Karst reservoir filling analysis method and system based on spectral decomposition and machine learning

文档序号：1936026 发布日期：2021-12-07 浏览：13次中文

阅读说明：本技术 基于频谱分解和机器学习的岩溶储层充填分析方法和系统 (Karst reservoir filling analysis method and system based on spectral decomposition and machine learning ) 是由张江云田飞底青云郑文浩王中兴杨永友张文秀于 2021-09-13 设计创作，主要内容包括：本发明属于数据识别、记录载体的处理领域,具体涉及了一种基于频谱分解和机器学习的岩溶储层充填分析方法和系统,旨在解决现有的石油勘探技术无法预测横向变化快的储层、无法识别大范围内复杂盆地碳酸盐岩洞穴型储层的发育特征的问题。本发明包括：获取标准化测井曲线数据；通过混合相位子波反褶积和扩散滤波,获得高精度的三维地震振幅数据体；确定能够反映整体地质情况的最优样本数参量进而确定时深转化关系和对储层敏感的特征参数；通过波阻抗解释结论门槛值计算缝洞型储层结构特征与空间分布规律；通过解释结论门槛值进行交会分析获得古岩溶洞穴充填物三维空间形态特征。本发明达到识别大范围内复杂盆地碳酸盐岩岩溶洞穴型储层的发育特征的效果提高了刻画的精度。(The invention belongs to the field of data identification and record carrier processing, and particularly relates to a karst reservoir filling analysis method and system based on spectral decomposition and machine learning, aiming at solving the problems that the existing oil exploration technology cannot predict a reservoir with rapid transverse change and cannot identify the development characteristics of a complex basin carbonate cave type reservoir in a large range. The invention comprises the following steps: acquiring standardized logging curve data; obtaining a high-precision three-dimensional seismic amplitude data volume through deconvolution and diffusion filtering of mixed phase wavelets; determining an optimal sample number parameter capable of reflecting the overall geological condition so as to determine a deep transformation relation and characteristic parameters sensitive to a reservoir; calculating structural characteristics and a spatial distribution rule of the fracture-cavity reservoir through a wave impedance interpretation conclusion threshold value; and performing intersection analysis by explaining a conclusion threshold value to obtain three-dimensional space morphological characteristics of the paleo-karst cave filling. The method achieves the effect of identifying the development characteristics of the complex basin carbonate karst cave type reservoir in a large range, and improves the precision of the carving.)

1. A method for analyzing a karst reservoir filling based on spectral decomposition and machine learning, the method comprising:

step S100, obtaining original geophysical logging data: obtaining raw logging data for each sample well by a logging device, comprising: measuring the natural potential SP of each sample well through a measuring electrode, measuring the natural gamma GR of each sample well through a natural gamma underground device and a natural gamma ground instrument, and obtaining the well diameter CAL of each sample well through a well diameter arm; obtaining resistivity curve data by conventional logging equipment: deep lateral logging RLLD, shallow lateral logging RLLS and micro lateral logging RLLM, and physical property characterization curve data: a compensated neutron CNL, a compensated acoustic curve AC and a density curve DEN;

step S200, acquiring seismic data, acquiring original seismic wave reflection signal data through a seismic wave excitation device and a receiving device, and acquiring isochronous three-dimensional spread of a target layer position mark layer according to the waveform of the original seismic wave reflection signal data;

step S300, preprocessing original geophysical logging data: drawing logging curve data based on all original logging data of the sample well, and performing abnormal value processing and standardization processing to obtain standardized logging curve data;

step S400, seismic data preprocessing: based on the seismic wave reflection signal data, obtaining a high-precision three-dimensional seismic amplitude data volume through mixed phase wavelet deconvolution and diffusion filtering;

