Seismic quantitative prediction method of shale TOC based on sensitive parameter volume

Abstract

The present invention discloses a shale TOC seismic quantitative prediction method based on a sensitive parameter body. For a target layer whose TOC content is to be measured, a well logging curve with a high correlation with the TOC content is analyzed and found as a sensitive parameter; sample data is constructed using the sensitive parameter; the sample data is used as input, and the TOC content at the depth corresponding to the sample data is used as output, and an RBF neural network prediction model is obtained by training; the sensitive parameter and post-stack three-dimensional seismic data are then inverted to obtain a sensitive parameter body, and a prediction sample is constructed using the sensitive parameter body, and the prediction sample is input into the prediction model to calculate the corresponding TOC value, thereby realizing the prediction of the TOC content of the target layer. The present invention overcomes the limitation of the prior art that only a few elastic parameters and seismic attributes of pre-stack inversion can be used for seismic prediction of TOC content, as well as the disadvantage of not being widely used. More parameters with a high correlation with the TOC content can be used to realize high-precision prediction of the shale TOC content, and the scope of application is wider and the prediction is easier to realize.

Description

Technical Field

[0001] The invention relates to an earthquake prediction method, and in particular to a shale TOC earthquake quantitative prediction method based on a sensitive parameter body. Background Art

[0002] In recent years, due to the rapid development of unconventional oil and gas exploration and development, shale reservoir evaluation has received widespread attention worldwide. Total organic carbon (TOC) is a key indicator for evaluating the abundance of organic matter and hydrocarbon generation capacity of source rocks. Its content has many effects on shale gas reservoirs. TOC content has a great influence on the porosity, wettability, permeability and microstructure of shale reservoirs; in addition, organic carbon content is also an important factor affecting matrix porosity and controlling gas adsorption in shale gas reservoirs; organic carbon, as a component of the rock matrix, its content also strongly affects the geomechanical properties of shale. Therefore, an accurate and reliable shale TOC prediction method plays a very important role in the exploration and development of unconventional oil and gas reservoirs.

[0003] At present, many logging prediction methods are based on the quantitative relationship between resistivity, porosity and TOC content, which can achieve high-accuracy TOC content prediction. However, it is difficult to use such methods for seismic quantitative prediction of TOC, because it is difficult for seismic prediction methods to establish a close correlation between seismic data and TOC content, like logging parameters such as resistivity and porosity. The seismic parameters that can be used for prediction at present include seismic attributes such as amplitude, instantaneous attributes, and frequency, as well as prestack inversion parameters related to elasticity such as density and wave impedance, λρ, and μρ. However, the response mechanism between most of these seismic attributes and TOC content is not clear, and the correlation mechanism between different seismic attributes and organic carbon characteristics needs to be studied in depth to ensure the rationality of the results. At the same time, prestack elastic parameters are not easy to obtain and need to be obtained through prestack inversion, rock physics, etc. In addition, under complex geological conditions, using only a few elastic parameters for seismic prediction of TOC content will be limited by data conditions and not extensive. Summary of the invention

[0004] The purpose of the present invention is to provide a method for solving the above problems, using multiple parameters to predict TOC content, and having a fast prediction speed and high accuracy for shale TOC seismic quantitative prediction based on sensitive parameter bodies.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows: a shale TOC seismic quantitative prediction method based on sensitive parameter body, comprising the following steps;

[0006] (1) determining a target layer in a formation whose TOC content is to be measured, obtaining well logging data and post-stack three-dimensional seismic data of the target layer, determining M depths at equal intervals on the target layer, and obtaining the TOC content of the core at each depth, wherein the well logging data includes a plurality of well logging curves;

[0007] (2) performing a correlation analysis between each well logging curve and the TOC content to obtain a correlation coefficient between each well logging curve and the TOC content, presetting a threshold, and retaining well logging curves with correlation coefficients greater than the threshold as sensitive parameters, wherein the number of sensitive parameters is N and they are marked as the 1st to the Nth;

[0008] (3) Construct sample data;

[0009] A sample data is constructed at each depth of the target layer, and the sample data at the jth depth is L j ={L 1j , L 2j , …, L ij , …, L Nj}; where L ij Represents the value of the i-th sensitive parameter at the j-th depth, i = 1 to N, j = 1 to M;

[0010] (4) Establishing an RBF neural network, taking the sample data as input and the TOC content at the depth corresponding to the sample data as output, and training to obtain the RBF neural network prediction model;

