Method for predicting organic carbon content of hydrocarbon source rock in earthquake based on multi-layer perceptron

Abstract

The invention relates to a multi-layer perceptron-based earthquake hydrocarbon source rock organic carbon content prediction method. The method comprises the following steps: acquiring organic carbon content data of a rock sample; extracting seismic attribute data of a well-side seismic channel of a well where the rock sample is located; carrying out correlation analysis on the organic carbon content data of the rock sample and the seismic attribute data of the seismic trace beside the well, and preferably selecting optimal seismic attribute data with high correlation; training the organic carbon content data and the optimized seismic attribute data of the rock sample through a multi-layer sensor to obtain a multi-layer sensor model for predicting the organic carbon content through seismic attributes; and inputting the optimal seismic attribute data of the whole reservoir in the research area into the multi-layer perceptron model, and further predicting the organic carbon content data of the whole reservoir in the research area. Based on seismic exploration data widely implemented in early-stage exploration of an oil and gas field and a small amount of core sample data, the hydrocarbon source rock integrity evaluation of the region is completed, and the method has important significance on hydrocarbon source rock evaluation and exploration decision guidance.

Description

Technical Field

[0001] The present invention belongs to the technical field of petroleum geological exploration, and in particular relates to a method for seismic prediction of organic carbon content in hydrocarbon source rocks based on a multi-layer perceptron. Background Art

[0002] High-quality source rocks play a decisive role in the formation and distribution of oil and gas reservoirs, and therefore play a crucial role in oil and gas exploration and development. Source rock research encompasses parameters such as organic matter abundance, organic matter type, and organic matter maturity. Key indicators for evaluating organic matter abundance include total organic carbon content (TOC%), chloroform bitumen "A" (%), total hydrocarbon content (HC), and hydrocarbon generation potential (S1+S2 mg / g). In practical production, TOC is often used as a standard parameter for source rock quality, and TOC is positively correlated with the hydrocarbon generation capacity of source rocks.

[0003] TOC is typically obtained through analytical testing. However, early-stage research often faces challenges such as a scarcity of rock samples, poor sampling conditions, and high experimental costs. This results in the use of a small amount of discontinuously distributed TOC data to evaluate source rock quality across an entire region or a specific formation, which inherently introduces significant errors.

[0004] Seismic exploration is generally required in early exploration. Seismic exploration involves excitation of seismic waves on the ground, which are reflected at the subsurface impedance interface. The reflected waves are then received at the surface, carrying information about the subsurface's physical properties. Seismic attribute technology originated in the 1960s and matured in the 1990s. The geological significance of seismic attributes has gradually become clearer, revealing information about lithology, physical properties, and stratigraphic structure contained in seismic data. This has laid the theoretical foundation for predicting the organic carbon content of source rocks using seismic attributes. Summary of the Invention

[0005] In order to solve the above problems, the present invention proposes a method for seismic prediction of organic carbon content in hydrocarbon source rocks based on a multi-layer perceptron.

[0006] The technical solution of the present invention is: A method for seismic prediction of organic carbon content in source rocks based on a multi-layer perceptron is as follows: Obtain organic carbon content data of rock samples; Extract seismic attribute data of the seismic trace near the well where the rock sample is located; Perform correlation analysis on the organic carbon content data of rock samples and the seismic attribute data of wellside seismic traces, and select the preferred seismic attribute data with high correlation; By training the organic carbon content data of rock samples and the preferred seismic attribute data of the multilayer perceptron, a multilayer perceptron model for predicting organic carbon content using seismic attributes was obtained; the preferred seismic attribute data of the entire reservoir in the study area were input into the multilayer perceptron model, and then the organic carbon content data of the entire reservoir in the study area was predicted.

[0007] Before the correlation analysis, the seismic attribute data of the borehole seismic trace in the time domain are first converted to the depth domain, the rock samples in the depth domain are matched with the seismic attribute data of the borehole seismic trace in the time domain, and then the organic carbon content data of the rock samples are correlated with the seismic attribute data of the borehole seismic trace.