step S500, well seismic calibration and characteristic parameter selection: acquiring a wave impedance curve of a sample well based on a compensation acoustic curve AC and a density curve DEN in the standardized logging curve data, further calculating a reflection coefficient curve, acquiring the preferred frequency of a Rake wavelet to keep the same as the main frequency of a high-precision three-dimensional seismic amplitude data body, performing convolution operation on the Rake wavelet and the reflection coefficient curve to obtain a synthetic seismic record, comparing the depth data of a target layer position mark layer with the isochronous three-dimensional spread of the target layer position mark layer for well seismic calibration, calculating the correlation between the synthetic seismic record and a well-side seismic channel waveform, judging that a well seismic result is qualified when the correlation is greater than or equal to a preset first threshold, and acquiring the time-depth conversion relation between the logging curve data and the seismic record and characteristic parameters sensitive to a reservoir;

the characteristic parameter sensitive to the reservoir is obtained by the method comprising the following steps:

drawing a histogram by responding to logging parameters generated by different geologic bodies beside a well, and selecting the standardized logging curve data as characteristic parameters sensitive to a reservoir when the numerical value of the standardized logging curve data can distinguish data points above a second threshold value preset by different logging interpretation conclusions; the characteristic parameters sensitive to the reservoir at least comprise wave impedance IMP, well diameter CAL, natural gamma GR, natural potential SP, resistivity curve data: deep lateral logging RLLD, shallow lateral logging RLLS and micro lateral logging RLLM, and the data of the physical property characterization curve: one or more of a compensated neutron CNL, a compensated acoustic curve AC, a density curve DEN;

step S600, constructing an isochronous trellis model: constructing an isochronous trellis model based on sedimentary stratum rules reflected by the high-precision three-dimensional seismic amplitude data volume and the time-depth conversion relation;

s700, simulating the parameters of the reservoir layer between wells: determining an optimal sample number parameter capable of reflecting the overall geological condition based on the standardized logging curve data of each sample well, selecting characteristic parameter data sensitive to a reservoir stratum of the sample well with the highest seismic waveform correlation of the optimal sample number parameter to construct an initial model, continuously correcting parameters of the initial model, and outputting a high-precision characteristic value simulation result data body, wherein the high-precision characteristic value simulation result data body is a data body in one-to-one correspondence with the characteristic parameters sensitive to the reservoir stratum;

step S800, depicting the boundary of the karst cave system between wells: drawing a wave impedance curve histogram based on a wave impedance curve IMP of a sample well in characteristic parameters sensitive to a reservoir, and determining and distinguishing an ancient karst cave from a surrounding rock wave impedance threshold value;

based on the paleo-karst cave and the surrounding rock wave impedance threshold value, defining a region exceeding the paleo-karst cave and surrounding rock wave impedance threshold value in a high-precision characteristic value simulation result data body as carbonate rock bedrock, and defining a region lower than the paleo-karst cave and surrounding rock wave impedance threshold value as a paleo-karst cave reservoir development position to obtain paleo-karst cave reservoir structural characteristics and a spatial distribution rule;

step S900, depicting a lithologic property boundary filled in the cave: depicting the lithologic property boundary filled in the cave: constructing an intersection layout based on the characteristic parameters sensitive to the reservoir, framing data points in the intersection layout as well logging interpretation sample points according to the lithological information and the physical property information of the individual depth section, and obtaining interpretation conclusion threshold values of the filling degree and the filling type;

performing intersection analysis on the high-precision characteristic value simulation result data volume through the interpretation conclusion threshold value to obtain the filling degree and filler combination three-dimensional space morphological characteristics in the cave;

step S1000, describing the structure and filling of the ancient karst cave: and carving development characteristics of the space distribution and the internal filling of the paleo-karst cave by adopting a lithologic shielding technology and a three-dimensional carving technology based on the structural characteristics and the space distribution rule of the paleo-karst cave reservoir and based on the internal filling degree of the cave and the three-dimensional space morphological characteristics of filler combination.