[0011] (5) For the 1st to Nth sensitive parameters, each sensitive parameter is used as a constraint, and the corresponding sensitive parameter volume is obtained by using the post-stack 3D seismic data inversion method, which is marked as the 1st to Nth sensitive parameter volume;

[0012] (6) constructing a prediction sample, including steps (61)-(65);

[0013] (61) Each sensitive parameter volume has the same size, and its line number, channel number, and sampling point are P, Q, and H respectively. The whole volume is a three-dimensional data volume of P×Q×H;

[0014] (62) For the first sensitive parameter body, it is organized into a K×H two-dimensional matrix, where K=P×Q, and then converted into a 1×L one-dimensional array, where L=K×H;

[0015] (63) Processing the second to Nth sensitive parameter bodies into one-dimensional arrays in sequence; and in each one-dimensional array, the coordinate values ​​corresponding to the elements at the same position are the same;

[0016] (64) The one-dimensional vectors corresponding to the first to Nth data bodies are used as the first to Nth row data of the matrix to form an N×L prediction matrix;

[0017] (65) Extract each column of the prediction matrix to form a prediction sample, and obtain L prediction samples in total;

[0018] (7) sequentially sending L prediction samples into the RBF neural network prediction model, outputting L TOC values, and for each TOC value, using the coordinates of the corresponding prediction sample as its coordinates to obtain a TOC value with coordinates; the coordinates are the line number, track number and sampling point;

[0019] (8) The obtained one-dimensional TOC array is transformed to form a three-dimensional TOC data body of P×Q×H according to the coordinates to complete the TOC content prediction of the target layer.

[0020] Preferably, the logging curves include curves of density, acoustic wave time difference, porosity, resistivity, potassium content and uranium content, and the TOC content of the core is obtained by geochemical analysis of the total organic carbon content of the shale core.

[0021] Preferably, the correlation analysis in step (2) is performed using SPSS statistical analysis software.

[0022] About sensitive parameters and sensitive parameter bodies:

[0023] An article published in the journal Natural Gas Geoscience in March 2019 titled “Seismic quantitative prediction method and application of TOC content in shales of the Wufeng Formation and Longmaxi Formation in Weiyuan area, Sichuan Basin” mentioned TOC sensitive parameters and how to combine TOC sensitive parameters with seismic data to obtain a three-dimensional sensitive parameter body.

[0024] In the present invention, the RBF neural network consists of an input layer, a hidden layer, and an output layer, and has the significant advantages of optimal fitting performance and global fitting. It is very suitable for fitting nonlinear data, and can train a complex nonlinear relationship network model between TOC and other multiple parameters, improve TOC prediction accuracy, and realize shale TOC seismic quantitative prediction.

[0025] When using post-stack 3D seismic data to invert sensitive parameter volumes with each sensitive parameter as a constraint, well logging data and seismic data are needed. However, well logging is in the depth domain, while seismic data is in the time domain. Therefore, it is necessary to convert them through synthetic records to find the correct time-depth relationship to ensure that they correspond to each other at the same depth, and to ensure that the inversion results are accurate and will not have large deviations. Here, seismic phase-controlled inversion is to establish a mapping relationship between seismic waveform structure and well logging curve structure through dynamic clustering analysis of seismic waveforms, and to achieve high-resolution seismic waveform indication simulation of non-wave impedance curves such as natural gamma, so as to obtain the sensitive parameter volume required for prediction.

[0026] For multiple sensitive parameter bodies, we split each sensitive parameter body in the order of body, surface, and line, and finally split each three-dimensional data body into a one-dimensional array. Each element in the one-dimensional array corresponds to the line number, channel number, and sampling point before the split. These one-dimensional arrays are then used to construct prediction samples for prediction, and a set of one-dimensional arrays about TOC content is obtained, which are then reconstructed into three-dimensional data bodies in the order of line, surface, and body.

[0027] Compared with the prior art, the advantages of the present invention are:

[0028] (1) The present invention overcomes the limitation of the prior art that only a few elastic parameters and seismic attributes can be used to predict TOC content, as well as the defect of lack of extensiveness. Multiple parameters with high correlation with TOC content are used together for prediction, and more sensitive parameter bodies can be obtained. Therefore, compared with the prior art, the present invention can use more parameters with high correlation with TOC content to achieve high-precision prediction of shale TOC content, with a wider range of application and more accurate prediction. Easy to implement .