[0008] The seismic attribute data of the wellbore seismic trace include instantaneous amplitude Instant Amp, instantaneous phase InstantPhase, instantaneous frequency Instant Freq, weighted instantaneous frequency Instant WeightedFreq, instantaneous bandwidth InstantFreq Band, instantaneous domain frequency Instant Domain Freq, arc length Arc Length, thin layer detector ThinIndicator, half-time energy Energy Half Time, high-order cumulant pseudo-entropy Pseudo_Entropy, high-order cumulant variance Variance, high-order cumulant Skew and high-order cumulant kurtosis Kurtosis.

[0009] The organic carbon content data of the rock sample is obtained by performing pyrolysis analysis on the rock sample.

[0010] The specific process of matching the rock sample in the depth domain with the seismic attribute data of the borehole seismic trace in the time domain is as follows: calculating the wave impedance based on the acoustic time difference logging data and density logging data of the well where the rock sample is located, and then calculating the reflection coefficient sequence; performing a convolution operation on the seismic wavelet and the reflection coefficient sequence to obtain a simulated seismic record, displaying the simulated seismic record and the seismic attribute data of the borehole seismic trace side by side, and adjusting the simulated seismic record by translation until the waveform of the simulated seismic record is optimally matched with the waveform of the seismic attribute data of the borehole seismic trace, that is, achieving borehole seismic calibration, that is, completing the matching of the rock sample in the depth domain with the seismic attribute data of the borehole seismic trace in the time domain.

[0011] The correlation analysis is obtained by calculating the Pearson correlation coefficient. The larger the Pearson correlation coefficient, the higher the correlation. The calculation process of the Pearson correlation coefficient is: Where: r is the Pearson correlation coefficient, dimensionless; X is the earthquake attribute observation value; is the average value of earthquake attribute observations; Y is the observed value of organic carbon content; is the average value of the observed organic carbon content.

[0012] The correlation analysis was performed using the curve fitting module of SPSS software.

[0013] The dependent variable of the multilayer perceptron is the organic carbon content data of the rock sample, and the seismic attribute data of the wellbore seismic trace is the covariate.

[0014] The preferred seismic attribute data are the first four types of wellbore seismic attribute data with the highest to lowest correlation.

[0015] The technical effects of the present invention are: The present invention creatively applies multi-layer perceptron technology to the quantitative calculation of organic carbon content predicted by seismic attribute data, completing the complete process of establishing a multi-layer perceptron model for predicting organic carbon content in source rocks using seismic attributes. The advantages of the present invention are strong operability and good applicability. Based on seismic exploration data widely implemented in the early stage of oil and gas field exploration and a small amount of core sample data, the present invention completes the overall evaluation of regional source rocks, which is of great significance for source rock evaluation and guiding exploration decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is the fitting result diagram of instantaneous amplitude Instant Amp and TOC curve.

[0017] Figure 2 This is the arc length and TOC curve fitting result graph.

[0018] Figure 3 This is the fitting result diagram of instantaneous frequency bandwidth Instant Freq Band and TOC curve.

[0019] Figure 4 This is the fitting result diagram of the instantaneous domain frequency Instant Domain Freq and TOC curve.

[0020] Figure 5 This is the result of the Energy Half Time and TOC curve fitting.

[0021] Figure 6 This is the result of high-order cumulative pseudo-entropy and TOC curve fitting.

[0022] Figure 7 This is the fitting result diagram of instantaneous frequency Instant Freq and TOC curve.

[0023] Figure 8 This is the result of high-order cumulative kurtosis and TOC curve fitting.

[0024] Figure 9 This is the fitting result diagram of instantaneous phase and TOC curve.

[0025] Figure 10 This is the result of the high-order cumulative torsion Skew and TOC curve fitting.