2. The method for karst reservoir filling analysis based on spectral decomposition and machine learning according to claim 1, wherein the measuring the natural potential SP of each sample well through the measuring electrode and the measuring the natural gamma GR of each sample well through the natural gamma downhole device and the natural gamma surface instrument specifically comprise:

measuring the natural potential SP of each sample well through the measuring electrode:

arranging a measuring electrode N on the ground, and arranging a measuring motor M underground through a cable;

lifting the measuring electrode M along the well axis to measure the change of the natural potential along with the well depth;

the calculation method of the natural potential value comprises the following steps:

wherein the content of the first and second substances,the total natural potential is the total natural potential,in order to be a diffusion potential coefficient,in order to obtain a diffusion adsorption potential coefficient,the resistivity value of the slurry filtrate is shown as,is the formation water resistivity value;

the natural gamma GR of each sample well is measured through the natural gamma underground device and the natural gamma ground instrument:

the natural gamma downhole device comprises a detector, an amplifier and a high-voltage power supply;

acquiring natural gamma rays through a detector, converting the natural gamma rays into electric pulse signals, and amplifying the electric pulse signals through an amplifier;

the ground instrument converts the electric pulse count formed every minute into a potential difference for recording.

3. The method for analyzing the filling of the karst reservoir based on the spectral decomposition and the machine learning according to claim 1, wherein the step S300 comprises:

step S310, drawing original logging curve data based on the original logging data;

step S320, based on the original logging curve data, removing outliers to obtain logging curve data with the outliers removed;

and S330, superposing single logging curve histogram data of all sample well positions in the work area based on the logging curve data without outliers, and obtaining standardized logging curve data by integrating threshold values.

4. The method for analyzing the filling of the karst reservoir based on the spectral decomposition and the machine learning according to claim 1, wherein the step S400 specifically includes:

step S410, based on the seismic wave reflection signal data, expressing a frequency domain seismic record convolution model as:

wherein the content of the first and second substances,representing the fourier transformed seismic records,representing the wavelet after the fourier transform,a frequency spectrum representing the fourier transformed reflection coefficient,represents angular frequency;

step S420, logarithm taking two sides of the equation of the frequency domain seismic record convolution model is converted into a linear system, and a linear seismic record convolution model is obtained:

step S430, performing inverse Fourier transform on the linear seismic record convolution model to obtain a cepstrum sequence:

wherein the content of the first and second substances,representing a repeating spectral sequence of seismic waveform recordings,a cepstrum sequence representing seismic wavelets,a repeating spectral sequence representing the reflection coefficients of the formation,representing a seismic waveform recording time;

step S440, based on the cepstrum sequence, performing wavelet and reflection coefficient separation through a low-pass filter, and extracting a wavelet amplitude spectrum;

step S450, obtaining the amplitude spectrum of the simulated seismic wavelet by a least square method:

wherein, least square method is used for fitting parametersIs a constant number of times, and is,a representation of the amplitude spectrum of the wavelet is obtained,anda polynomial expression representing f, which represents the frequency of the seismic wave;

step S460, obtaining a wavelet maximum phase component and a wavelet minimum phase component based on the simulated seismic wavelet amplitude spectrum;

wavelet setting deviceHas a maximum phase component ofThe minimum phase component isWavelet of fundamental waveComprises the following steps:

the magnitude spectrum is represented in the cepstrum as:

wherein the repetition spectrum of the amplitude spectrumSymmetrically displayed on the positive and negative axes of the match score,for maximum phase component of seismic waveletThe cepstrum of the corresponding minimum phase function,for minimum phase component of seismic waveletsThe corresponding cepstrum of the maximum phase function;

step S470, determining a group of mixed phase wavelet sets with the same amplitude spectrum based on the cepstrum in the amplitude spectrum, continuously adjusting the parameters of Shu' S wavelets, maintaining low frequency, expanding high frequency and properly improving dominant frequency to construct an expected output wavelet form, searching for an optimal balance point between resolution and fidelity by taking a signal-to-noise ratio spectrum as a reference under the control of a well curve, and obtaining waveform data after shaping;