[0029] (2) By adopting the neural network approach, the nonlinear fitting of input sensitive parameters and output TOC content is achieved, which has higher prediction accuracy than traditional modeling methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of the present invention;

[0031] Figure 2 It is a schematic diagram of sample data composition in the target layer;

[0032] Figure 3a It is a comparison chart of the measured TOC value and the predicted TOC value of the training data;

[0033] Figure 3b It is a comparison chart of the TOC measured value and the TOC predicted value of the test data;

[0034] Figure 4 It is the intersection analysis diagram of the TOC measured value and the TOC predicted value of the training data;

[0035] Figure 5 It is the intersection analysis diagram of the TOC measured value and the TOC predicted value of the test data;

[0036] Figure 6 This is a well-connected cross-sectional diagram of the predicted TOC content in the study area in Example 2 of the present invention;

[0037] Figure 7 It is the predicted dragon-1 subsegment S111 of the present invention. 1 The average TOC content plane diagram;

[0038] Figure 8 It is the Long-2 subsegment S111 predicted by the present invention. 2 Plane diagram of average TOC content. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the accompanying drawings.

[0040] Example 1: See Figure 1-Figure 5 , a shale TOC seismic quantitative prediction method based on sensitive parameter body, comprising the following steps;

[0041] (1) determining a target layer in a formation whose TOC content is to be measured, obtaining well logging data and post-stack three-dimensional seismic data of the target layer, determining M depths at equal intervals on the target layer, and obtaining the TOC content of the core at each depth, wherein the well logging data includes a plurality of well logging curves;

[0042] (2) performing a correlation analysis between each well logging curve and the TOC content to obtain a correlation coefficient between each well logging curve and the TOC content, presetting a threshold, and retaining well logging curves with correlation coefficients greater than the threshold as sensitive parameters, wherein the number of sensitive parameters is N and they are marked as the 1st to the Nth;

[0043] (3) Construct sample data;

[0044] A sample data is constructed at each depth of the target layer, and the sample data at the jth depth is L j ={L 1j , L 2j , …, L ij , …, L Nj}; where L ij Represents the value of the i-th sensitive parameter at the j-th depth, i = 1 to N, j = 1 to M;

[0045] (4) Establishing an RBF neural network, taking the sample data as input and the TOC content at the depth corresponding to the sample data as output, and training to obtain the RBF neural network prediction model;

[0046] (5) For the 1st to Nth sensitive parameters, each sensitive parameter is used as a constraint, and the corresponding sensitive parameter volume is obtained by using the post-stack 3D seismic data inversion method, which is marked as the 1st to Nth sensitive parameter volume;

[0047] (6) constructing a prediction sample, including steps (61)-(65);

[0048] (61) Each sensitive parameter volume has the same size, and its line number, channel number, and sampling point are P, Q, and H respectively. The whole volume is a three-dimensional data volume of P×Q×H;

[0049] (62) For the first sensitive parameter body, it is organized into a K×H two-dimensional matrix, where K=P×Q, and then converted into a 1×L one-dimensional array, where L=K×H;

[0050] (63) Processing the second to Nth sensitive parameter bodies into one-dimensional arrays in sequence; and in each one-dimensional array, the coordinate values ​​corresponding to the elements at the same position are the same;

[0051] (64) The one-dimensional vectors corresponding to the first to Nth data bodies are used as the first to Nth row data of the matrix to form an N×L prediction matrix;

[0052] (65) Extract each column of the prediction matrix to form a prediction sample, and obtain L prediction samples in total;

[0053] (7) sequentially sending L prediction samples into the RBF neural network prediction model, outputting L TOC values, and for each TOC value, using the coordinates of the corresponding prediction sample as its coordinates to obtain a TOC value with coordinates; the coordinates are the line number, track number and sampling point;

[0054] (8) The obtained one-dimensional TOC array is transformed to form a three-dimensional TOC data body of P×Q×H according to the coordinates to complete the TOC content prediction of the target layer.

[0055] The logging curves include curves of density, acoustic wave time difference, porosity, resistivity, potassium content and uranium content. The TOC content of the core is obtained by geochemical analysis of the total organic carbon content of the shale core.

[0056] The correlation analysis in step (2) is specifically performed using SPSS statistical analysis software.

[0057] Wherein: In step (1), assuming that the target layer is located at a depth of 2200-2300 meters underground, the total thickness of the target layer is 100 meters. We sample at equal intervals of 1 meter, so the number of sampling points at the depth is 100.

