point and the known well points. The absolute value of the waveform clustering values ​​subtracted from each other is taken to represent the degree of difference in seismic facies between the two points. Based on the understanding of seismic facies, if the seismic waveform difference between two points is small, then the sedimentary formations at the two points are more likely to be similar. Therefore, the waveform difference factor can influence the weight of the known well points in the interpolation prediction process.

[0084] The cumulative waveform difference in this step refers to the cumulative sum of the absolute differences between the cluster values ​​of each pair of adjacent cluster points spanned by the line segment connecting the predicted point and the known well points. Its significance is to quantitatively characterize the complexity of the seismic phase changes between two points on a plane. According to geological sedimentary laws, if the complexity of the seismic phase changes between two points is low, the probability that the two points belong to similar sedimentary systems will increase, and the stratigraphic sedimentary characteristics of the two points will have a greater probability of similarity. Therefore, the cumulative waveform difference factor can influence the weight of the known well points in the interpolation prediction process.

[0085] The above analysis shows that when interpolating known well information to predict unknown point parameters, three factors—planar distance, waveform difference, and cumulative waveform difference—all influence the weighting of known wells. The impact of these three factors varies depending on the geological sedimentary systems and specific characteristics of the wellbore areas. Therefore, the influence of these three factors in assigning sample point weights needs to be optimized based on the specific geological conditions. To facilitate adjustment of the weighting of the three factors, it is necessary to first normalize the weights of each factor.

[0086] The normalized calculation method for the planar distance factor is used as an example to illustrate this. First, the distance span between the maximum and minimum planar distances between the predicted point and each participating well is calculated. The planar distance between the participating wells and the predicted point is then divided by the distance span. The resulting value is squared and then reciprocated. (In special cases, to avoid a denominator of 0, this can be approximated by using a very small positive number in the denominator.) This yields the interpolation weight for each participating well. The total interpolation weight is then summed for each participating well to obtain the total interpolation weight. The normalized interpolation weight for the planar distance factor is then divided by the total interpolation weight for each participating well. This calculation method limits the interpolation weight for each participating well to a value between 0 and 1, and the sum of the interpolation weights for all participating wells is 1.

[0087] The weight normalization calculation method for waveform difference factors and waveform cumulative difference factors is similar to the weight normalization calculation method for the above-mentioned plane distance. It is only necessary to replace the plane distance value with the waveform difference value and waveform cumulative difference value derived from the waveform clustering results respectively.

[0088] After normalizing the interpolation weights of the three factors respectively, the limit range of the interpolation weight values ​​of the known wells involved in the prediction is 0-1, and the sum of the interpolation weight values ​​of all the known wells involved in the prediction of each factor is 1.

[0089] b) Based on the normalized weights of the three factors (plane distance weight, waveform difference weight, and waveform cumulative difference weight), combined with geological knowledge, a weighted mixture is performed to obtain a comprehensive weight.

[0090] For a predicted point, it is necessary to determine the interpolation weight for each known well involved in the prediction. This interpolation weight is derived by proportionally weighting the weights of three factors: planar distance weight, waveform difference weight, and cumulative waveform difference weight. During this weighted blending process, the mixing ratio of the three factor weights is roughly determined based on the geological and sedimentary characteristics of the study area. The sum of the mixing ratios of the three factor weights is maintained at 1, and the proportions of the three factors increase and decrease in a proportional manner.

[0091] If the mixing ratio of the three-factor weights is uncertain at the beginning of the study, different combination tests of the weight ratios of the three factors can be carried out, and the mixing scheme can be optimized based on the degree of conformity between the final interpolation prediction results map and the existing geological understanding. It is also possible to consider extracting some known wells from participating in the interpolation prediction, but only using them to test the reliability of the prediction results, so as to optimize and adjust the mixing ratio of the three factors and obtain a prediction result with relatively high reliability.

[0092] c) Based on the comprehensive weights and using the reservoir thickness information of known well sampling points, the distribution of reservoir thickness parameters is comprehensively predicted.

[0093] In this embodiment, based on the reservoir thickness information of known well points, the thickness parameters of the target reservoir are interpolated and predicted under the control of the weights of three factors (plane distance weight, waveform difference weight, and waveform cumulative difference weight). By selecting a comprehensive weight obtained by mixing the plane distance weight of 0.5, the waveform difference weight of 0.25, and the waveform cumulative difference weight of 0.25, the thickness distribution plane of the target reservoir is predicted, as shown in the attached figure. Figure 10 shown.

[0094] If the waveform difference and waveform cumulative difference are not considered and only the thickness information of known wells is used for inverse distance weighted interpolation, the reservoir thickness distribution obtained is as shown in the attached figure. Figure 11 As shown. Figure 10 , Attachment Figure 11A comparative analysis shows that the reservoir thickness map obtained by the three-factor mixed weight prediction has many advantages: the trend of sandstone spreading and depositing from the southeastern provenance to the northwest is clearer, the abnormal "bull's eye" phenomenon affected by the distribution of known wells is less likely to occur near the well point, the variation details of the reservoir thickness distribution are richer, and in some areas lacking known well control, the thickness variation can be predicted relatively reasonably. The accuracy and reliability of the prediction results are improved, and they can be more effectively used for closure identification and well deployment research.

[0095] Based on the reservoir thickness predictions derived from a combination of planar distance, waveform differences, and cumulative waveform differences, combined with known structural information indicating a structural contour approximately 45 degrees northeast of the region and the known dry sand body of Well A4, a lithologic trap (shown by the dashed line) was identified southwest of Well A4. A proposed well was then drilled at a higher point within this lithologic trap. This proposed well was subsequently drilled and encountered a 2-meter-thick, pure oil-rich reservoir in the target layer, confirming the reliability of the reservoir distribution predictions and lithologic trap identification results. In contrast, the proposed well was not located within the lithologic trap on the planar reservoir thickness predictions using only inverse distance weighted interpolation. In contrast, the reservoir thickness predictions using the three-factor hybrid weighting method proposed in the present invention's differential resonance-based reservoir prediction method demonstrate stronger predictability and higher reliability.

[0096] It can be seen from this embodiment that the reservoir prediction method based on differential resonance proposed in the present invention can, starting from known well data, use Manhattan distance to quantitatively analyze the differential resonance phenomenon between the coarsened well curve at a relatively macroscopic scale and the target reservoir parameters at a relatively microscopic scale; on the basis of this understanding, combined with formation velocity information, well-seismic time-depth calibration results, and seismic layer interpretation results, the seismic layers and time windows that sensitively reflect reservoir changes can be determined; accordingly, waveform clustering analysis is performed on the seismic waveform within the seismic time window to obtain waveform clustering plane data; based on the waveform clustering results, according to the reservoir parameters of the known well points, under the mixed control of three factor weights (plane distance weight, waveform difference weight, waveform cumulative difference weight), the target reservoir parameters are more reasonably and reliably interpolated and predicted, and efficiently applied to oil exploration and development research.

Claims

1. A reservoir prediction method based on differential resonance, characterized by: The following steps are involved: 1) Smoothing and coarsening the wave impedance logging curves of known wells in the work area; 2) Then, based on the roughened and smoothed wave impedance logging curve of the known well, a logging curve segment that resonates differentially with the reservoir parameters is selected near the target layer; 3) According to the preferred differential resonance logging curve segment, determine the seismic layer and time window corresponding to the logging curve resonance segment; 4) Based on the seismic horizon and time window determined in step 3), the seismic waveform clustering attributes are extracted based on the Manhattan distance of the -90-degree phase seismic waveform. Then, combined with the reservoir parameters of the known wells, the reservoir parameters of the target reservoir are interpolated and predicted by mixing the weights of the three difference factors; In step 2), the method of selecting a well logging curve segment near the target layer that has a differential resonance with the reservoir parameters includes the following steps: Taking the vertical depth of the target layer at the center of each known well as the reference depth, a series of depth windows are constructed by setting the vertical width value of the depth window and the upper and lower offset depth values ​​near the reference depth. Then, within each depth window, the Manhattan distance between the roughened well log curves of every two known wells is calculated. Calculate the absolute value of the reservoir parameter difference between every two known wells; For each depth window, a crossplot is drawn with the absolute value of the reservoir parameter difference between wells as the horizontal axis and the Manhattan distance of the well logging curve between wells as the vertical axis. By fitting the scattered points in the crossplot with a linear formula, the slope of the Manhattan distance changing with the reservoir parameter difference between wells is obtained. Compare the slope values ​​of all depth windows, and select the logging curve segment corresponding to the depth window with the largest slope change value as the preferred differential resonance logging curve segment; In step 4), the three difference factors are plane distance, waveform difference, and waveform cumulative difference; in the interpolation prediction process, for each point to be predicted, the weights of the three difference factors of plane distance, waveform difference, and waveform cumulative difference are calculated respectively, and each difference factor is normalized to obtain the interpolation weight value of each difference factor for each known well participating in the prediction. Then, a weighted blending is performed in combination with geological knowledge to obtain the comprehensive weight of each known well participating in the prediction corresponding to the point to be predicted. Then, based on the comprehensive weight, the reservoir parameters of the corresponding prediction point are interpolated and predicted; The plane distance is the distance between the point to be predicted and the known well points involved in the prediction, the waveform difference is the difference in waveform clustering results between the point to be predicted and the known well points involved in the prediction, and the waveform cumulative difference is the cumulative sum of the absolute values ​​of the differences in class values ​​of each two adjacent clustering points spanned by the line segment connecting the point to be predicted and the known well points involved in the prediction.

2. The reservoir prediction method based on differential resonance according to claim 1, characterized in that: The reservoir parameters include reservoir thickness, reservoir velocity, reservoir density, reservoir wave impedance, reservoir permeability or reservoir porosity.

3. The reservoir prediction method based on differential resonance according to claim 1, characterized in that: In step 1), the smoothing and coarsening process includes the following steps: obtaining -90-degree phase seismic data of the work area, calculating the vertical two-way time width of the seismic event axis near the target layer, drawing a probability distribution diagram, and determining the peak vertical two-way time corresponding to the peak position point of the vertical two-way time width distribution; then using the peak vertical two-way time to perform time-to-depth conversion to obtain the corresponding depth domain vertical thickness, and then performing mean smoothing processing on the logging wave impedance curve with a depth window of 0.2 to 0.3 times the depth domain vertical thickness.

4. The reservoir prediction method based on differential resonance according to claim 1, characterized in that: In step 2), the Manhattan distance between the smoothed and roughened logging curves of each two known wells is calculated according to the following Manhattan distance formula: M is the Manhattan distance between the smoothed and roughened log curves of two known wells, A i is the value of the smoothed and coarsened logging curve of a well at the i-th sampling point, B i is the value of the smoothed and coarsened logging curve of another well at the i-th sampling point, N is the number of vertical depth sampling points in the logging curve layer within the depth window, and i is a natural number of vertical depth sampling points traversed within the window.

Images