step S480, constructing a tensor diffusion model based on the shaped waveform data:

wherein the content of the first and second substances,it is shown that the time of diffusion,denotes a divergence operator, D denotes a tensor-type diffusion coefficient of the diffusion filter, U denotes a diffusion filtering result,to representThe result of diffusion filtering when =0,to representThe waveform data after the shaping at that time is used as an initial condition of the tensor diffusion model,a gradient representing a result of the diffusion filtering;

constructing a gradient structure tensor based on the tensor diffusion model:

where, U represents the result of the diffusion filtering,representing a gradient vector tensor product;

the expression scale isGaussian function of (d):

wherein r represents the calculated radius;

the eigenvectors of the structure tensor are:

wherein the content of the first and second substances,、andthe 3 eigenvectors, expressed as gradient structure tensors, can be considered as local orthogonal coordinate systems,pointing in the direction of the gradient of the seismic signal,andwith the plane of composition parallel to the seismic signalThe local structural features of (a) the structure,、andare respectively connected with、Andcorresponding three characteristic values;

step S490, calculating a linear structure confidence metric, a planar structure confidence metric, and a diffusion tensor, respectively, based on the feature vectors of the structure tensor;

the presence structure confidence measureComprises the following steps:

the planar structure confidence measureComprises the following steps:

the diffusion tensor D is:

wherein the content of the first and second substances,、andthree non-negative eigenvalues representing the diffusion tensor, each representing a diffusion filter edge、Andthe filter strengths of the three characteristic directions;

5. The method for karst reservoir filling analysis based on spectral decomposition and machine learning of claim 4, wherein the calculation method for simulating the amplitude spectrum of the seismic wavelet is as follows:

positioning a maximum value of an amplitude spectrum in seismic wave reflection signal data and a frequency corresponding to the maximum value;

obtaining parameters by fitting the maximum value of the seismic signal amplitude spectrum and the simulated seismic wavelet amplitude spectrum in a least square method modeAndobtaining the corresponding frequency amplitude value of the fitted maximum value by the coefficients of the polynomial;

dividing the maximum value of the seismic signal amplitude spectrum by the fitted amplitude value of the corresponding frequency, and further using a quotient fitting polynomialThe coefficient of (a).

6. The method for karst reservoir filling analysis based on spectral decomposition and machine learning according to claim 1, wherein the step S700 specifically comprises steps S710 to S790:

step S710, selecting a sample well as a reference target well, and setting an initial sample number parameter to be 1;

s720, selecting the sample well standardized logging curve characteristic parameter data and the reference target well standardized logging curve characteristic parameter data with the quantity as the sample number parameters according to the waveform similarity principle to perform correlation analysis to obtain a characteristic parameter correlation value of the sample number parameter-reference target well;

step S730, increasing the sample number parameters 1 by 1, repeating the method of the step S720 to obtain the correlation values of the sample number parameters and the reference target well curve characteristic parameters corresponding to the sample number parameters, and connecting the correlation values of all the sample number parameters and the reference target well curve characteristic parameters to obtain a correlation curve of the correlation of the reference well curve characteristic parameters along with the change of the sample number parameters;

step S740, selecting another sample well as a reference target well, repeating the steps S710-S730 to obtain a correlation curve of the correlation of the characteristic parameters of the multiple reference wells along with the change of the sample number parameters, fitting the correlation curves of the correlation of the characteristic parameters of all the reference wells along with the change of the sample number parameters into an overall correlation curve, selecting an inflection point at which the correlation in the overall correlation curve increases along with the increase of the sample parameters and finally remains stable, and determining the optimal sample number parameters;

step S750, based on the high-precision three-dimensional seismic amplitude data volume and the isochronous grid model, calculating waveform correlation between a point to be detected and a sample well position, sorting the waveform correlation from large to small, and selecting characteristic parameter data sensitive to a reservoir of a sample well with the highest seismic waveform correlation of an optimal sample quantity parameter strip; constructing an initial model based on the sample well corresponding to the seismic waveform characteristic data of the sample well with the highest correlation through an inter-well characteristic parameter interpolation mode;

