A high-resolution reservoir prediction method under seismic full-wave information mining

Abstract

The application discloses a high-resolution reservoir prediction method under seismic full-wave information mining, and comprises the following steps: firstly, fine seismic horizon interpretation and stratum framework construction under the guidance of isochronous stratum sequence; secondly, improving the seismic resolution by using well-free seismic resolution improvement technology, taking the well point correlation curve obtained by the automatic searching technology under the tolerance control as a quality control means; thirdly, fully utilizing the seismic information, extracting various main attributes and derived attributes, and then selecting attribute matrix combination to form "mass data" information related to the stratum; fourthly, taking the well logging curve of each well as a target learning training, and effectively improving the training effect and the coincidence degree of well point fitting under the premise of optimizing the algorithm and the tolerance control; and finally, generating a reservoir seismic prediction body by optimizing the training scheme. The application solves the industry problem of effective prediction of thin layers in exploration and development.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of oil and gas reservoir exploration and development, and particularly relates to a high-resolution reservoir prediction method under seismic full-wave information mining. BACKGROUND

[0002] Traditional seismic reservoir inversion includes deterministic inversion and geostatistical inversion, wherein the longitudinal resolution of deterministic seismic inversion is equivalent to the seismic resolution, such as 2000m buried stratum, 10-60HZ effective frequency band of conventional seismic acquisition, main frequency of about 30HZ, stratum interval velocity of about 2500-3000m / s, and the highest longitudinal resolution of seismic inversion of about 10-13m; and the geostatistical inversion can reach a longitudinal resolution of 2-3m, and the result is greatly affected by the size of the vertical and horizontal ranges of the variogram and the selection of the simulation method, and the inversion result model increases linearly with the increase of the longitudinal resolution, which seriously affects the need for fine understanding of thin reservoirs (reservoirs and interbedded layers) at the present stage. Therefore, a new method for effectively predicting thin layers needs to be found. SUMMARY

[0003] The purpose of the present application is to combine wells and seismic data to effectively break through the low resolution of conventional seismic reservoir inversion, and to provide a method for effectively predicting the distribution of thin reservoirs and interbedded layers by fully utilizing well and seismic data.

[0004] The technical purpose of the present application is achieved by the following technical scheme.

[0005] A high-resolution reservoir prediction method under seismic full-wave information mining, comprising the following steps:

[0006] Step 1: Seismic horizon fine interpretation and longitudinal boundary control

[0007] In step 1: first, complete seismic horizon calibration by means of fine synthetic seismogram, and check and adjust the seismic horizon calibration scheme by connecting to the backbone profile; then, complete fine seismic horizon interpretation by mutual verification of coherent volume, isochronous slice, waveform variable area profile, color profile, and three-dimensional visualization; secondly, build a stratigraphic framework using the interpreted horizon and faults, which is used for subsequent search boundary control in the vertical direction by automatic searching.

[0008] Step 2: Determine the sensitive parameters of reservoirs and non-reservoirs

[0009] In step 2: obtain the sensitivity attributes of reservoirs and non-reservoirs through rock physics analysis, which can be elastic attributes such as P-wave velocity, S-wave velocity or density, or non-elastic attributes such as porosity and permeability;

[0010] Step 3: Seismic resolution improvement processing under automatic quantitative quality control

[0011] In step 3: Using well-free seismic topology extension technology, while maintaining the overall characteristics of the seismic wavelet spectrum, the seismic wavelet spectrum distribution is extended towards higher frequencies to improve seismic resolution. Boundary control (1×1 seismic horizons after stratigraphic interpretation interpolation) is used. The seismic traces near the well after topology extension are correlated vertically with the well AC / DEN curves and the synthetic seismic records generated from the corresponding seismic wavelets, with a certain tolerance (automatic vertical position adjustment). Automatic search technology is used to determine the overall correlation (average correlation across multiple wells) between the seismic traces near the well and the synthetic records under different high-frequency topology bodies, and the seismic bodies with higher correlation are selected as the final result (see appendix). Figure 1 Appendix Figure 2 ), and generate new time-depth relationships for each well based on the optimal matching results of forward homing.

[0012] Step 4: Automatic extraction of seismic attributes and transformation of various derived attributes

[0013] In step 4, under the original seismic body, different seismic bodies and the original seismic body are obtained using frequency division technology. Then, the seismic properties of the frequency division body and the original seismic body can be divided into three major categories: amplitude, frequency and phase, which can generate more than 30 kinds of properties. The seismic properties are then combined in a matrix.