[0058] In step (2), based on experience, we can set the threshold value at above 0.4 or 0.5. There are dozens to hundreds of logging curves in the logging data. It is a large workload to perform correlation analysis on each logging curve and TOC content. In order to simplify the calculation amount, we can preliminarily screen a batch of logging curves related to TOC content based on experience, and then use the preliminarily screened logging curves to perform correlation analysis with TOC content, and retain the logging curves with correlation coefficients greater than the threshold as sensitive parameters. Assume that in this embodiment, we analyze and obtain the following logging curves with high correlation with TOC content, including: density curve, unit is g / cm3; acoustic wave time difference curve, unit is μs / ft; resistivity curve, unit is Ω·m; porosity curve, unit is %; potassium content curve, unit is %; uranium content curve, unit is μg / g; the correlations are 0.871, 0.709, 0.491, 0.637, 0.758, and 0.707, respectively. Then N = 6, and the above 6 well logging curves are marked as No. 1 to No. 6 respectively;

[0059] In step (3), the sample data at the first depth L1 = {L 11 , L 21 , L 31 , L 41 , L 51 , L 61 ,};L 11 -L 61 They represent the values ​​of density curve, acoustic time difference curve, resistivity curve, porosity curve, potassium content curve and uranium content curve at the depth respectively. The sample data of other depths can be obtained in the same way.

[0060] In step (4), the RBF neural network structure is a three-layer feedforward neural network with a single hidden layer, which is a local approximation network; it consists of three layers: input layer, hidden layer, and output layer. The input layer is the bridge connecting the external environment and the network; the hidden layer usually has a higher dimension and has the function of mapping vectors from low dimensions to high dimensions, which can make nonlinear data that is linearly inseparable in low dimensions become linearly separable in high dimensions; the third layer is the output layer, and the hidden layer space is linearly mapped to the output layer space.

[0061] In order to obtain more sample data, we selected five wells in a shale exploration area and collected data from the five wells to construct sample data, and then all the sample data constituted a data set. Then the sample data in the data set was divided into training data and test data. In actual operation, we used the data of four wells as training data, accounting for 80%, to create a neural network model, and the data of another well as test data, accounting for 20%, to verify the reliability of the model. During training, the sample data is used as input, and the TOC content at the corresponding depth of the sample data is output. The optimal neural network model is obtained by adjusting the number of neurons and the expansion constant of the neural network for training, which is used as the RBF neural network prediction model to be used in the next step; the expansion constant can be determined according to the dispersion of the data center. The farther the input of the neuron is from the center of the radial basis function, the lower the activation degree of the neuron.

[0062] When obtaining training data or test data, we use the measured TOC content of the core as the TOC measured value, and then the TOC content corresponding to the output of the prediction model is called the TOC predicted value. We compare the TOC measured value of the training data with the TOC predicted value obtained by the prediction model to obtain Figure 3a , the TOC measured value of the test data is compared with the TOC predicted value obtained by the prediction model Figure 3b As can be seen from the results in the figure, the predicted values ​​are very close to the measured values, with high prediction accuracy. The prediction results of the test wells show that the model is stable and reliable, with strong and stable prediction capabilities.

[0063] We perform an intersection analysis on the TOC measured value and the TOC predicted value of the training data and obtain Figure 4 , Figure 4 In the , the correlation between the measured TOC value and the predicted TOC value of the training data reached 0.923;

[0064] Similarly, the TOC measured value of the test data is analyzed with the TOC predicted value to obtain Figure 5 , Figure 5 In the test data, the correlation between the measured TOC value and the predicted TOC value reached 0.936.

[0065] Example 2: See Figures 6 to 8 Based on Example 1, a specific application of the method of the present invention is given. We selected a shale gas block in southern Sichuan as the research area, and took the Lower Silurian Longmaxi Formation shale as the research target to predict its TOC content. This study mainly focuses on the first section of the Longmaxi Formation S1l1 and the second section of the Longmaxi Formation S1l2. The first section of the Longmaxi Formation is further subdivided into the Long 1 1 subsection S1l1. 1 Helong 12 subsection S1l1 2 .

[0066] The TOC content prediction of the well section in the study area is as follows Figure 6 As shown. S1l1 1 The TOC content of the section is relatively high, mainly between 1% and 5%, and shows a trend of high at the bottom and low at the top in the vertical direction. The TOC content at the bottom can be as high as 4% or above, and the TOC content decreases upward to 1% to 3%; in the lateral direction, the TOC distribution heterogeneity is relatively small, the overall TOC value is high, and the distribution is relatively stable, but there are local changes.