step S760, selecting characteristic parameters sensitive to the reservoir corresponding to the curve parameters to be simulated of the sample well with the highest correlation degree of the seismic waveform of the parameter strip of the optimal sample number as prior information based on the initial model; the curve parameters to be simulated are curve parameters which correspond to the characteristic parameters sensitive to the reservoir layer one by one, at least comprise wave impedance IMP, well diameter CAL, natural gamma GR, natural potential SP and resistivity curve data: deep lateral logging RLLD, shallow lateral logging RLLS and micro lateral logging RLLM, and the data of the physical property characterization curve: one or more of compensated neutrons CNL, compensated acoustic wave AC, density curve DEN;

step S770, performing matched filtering on the initial model and the prior information to obtain a maximum likelihood function;

step S780, based on the maximum likelihood function and prior information, obtaining the posterior probability statistical distribution density under a Bayes frame, and sampling the posterior probability statistical distribution density to obtain a target function;

step S790, taking the target function as the input of the initial model, sampling posterior probability distribution by a Markov chain Monte Carlo method MCMC and Metropolis-Hastings sampling criterion, continuously optimizing parameters of the initial model, selecting a solution of the target function when the maximum value is taken as random realization, taking the average value of multiple random realization as expected value output, and taking the expected value output as a high-precision characteristic value simulation result data body; and parameters in the high-precision characteristic value simulation result data volume correspond to the characteristic parameters sensitive to the reservoir one by one.

7. The method for karst reservoir filling analysis based on spectral decomposition and machine learning according to claim 6, wherein the step S780 specifically includes:

step S781, satisfying the rule of Gaussian distribution by white noise, and expressing the parameters of the high-precision characteristic value simulation result data body as follows:

y represents parameters of a logging curve high-precision characteristic value simulation result data volume, X represents actual characteristic parameter values of an underground stratum to be solved, and N represents random noise;

step S782, becauseAlso satisfying a gaussian distribution, the initial objective function can be determined as:

wherein the content of the first and second substances,a function relating to a posteriori information is represented,showing that the characteristic curve of the sample well is matched and filtered by selecting the sample well based on the optimal sample number, the posterior probability statistical distribution density is obtained, and the expected value of the characteristic parameter is further calculated,a covariance representing white noise;

step S783, based on the initial objective function, introducing prior information into the objective function through maximum posterior estimation, and obtaining a stable objective function as follows:

wherein the content of the first and second substances,representing the wave impedance or characteristic parameter to be simulated,representing functions related to prior information such as geological and well log data,representation for coordinationAndthe smoothing parameters of the mutual influence between them.

8. The method for karst reservoir filling analysis based on spectral decomposition and machine learning according to claim 6, wherein the step S790 is as follows:

s791, setting M as a target space, n as the total number of samples, and M as the number of samples when the Markov chain tends to be stable;

step S792, presetting a Markov chain to make the Markov chain converge to stable distribution;

step S793, from a certain point in MStarting from, sampling simulation is performed through a Markov chain to generate a point sequence:；

step S794, functionThe expected estimate of (c) is:

wherein n represents the total number of generated samples, m represents the number of samples when the Markov chain reaches a plateau, and k represents an accumulation parameter;

step S795, select a transfer functionAnd an initial valueIf the parameter value at the beginning of the ith iteration isThen, the ith iteration process is:

fromExtract an alternative valueCalculating alternative valuesProbability of acceptance of：

Step S796 ofIs arranged atBy probabilityIs arranged at；

Step S797, continuously disturbing the parameters of the initial model, repeating the method of the steps C692-C696 until the preset iteration number n is reached, and obtaining a posterior sampleAnd further calculating each order matrix of posterior distribution to obtain an expected output value, and outputting the expected value as a high-precision characteristic value simulation result data body.