[0026] Figure 11 This is the result of fitting the thin layer detector Thin Indicator and TOC curve.

[0027] Figure 12 This is the result of high-order cumulative variance and TOC curve fitting.

[0028] Figure 13 This is the result of fitting the weighted instantaneous frequency Instant Weighted Freq and TOC curve.

[0029] Figure 14 This is a flowchart of a method for seismic prediction of organic carbon content in source rocks based on a multi-layer perceptron.

[0030] Figure 15 This is the prediction result diagram of the organic carbon content data of the entire reservoir. DETAILED DESCRIPTION

[0031] Example 1 A method for seismic prediction of organic carbon content in source rocks based on a multi-layer perceptron is as follows: Obtain organic carbon content data of rock samples; Extract seismic attribute data of the seismic trace near the well where the rock sample is located; Perform correlation analysis on the organic carbon content data of rock samples and the seismic attribute data of wellside seismic traces, and select the preferred seismic attribute data with high correlation; By training the organic carbon content data of rock samples and the preferred seismic attribute data of the multilayer perceptron, a multilayer perceptron model for predicting organic carbon content using seismic attributes was obtained; the preferred seismic attribute data of the entire reservoir in the study area were input into the multilayer perceptron model, and then the organic carbon content data of the entire reservoir in the study area was predicted.

[0032] Example 2 On the basis of Example 1, the following further aspects are included: Before the correlation analysis, the seismic attribute data of the borehole seismic trace in the time domain are first converted to the depth domain, the rock samples in the depth domain are matched with the seismic attribute data of the borehole seismic trace in the time domain, and then the organic carbon content data of the rock samples are correlated with the seismic attribute data of the borehole seismic trace.

[0033] The specific process of matching the rock sample in the depth domain with the seismic attribute data of the borehole seismic trace in the time domain is as follows: calculating the wave impedance based on the acoustic time difference logging data and density logging data of the well where the rock sample is located, and then calculating the reflection coefficient sequence; performing a convolution operation on the seismic wavelet and the reflection coefficient sequence to obtain a simulated seismic record, displaying the simulated seismic record and the seismic attribute data of the borehole seismic trace side by side, and adjusting the simulated seismic record by translation until the waveform of the simulated seismic record is optimally matched with the waveform of the seismic attribute data of the borehole seismic trace, that is, achieving borehole seismic calibration, that is, completing the matching of the rock sample in the depth domain with the seismic attribute data of the borehole seismic trace in the time domain.

[0034] The correlation analysis is obtained by calculating the Pearson correlation coefficient. The larger the Pearson correlation coefficient, the higher the correlation. The calculation process of the Pearson correlation coefficient is: Where: r is the Pearson correlation coefficient, dimensionless; X is the earthquake attribute observation value; is the average value of earthquake attribute observations; Y is the observed value of organic carbon content; is the average value of the observed organic carbon content.

[0035] The correlation analysis was performed using the curve fitting module of SPSS software.

[0036] Example 3 On the basis of Example 2, it also includes: the seismic attribute data of the wellbore seismic trace includes instantaneous amplitude Instant Amp, instantaneous phase Instant Phase, instantaneous frequency Instant Freq, weighted instantaneous frequency InstantWeighted Freq, instantaneous bandwidth Instant Freq Band, instantaneous dominant frequency Instant Domain Freq, arc length ArcLength, thin layer detector ThinIndicator, half-time energy Energy Half Time, high-order cumulative pseudo-entropy Pseudo_Entropy, high-order cumulative variance Variance, high-order cumulative twist Skew and high-order cumulative kurtosis Kurtosis.

[0037] Example 4 Based on Example 3, the method further includes: obtaining the organic carbon content data of the rock sample by performing pyrolysis analysis on the rock sample; the dependent variable of the multilayer perceptron is the organic carbon content data of the rock sample, and the seismic attribute data of the borehole seismic trace is used as a covariate; and the preferred seismic attribute data is the top four types of borehole seismic attribute data, ranked from highest to lowest correlation.