[0067] S1l1 2 The TOC content of the section is relatively low, less than 2% overall, and only in the local area at the bottom the TOC content is greater than 2%. The TOC content varies horizontally, fluctuating and showing an unstable state.

[0068] The TOC content of S1l2 section is low, with the TOC content in almost most areas below 1%, and only in some local areas above 1% and below 5%. The TOC content varies slightly in the horizontal direction.

[0069] Figure 7 For S1l1 1 The average TOC content of the segment is shown in the planar distribution diagram. 1 The TOC content is relatively high, with the southeast being high and the northwest being relatively low. It is distributed in a northeast strip shape, with the average value reaching up to 5%. The TOC content around wells B and E is relatively high, while the TOC content near wells A, C, and D is relatively low.

[0070] Figure 8 For S1l1 2 The TOC content plane distribution diagram shows low TOC content overall. The TOC content well profile prediction diagram also shows that S1l1 2 The TOC content in most areas is at low values, resulting in a low average TOC content in this section.

[0071] From the plane prediction map, we can see that the high TOC areas in this section are mainly distributed in the northeast and southwest, with an average value of approximately 2-3%, and other areas are almost all below 2%.

[0072] Through analysis, it is believed that S1l1 in the southeast of the study area 1 The TOC distribution heterogeneity is small, the average content is high, and the distribution range is wide, making it a favorable area for shale gas exploration and development in this area.

[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A shale TOC seismic quantitative prediction method based on sensitive parameter bodies, characterized by: The steps include: (1) determining a target layer in a formation whose TOC content is to be measured, obtaining well logging data and post-stack three-dimensional seismic data of the target layer, determining M depths at equal intervals on the target layer, and obtaining the TOC content of the core at each depth, wherein the well logging data includes a plurality of well logging curves; (2) performing a correlation analysis between each well logging curve and the TOC content to obtain a correlation coefficient between each well logging curve and the TOC content, presetting a threshold, and retaining the well logging curves with a correlation coefficient greater than the threshold as sensitive parameters. The number of sensitive parameters is N, and they are marked as the 1st to the Nth; (3) Construct sample data; A sample data is constructed at each depth of the target layer, and the sample data at the jth depth is L j ={L 1j , L 2j , … , L ij , … , L N j }; where L ij Represents the value of the i-th sensitive parameter at the j-th depth, i=1~N, j=1~M; (4) Establishing an RBF neural network with sample data as input and the TOC content at the depth corresponding to the sample data as output, and training to obtain the RBF neural network prediction model; (5) For the 1st to Nth sensitive parameters, each sensitive parameter is used as a constraint, and the corresponding sensitive parameter volume is obtained by using the post-stack 3D seismic data inversion method, which is marked as the 1st to Nth sensitive parameter volume; (6) constructing a prediction sample, including steps (61)-(65); (61) Each sensitive parameter volume has the same size, and its line number, channel number and sampling point are P, Q, and H respectively. The whole volume is a three-dimensional data volume of P×Q×H; (62) For the first sensitive parameter body, organize it into a K×H two-dimensional matrix, where K=P×Q, and then convert it into a 1×L one-dimensional array, where L=K×H; (63) Process the second to Nth sensitive parameter bodies into one-dimensional arrays in sequence; and in each one-dimensional array, the coordinate values ​​corresponding to the elements at the same position are the same; (64) The one-dimensional vectors corresponding to the first to Nth data bodies are used as the first to Nth row data of the matrix, respectively, to form an N×L prediction matrix; (65) Extract each column of the prediction matrix to form a prediction sample, and obtain L prediction samples in total; (7) L prediction samples are sequentially sent to the RBF neural network prediction model, and L TOC values ​​are output. For each TOC value, the coordinates of the corresponding prediction sample are used as its coordinates to obtain a TOC value with coordinates; the coordinates are the line number, track number and sampling point; (8) The obtained one-dimensional TOC array is transformed to form a three-dimensional TOC data volume of P×Q×H according to the coordinates to complete the TOC content prediction of the target layer.

2. The shale TOC seismic quantitative prediction method based on sensitive parameter volume according to claim 1 is characterized by: The logging curves include density, acoustic wave time difference, porosity, resistivity, potassium content and uranium content curves, and the TOC content of the core is obtained by geochemical analysis of the total organic carbon content of the shale core.

3. The shale TOC seismic quantitative prediction method based on sensitive parameter volume according to claim 1 is characterized by: The correlation analysis in step (2) is specifically performed using SPSS statistical analysis software.

Images