9. The method for analyzing the filling of the karst reservoir based on the spectrum decomposition and the machine learning of claim 1, wherein the step S900 comprises:

step S910, constructing a cross map template by taking the wave impedance value and the natural GR value as characteristic parameters sensitive to the reservoir;

step S920, based on the cross plot template, dividing sample points into full filling, half filling and bedrock according to the lithological information and physical property information of the individual depth section;

step S930, selecting half-filling sample points with better reservoir performance in a frame mode, and calculating to obtain a two-dimensional intersection graph interpretation conclusion threshold value;

step S940, inputting the two-dimensional intersection chart interpretation conclusion threshold value into a three-dimensional natural GR characteristic parameter simulation data body and a wave impedance inversion data body, and depicting the spatial structure form of a semi-filled reservoir to obtain the three-dimensional spatial form characteristics of the semi-filled part of the ancient karst cave;

step S900 further includes the step of acquiring other cave characteristics:

and step S950, dividing sample points into bedrocks, sedimentary filling, chemical filling, collapse filling and mixed filling according to the lithological information and the physical property information of the individual depth section based on the intersection map template, framing various filler sample points, respectively calculating corresponding interpretation conclusion threshold values, respectively depicting corresponding spatial structure forms, and obtaining three-dimensional spatial form characteristics of the paleo-karst cave filler.

10. A system for karst reservoir filling analysis based on spectral decomposition and machine learning, the system comprising: the system comprises an original geophysical logging data acquisition module, a seismic data acquisition module, an original geophysical logging data preprocessing module, a seismic data preprocessing module, a well seismic calibration and characteristic parameter selection module, an isochronous grid model construction module, an inter-well characteristic parameter simulation module, an inter-well karst cave system boundary delineation module, a karst cave internal filling lithology boundary delineation module and an ancient karst cave structure and filling characteristic three-dimensional description module;

the original geophysical logging data acquisition module is configured to acquire original logging data of each sample well through logging equipment, and comprises: measuring the natural potential SP of each sample well through a measuring electrode, measuring the natural gamma GR of each sample well through a natural gamma underground device and a natural gamma ground instrument, and obtaining the well diameter CAL of each sample well through a well diameter arm; obtaining resistivity curve data by conventional logging equipment: deep lateral logging RLLD, shallow lateral logging RLLS and micro lateral logging RLLM, and physical property characterization curve data: a compensated neutron CNL, a compensated acoustic curve AC and a density curve DEN;

the seismic data acquisition module is configured to acquire original seismic wave reflection signal data through a seismic wave excitation device and a receiving device, and acquire isochronous three-dimensional spread of a target layer position mark layer according to the waveform of the original seismic wave reflection signal data;

the original geophysical logging data preprocessing module is configured to draw logging curve data based on original logging data of the sample well, perform abnormal value processing and standardization processing, and obtain standardized logging curve data;

the seismic data preprocessing module is configured to obtain a high-precision three-dimensional seismic amplitude data volume through mixed phase wavelet deconvolution and diffusion filtering based on the seismic wave reflection signal data;

the well-seismic calibration and characteristic parameter selection module is configured to obtain a wave impedance curve of the sample well based on a compensation acoustic curve AC and a density curve DEN in the standardized well-logging curve data, further calculating a reflection coefficient curve, acquiring the preferred frequency of the Rake wavelet to make the Rake wavelet consistent with the main frequency of the high-precision three-dimensional seismic amplitude data body, performing convolution operation on the Rake wavelet and the reflection coefficient curve to obtain a synthetic seismic record, comparing the depth data of the target horizon mark layer with the isochronous three-dimensional spread of the target horizon mark layer to carry out well seismic calibration, calculating the correlation degree of the synthetic seismic record and the waveform of seismic channels beside the well, when the correlation degree is greater than or equal to a preset first threshold value, judging that the well-seismic calibration result is qualified, and obtaining a time-depth conversion relation between logging curve data and seismic records and characteristic parameters sensitive to a reservoir;