[0038] Specific experimental cases A method for seismic prediction of organic carbon content in source rocks based on a multi-layer perceptron is as follows.

[0039] Step 1: Sampling source rocks in the study area; The rock samples come from the YHC1 well in the Hari Sag of the Suhongtu Depression in the central part of the Yin'e Basin. There are 45 rock samples. The rock sample stratum is the Lower Cretaceous, and the lithology is mainly dark gray calcareous mudstone and gray dolomitic mudstone.

[0040] Step 2: Pyrolysis analysis of the rock samples was performed by the Geochemistry Laboratory of Yangtze University. The organic carbon content data obtained are shown in Table 1. Table 1 Organic carbon content data of rock samples .

[0041] Step 3: Extract common types of seismic attribute data and extract seismic attribute data of wellside seismic traces; After loading the 3D post-stack seismic data into the seismic processing and interpretation software, thirteen commonly used post-stack seismic attribute data are extracted, namely: Instant Amp, Instant Phase, Instant Frequency, Instant Weighted Freq, Instant Freq Band, Instant Domain Freq, Arc Length, Thin Indicator, Energy Half Time, Pseudo-Entropy, Variance, Skew, and Kurtosis. The rock sample comes from the YHC1 well. A set of seismic attribute data of the YHC1 well's near-well seismic traces are extracted, and thirteen types of post-stack seismic attribute data are obtained.

[0042] Step 4: Convert the seismic attribute data of the borehole seismic trace in the time domain to the depth domain, and match the rock sample in the depth domain with the seismic attribute data of the borehole seismic trace in the time domain; The wave impedance is calculated based on the acoustic time difference logging data and density logging data of Well YHC1, and then the reflection coefficient sequence is calculated. The appropriate seismic wavelet is selected and convolved with the reflection coefficient sequence to obtain a simulated seismic record. The simulated seismic record is displayed side by side with the seismic attribute data of the wellbore seismic trace. The simulated seismic record is adjusted by translation until the waveform of the simulated seismic record and the waveform of the seismic attribute data of the wellbore seismic trace are optimally matched. In other words, wellbore seismic calibration is achieved, and the rock sample in the depth domain is matched with the seismic attribute data of the wellbore seismic trace in the time domain.

[0043] Step 5: Use the curve fitting module of SPSS software to perform correlation analysis between the organic carbon content data of the rock sample and the seismic attribute data of the wellbore seismic trace. The fitting diagram is shown in Figure 2-Figure 14 The correlation statistics of thirteen post-stack seismic attribute data and organic carbon content data of rock samples are shown in Table 2. Table 2 Correlation statistics between thirteen post-stack seismic attribute data and organic carbon content data of rock samples The four items with high correlation are selected as the preferred seismic attribute data: Arc Length, Thin Indicator, Energy Half Time and High-order Cumulative Variance.

[0044] Step 6: Use the multilayer perceptron to train the organic carbon content data of the rock sample and the preferred seismic attribute data to obtain a multilayer perceptron model for predicting organic carbon content using seismic attributes; input the preferred seismic attribute data of the entire reservoir in the study area into the multilayer perceptron model, and then predict the organic carbon content data of the entire reservoir in the study area. The prediction results of the organic carbon content of the well profile are shown as follows: Figure 15 .

[0045] Figure 15 In the figure, dark colors indicate high organic carbon content, which mainly appears at the bottom of the Yingen Formation and the second member of the Bashan Formation, which is consistent with the stratigraphic distribution of the main source rocks in the Hari Sag. At the same time, the white curve shows the organic carbon content on the well, which is consistent with the profile prediction value. The predicted results of organic carbon content have been well verified.