[0014] Step 5: Using massive amounts of information, automatically find the target, perform correlation fitting, and generate a high-resolution prediction volume.

[0015] In step 5, the target logging curves of each well are used for training. Since logging data is in the depth domain and seismic data is in the time domain, mapping logging data to seismic data using velocity fields is difficult due to various limitations. Achieving a perfect one-to-one correspondence (equichronous time-point correlation) between logging and seismic data is challenging, making it difficult to achieve high correlation using conventional correlation methods. By performing correlation in the vertical direction with a certain tolerance (automatic adjustment of vertical position), errors can be eliminated, effectively improving training results and the fit of well points. Generally, linear algorithms result in larger training fitting errors, while SVR algorithms using neural grids or vector machines show better fitting results. Using conventional vertical isochronous correlation, the correlation coefficient can reach over 0.8 while maintaining a correlation coefficient of approximately 0.3. Finally, the optimized training scheme generates a reservoir seismic prediction body.

[0016] Step 6: Characterization of thin reservoirs and interlayers

[0017] In step 6, the corresponding log curve shape of the thin reservoir from the conventional log curve can be known, the curve amplitude of the thin layer is reduced compared with the thick reservoir with the same property due to the influence of the adjacent layer; similarly, the attribute amplitude of the thin reservoir or the thin interlayer in the thin reservoir prediction result is also reduced, and the threshold value method will introduce too much information of the non-target body; therefore, the phenomenon of the reduction of the characteristic value in the prediction profile or the body is utilized, the top and bottom of the thin reservoir or the interlayer are explained by the artificial intervention method, and then the thickness of the thin reservoir or the interlayer is obtained, and it is particularly pointed out that the vertical stratum thickness is obtained by the seismic, and the real thickness or the vertical thickness of the stratum needs to be obtained by the combination of the well and the seismic (see the attached Figure 3 , and then the spatial distribution of the thin reservoir or the interlayer is effectively characterized.

[0018] Compared with the prior art, the advantages of the present application are that:

[0019] The present application provides a high-resolution reservoir prediction method under the seismic full-wave information mining, in the oil and gas field exploration and development stage, the existing seismic, logging and other data are fully utilized, the fine seismic horizon interpretation, the high resolution processing of the seismic data under the well control, the extraction and combination of the seismic full waveform information, the image fuzzy similarity discrimination, the automatic correlation search under the longitudinal tolerance control and the array transformation technology are utilized, the high-resolution instantaneous seismic attribute and the multiple iterations are fully utilized, the longitudinal resolution of the reservoir prediction is effectively improved, and the industry problem of improving the seismic high longitudinal resolution and the modeling in the present geological statistics inversion is effectively solved. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 Seismic frequency extension quantitative quality control curve and optimization diagram;

[0021] Figure 2 Automatic search under tolerance correlation and conventional isochronous comparison diagram;

[0022] Figure 3 Various stratum thickness diagrams in development;

[0023] Figure 4 The high-resolution reservoir prediction under the seismic full-wave information mining is the effect verification profile of A3H horizontal well in a certain oilfield in the South China Sea;

[0024] Figure 5 The high-resolution reservoir prediction under the seismic full-wave information mining is the effect verification profile of directional well in a certain oilfield in the South China Sea;

[0025] Figure 6 The high-resolution reservoir prediction under the seismic full-wave information mining is the effect verification profile of B04ST2 in a certain oilfield in the Bohai Sea;

[0026] Figure 7The effect verification profile of high resolution reservoir prediction under the seismic full wave information mining in E60 of a certain oilfield in Bohai;

[0027] Figure 8 The result map of high resolution reservoir prediction under the seismic full wave information mining;

[0028] Figure 9 The application effect map (1) of thin layer prediction method in a certain oilfield in Bohai;

[0029] Figure 10 The application effect map (2) of thin layer prediction method in a certain oilfield in Bohai.

[0030] For those skilled in the art, other related drawings can be obtained according to the above drawings without creative labor. DETAILED DESCRIPTION

[0031] The technical scheme of the present application will be further illustrated below in combination with specific embodiments.