the characteristic parameter sensitive to the reservoir is obtained by the method comprising the following steps:

the isochronous trellis model building module is configured to build an isochronous trellis model based on sedimentary stratum rules reflected by the high-precision three-dimensional seismic amplitude data volume and the time-depth conversion relation;

the inter-well characteristic parameter simulation module is configured to determine an optimal sample number parameter capable of reflecting the overall geological condition based on the standardized logging curve data of each sample well, select the characteristic parameter data sensitive to the reservoir of the sample well with the highest seismic waveform correlation of the optimal sample number parameter strip to construct an initial model, continuously correct the parameters of the initial model, and output a high-precision characteristic value simulation result data volume, wherein the high-precision characteristic value simulation result data volume is a data volume corresponding to the characteristic parameters sensitive to the reservoir one by one;

the inter-well karst cave system boundary delineation module is configured to draw a wave impedance curve histogram based on a wave impedance curve IMP in characteristic parameters of a sample well sensitive to a reservoir, and determine and distinguish a paleo-karst cave from a surrounding rock wave impedance threshold value;

based on the paleo-karst cave and the surrounding rock wave impedance threshold value, defining a region exceeding the impedance threshold value in a high-precision characteristic value simulation result data body as carbonate bedrock, and defining a region lower than the impedance threshold value as a paleo-karst cave reservoir development position to obtain paleo-karst cave reservoir structural characteristics and a spatial distribution rule;

the lithologic property boundary description module filled in the karst cave is configured to describe the lithologic property boundary filled in the cave: constructing an intersection layout based on the characteristic parameters sensitive to the reservoir, framing data points in the intersection layout as well logging interpretation sample points according to the lithological information and the physical property information of the individual depth section, and obtaining interpretation conclusion threshold values of the filling degree and the filling type;

the three-dimensional description module for the structure and the filling characteristics of the paleo-karst cave is configured to describe the structure and the filling of the paleo-karst cave: and carving development characteristics of the space distribution and the internal filling of the paleo-karst cave by adopting a lithologic shielding technology and a three-dimensional carving technology based on the structural characteristics and the space distribution rule of the paleo-karst cave reservoir and based on the internal filling degree of the cave and the three-dimensional space morphological characteristics of filler combination.

Technical Field

The invention belongs to the field of data identification and record carrier processing, and particularly relates to a karst reservoir filling analysis method and system based on spectral decomposition and machine learning.

Background

In recent years, deep-layer oil and gas resources in the basin are gradually becoming the center of gravity in the field of oil and gas exploration. Deep carbonate oil and gas exploration has huge potential, and the oil reservoirs account for 52 percent of the globally proven oil reserves and 60 percent of the global yield, and are one of the most important oil and gas target areas in many basins in the world. Although most of the oil and gas in China are produced from clastic rock reservoirs at present, carbonate rock reservoirs play more and more important roles. China marine carbonate reservoir oil and gas resources are rich, a large amount of petroleum resources are found in onshore marine basins such as Tarim, Ordos, Sichuan and Bohai Bay, and the found petroleum reserves are 340 multiplied by 10⁸t, however, the detemining rate was only 4.56% and 13.17%. Therefore, the marine carbonate reservoir has great oil and gas exploitation potential and is an important oil and gas exploration and succession field in China.

Researches show that the filling characteristics of the karst reservoir can greatly influence the oil and gas productivity, for example, an unfilled or semi-filled ancient karst cave is beneficial to oil and gas migration, can effectively store and well form during later-stage adjustment; full-filled paleo-karst caverns are difficult to form into an effective reservoir space. According to statistics of scholars, sand mud, cave collapse gravel, chemical filling and the like occupy more than 70% of the space of the ancient karst cave reservoir. The differences of lithology and physical properties of the fillers seriously affect the oil and gas reservoir storage space and seepage effect.