Claims

1. A method for predicting the organic carbon content of source rocks by earthquakes based on a multi-layer perceptron, characterized in that: Here’s how: Obtain organic carbon content data of rock samples; Extract seismic attribute data of the seismic trace near the well where the rock sample is located; Perform correlation analysis on the organic carbon content data of rock samples and the seismic attribute data of wellside seismic traces, and select the preferred seismic attribute data with high correlation; By training the organic carbon content data of rock samples and the preferred seismic attribute data of the multilayer perceptron, a multilayer perceptron model for predicting organic carbon content using seismic attributes was obtained; the preferred seismic attribute data of the entire reservoir in the study area were input into the multilayer perceptron model, and then the organic carbon content data of the entire reservoir in the study area was predicted.

2. The method for predicting organic carbon content of source rocks by earthquakes based on a multi-layer perceptron according to claim 1, characterized in that: Before the correlation analysis, the seismic attribute data of the borehole seismic trace in the time domain are first converted to the depth domain, the rock samples in the depth domain are matched with the seismic attribute data of the borehole seismic trace in the time domain, and then the organic carbon content data of the rock samples are correlated with the seismic attribute data of the borehole seismic trace.

3. The method for predicting the organic carbon content of source rocks by earthquakes based on a multi-layer perceptron according to claim 1, characterized in that: The seismic attribute data of the wellbore seismic trace include instantaneous amplitude, instantaneous phase, instantaneous frequency, weighted instantaneous frequency, instantaneous bandwidth, instantaneous main frequency, arc length, thin layer detection body, half-time energy, high-order cumulant pseudo-entropy, high-order cumulant variance, high-order cumulant torsion and high-order cumulant kurtosis.

4. The method for predicting the organic carbon content of source rocks by earthquakes based on a multi-layer perceptron according to claim 1, characterized in that: The organic carbon content data of the rock sample is obtained by performing pyrolysis analysis on the rock sample.

5. The method for predicting organic carbon content of source rocks by earthquakes based on a multi-layer perceptron according to claim 2, characterized in that: The specific process of matching the rock sample in the depth domain with the seismic attribute data of the borehole seismic trace in the time domain is as follows: calculating the wave impedance based on the acoustic time difference logging data and density logging data of the well where the rock sample is located, and then calculating the reflection coefficient sequence; performing a convolution operation on the seismic wavelet and the reflection coefficient sequence to obtain a simulated seismic record, displaying the simulated seismic record and the seismic attribute data of the borehole seismic trace side by side, and adjusting the simulated seismic record by translation until the waveform of the simulated seismic record is optimally matched with the waveform of the seismic attribute data of the borehole seismic trace, that is, achieving borehole seismic calibration, that is, completing the matching of the rock sample in the depth domain with the seismic attribute data of the borehole seismic trace in the time domain.

6. The method for predicting organic carbon content of source rocks by earthquakes based on a multi-layer perceptron according to claim 5, characterized in that: The correlation analysis is obtained by calculating the Pearson correlation coefficient. The larger the Pearson correlation coefficient, the higher the correlation. The calculation process of the Pearson correlation coefficient is: Where: r is the Pearson correlation coefficient, dimensionless; X is the earthquake attribute observation value; is the average value of earthquake attribute observations; Y is the observed value of organic carbon content; is the average value of the observed organic carbon content.

7. The method for predicting organic carbon content of source rocks by earthquakes based on a multi-layer perceptron according to claim 1, characterized in that: The correlation analysis was performed using the curve fitting module of SPSS software.

8. The method for predicting organic carbon content of source rocks by earthquakes based on a multi-layer perceptron according to claim 1, characterized in that: The dependent variable of the multilayer perceptron is the organic carbon content data of the rock sample, and the seismic attribute data of the wellbore seismic trace is the covariate.

9. The method for predicting organic carbon content of source rocks by earthquakes based on a multi-layer perceptron according to claim 1, characterized in that: The preferred seismic attribute data are the first four types of wellbore seismic attribute data with the highest correlation to the lowest.

Images