[0032] Embodiment 1

[0033] The new technical method of the present application is used to predict the reservoir and interlayer distribution of the medium-high permeability sandstone reservoir in HJ2-21 oil group of Hanjiang group of a certain oilfield in South China Sea. The HJ2-21 layer is drilled in 10 wells of the oilfield, the average thickness of HJ2-21 reservoir is 6m, the interlayer thickness is 0.5-1.5m. The target layer is buried about 2000m, the effective frequency band of the earthquake is 8-60HZ, the main frequency is about 30HZ, the formation velocity is about 2500-3000m / s, and the highest resolution of the seismic inversion in the longitudinal direction is about 10-13m.

[0034] 1. The specific process of predicting the HJ2-21 small layer reservoir and interlayer distribution

[0035] (1) Fine interpretation of formation position and fault and establishment of formation framework

[0036] The seismic horizon calibration is completed by means of fine synthetic seismic record, and the seismic horizon calibration scheme is checked and adjusted by connecting the backbone profile. Then, the fine seismic horizon interpretation is completed by mutual verification of coherent body, isochronous slice, waveform variable area profile, color profile, three-dimensional visualization and the like. Then, the formation framework is built by using the formation interpretation horizon and fault.

[0037] (2) The porosity is selected as the characteristic curve of the reservoir by the reservoir sensitivity analysis, and the spatial distribution prediction of the porosity is carried out by using the high resolution reservoir prediction method under the seismic full wave information mining.

[0038] (3) Seismic resolution enhancement was performed under quality control of 20 wells drilled in the study area. The original effective seismic frequency band was 8–60 Hz, the initial design maximum frequency for frequency upscaling was 150 Hz, and the frequency increment step was 5 Hz. Seismic resolution enhancement was achieved using well-free seismic frequency upscaling technology. The quality control curves (see attached) were used to improve the resolution. Figure 1 It can be seen that when the overlay frequency is 95 Hz, the correlation between the synthetic record and the well-side seismic trace decreases sharply as the overlay frequency increases. Therefore, the optimal overlay frequency is 95 Hz.

[0039] (4) Seismic attribute extraction and derived attribute transformation

[0040] Based on the frequency-spread seismic pure wave data volume, frequency division was first performed at 15Hz, 25Hz, 35Hz, 45Hz, 55Hz, 65Hz, 75Hz, 85Hz, and 95Hz. Secondly, more than 30 seismic attributes, including seismic waveform envelope, waveform combination, waveform kurtosis, amplitude, wave number, instantaneous amplitude, and instantaneous phase, were selected from these 10 seismic data volumes. Then, attribute matrices were combined to form a massive amount of stratigraphic-related information (see appendix). Figure 4 ).

[0041] (5) Target correlation fitting and prediction generation under massive information

[0042] Optimization of BP neural network algorithm based on porosity (with appendix) Figure 5 The data is used for learning. During the learning process, the error is eliminated by performing correlation in the vertical direction and giving a certain tolerance (automatic adjustment of the vertical position), which effectively improves the training effect and the fit of the well points. The fitting correlation can reach more than 0.85. Finally, the "massive data" is converted into a porosity volume based on the learning results using the optimized training scheme.

[0043] (6) Effect inspection and characterization of thin reservoirs and interlayers

[0044] By utilizing reservoir prediction profiles from interconnected wells and blind well inspections (see attached) Figure 6 Appendix Figure 7 It can be seen that the high-resolution reservoir prediction method using seismic full-wave information mining has a high degree of consistency with well point prediction results and rich lateral information variation of reservoirs and interlayers, effectively solving the problem of severe modeling and rigid profile appearance under high vertical resolution in geostatistical inversion.

[0045] The cross-section shows that the HJ2-21 reservoir has obvious interlayers, which can be interpreted with artificial intervention at the top and bottom.

[0046] 2. Application Effect

[0047] (1) A new understanding of the internal interlayer of HJ2-21

[0048] Because the total thickness of HJ2-21 layer is about 10m, the drilled well has drilled the reservoir thickness of 6m averagely, the conventional deterministic reservoir inversion cannot effectively identify the internal interlayer, and the internal interlayer of HJ2-21 layer predicted by using the geostatistical inversion is only distributed in the continuous development well area (mainly affected by the internal interlayer of HJ2-21 layer of edge well 1A, A15P1 and A3P1 in the development area not developed), and the result of high resolution reservoir prediction (attached Figure 8 ) from the full wave information of the earthquake can be seen that the internal interlayer of HJ2-21 layer is limited in planar distribution and has poor continuity.

[0049] (2) Implementation effect

[0050] The production dynamic confirms the reliability of the prediction of the internal interlayer of HJ2-21 layer, the production well of the layer is 10, has a unified reservoir pressure system, which shows that the internal interlayer of HJ2-21 layer is locally distributed in the plane, only affects the reservoir distribution and the complexity of the reservoir, and cannot form a local reservoir lithology body on the plane, which does not affect the deployment of the overall development well, but has a certain influence on the oil and gas production capacity of the development well.

[0051] Secondly, the thin layer prediction method has also been well applied in a certain oilfield in Bohai (attached Figure 9 , attached Figure 10 )

[0052] In summary, the method successfully solves the serious contradiction between longitudinal high resolution and modeling in geostatistical seismic inversion, solves the phenomenon of low correlation with well points in the training learning in the conventional machine learning reservoir prediction, effectively makes the prediction accuracy reach 1-1.5m, and meets the needs of the research on thin reservoir and thin interlayer.

[0053] The above has made an exemplary description of the present application, and it should be explained that, without departing from the core of the present application, any simple modification, change or other equivalent replacement which can not cost the creative labor of the person skilled in the art falls into the protection scope of the present application.

Claims

1. A high-resolution reservoir prediction method under seismic full-wave information mining, characterized in that, The method comprises the following steps: Step 1: fine seismic horizon interpretation and longitudinal boundary control; First, fine synthetic seismogram is used to calibrate seismic horizon, and the seismic horizon calibration scheme is checked and adjusted through the connection of the backbone profile; then, the fine seismic horizon interpretation is completed through the mutual verification of coherent volume, isochronous slice, waveform variable area profile, color profile and 3D visualization; secondly, the stratigraphic framework is built by using the horizon interpretation horizon and the fault, which is used for subsequent automatic searching of the boundary in the longitudinal direction under the image identification control; Step 2: determining the sensitive parameters of reservoir and non-reservoir; Through rock physical analysis, the sensitivity attributes of reservoir and non-reservoir are obtained, including P-wave velocity, S-wave velocity, density, porosity or permeability; Step 3: seismic resolution enhancement processing under automatic searching and quantitative quality control; The well-controlled seismic frequency extension technology is used to extend the seismic wavelet spectrum distribution to the high frequency end while maintaining the overall characteristics of the seismic wavelet spectrum, so as to improve the seismic resolution, and the boundary control is used to correlate the well seismic trace after seismic frequency extension with the well AC / DEN curve and the synthetic seismogram generated by the corresponding seismic wavelet after seismic frequency extension in the longitudinal direction with a certain tolerance, and the automatic searching technology is used to select the seismic volume with higher correlation as the final result according to the overall correlation between the well seismic trace and the synthetic seismogram under different high-frequency extension volumes, and generate new time-depth relationship of each well according to the optimal matching result of the positive dynamic searching; Step 4: automatic extraction of seismic attributes and transformation of various derived attributes; Under the original seismic volume, different seismic volumes and original seismic volumes are obtained by using frequency division technology, then the seismic attributes of the frequency division volume and the original seismic volume are divided into three categories of amplitude, frequency and phase, more than thirty kinds of attributes are derived, and the seismic attributes are combined in matrix form; Step 5: automatic searching target correlation fitting and high-resolution prediction volume generation under huge amount of information; The target logging curve of the drilled well in the study area is used for learning and training, the correlation with a certain tolerance is used to realize the elimination of errors, the training effect and the coincidence degree of well point fitting are effectively improved, the neural network is selected, the correlation degree is more than 0.8 under the condition of correlation coefficient 0.3 under the conventional longitudinal isochronous corresponding correlation, finally, the training scheme is optimized to generate the reservoir seismic prediction volume, and the optimized time-depth relationship of each well is generated again; Step 6: thin reservoir and interlayer description; The phenomenon of characteristic value reduction in the prediction profile or volume is used to interpret the top and bottom of the thin reservoir or interlayer through artificial intervention, and then the thickness of the thin reservoir or interlayer is obtained.

2. The high-resolution reservoir prediction method under seismic full-wave information mining according to claim 1, characterized in that, In step 6, the seismic obtains the vertical stratigraphic thickness, for the horizontal well or high angle deviated well, the true thickness or vertical thickness of the stratum is obtained under the combination of well and seismic, and then the spatial distribution of the thin reservoir or interlayer is effectively characterized.

Images