The carbonate rock ancient karst reservoir space distribution is complex and the mutual connection is tight. The key problem of the research of the paleo-karst reservoir stratum is how to accurately know the structure of the deep-buried paleo-karst system, how to effectively identify the position of the paleo-karst system by utilizing geophysical data and how to accurately describe the filling characteristics in the paleo-karst cave system.

Conventional geophysical exploration tools face the bottleneck problem of reservoir identification. Due to deposition, diagenesis and later-stage karst transformation, caves and cracks of the formed carbonate rock ancient karst reservoir are discrete and random, and the carbonate rock ancient karst reservoir has strong heterogeneity, so that the detection experience obtained from a shallow clastic rock reservoir and a porous carbonate rock reservoir cannot be directly used for exploring the ancient karst reservoir of a Tahe oil field. The sparse logging curve has high longitudinal resolution, but the detection range is limited, reservoir parameters far away from a shaft cannot be accurately judged, and the characteristics determine that the traditional logging detection means has high difficulty in quantifying the fine prediction of a reservoir body.

As the ancient karst cave type reservoir is buried deeply, the seismic wave energy absorption attenuation effect is strong, so that the seismic data dominant frequency is obviously reduced, the resolution is greatly reduced, and the signal-to-noise ratio in a deep time window is too low. And the drilling data and the core samples show that most caves are less than 5 meters in height, the maximum height is about 15 meters, and the resolution of the original seismic data cannot achieve the accurate identification effect. In addition, compared with a logging curve, the lithology physical property information content contained in the seismic attribute is low, and the reservoir filling characteristics cannot be effectively identified.

Therefore, how to fully utilize the geophysical information and establish the connection between a high-precision well logging interpretation method and large-range seismic detection data is the key for improving the reservoir prediction resolution and the longitudinal and transverse detection capability.

Well-to-seismic joint inversion can convert seismic reflection data into quantitative estimation of rock properties, and is widely applied to reservoir prediction. Well-to-seismic joint inversion based on geology statistics is characterized in that a geology statistics principle is utilized, a geologic mode is used as guidance, multi-class and multi-scale data such as well logging and earthquake are integrated according to a variation function changing along with distance, an ancient karst cave and compact surrounding rocks can be distinguished, and lithology and physical property parameters of a reservoir body can be represented on the basis. The geological statistical thought gives play to the geological advantages in reservoir modeling, the theoretical problems of detection precision and detection range in the geophysical prospecting field are solved, the limitation of seismic resolution is broken through, and the effect of random simulation of a thin reservoir is achieved. However, the method is preferred for sample wells that rely on a prior model and reference a variation function that varies with distance. The seismic data only play a role in optimizing the random simulation result, and the contribution of the seismic information to the sample optimization is not considered. The inversion result is directly controlled by the inter-well interpolation in the high-frequency section, the transverse prediction randomness is strong, and the reservoir with quick transverse change cannot be predicted.

Disclosure of Invention

The method aims to solve the technical problem of reservoir identification at present, namely the existing method for analyzing the filling characteristics of the oil exploration cave-type reservoir only takes geological data as optimization to a random simulation structure and does not consider the contribution of seismic information to well logging sample optimization, so that the inversion result is directly controlled by inter-well interpolation in a high-frequency section, the transverse prediction randomness is strong, and the reservoir with quick transverse change cannot be predicted. The invention provides a karst reservoir filling analysis method based on spectral decomposition and machine learning, which comprises the following steps:

the characteristic parameter sensitive to the reservoir is obtained by the method comprising the following steps:

step S900, depicting a lithologic property boundary filled in the cave: constructing an intersection layout based on the characteristic parameters sensitive to the reservoir, framing data points in the intersection layout as well logging interpretation sample points according to the lithological information and the physical property information of the individual depth section, and obtaining interpretation conclusion threshold values of the filling degree and the filling type;

In some preferred embodiments, the measuring the natural potential SP of each sample well by the measuring electrode and the measuring the natural gamma GR of each sample well by the natural gamma downhole device and the natural gamma surface instrument specifically include: