Depth domain reservoir parameter prediction method and system, storage medium and equipment

Abstract

The invention provides a depth domain reservoir parameter prediction method and system, a storage medium and equipment, and belongs to the field of geophysical interpretation. According to the method, reservoir parameter prediction learning models are constructed, optimized sensitive body attribute data are used as input, seismic facies are used as constraints, reservoir parameter prediction is carried out by using corresponding learning models under different waveform characteristics, and a prediction result of a three-dimensional space depth domain parameter body is obtained. According to the method, the problem that depth domain wavelets are difficult to solve in a conventional depth domain parameter prediction method is avoided, the learning model is established through the optimized reservoir parameter sensitive attributes, direct parameter prediction is realized under the guidance of geologic features, and the parameter prediction precision is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of geophysical interpretation, and in particular relates to a method, system, storage medium and equipment for predicting reservoir parameters in a depth domain. Background Art

[0002] With the development of prestack depth migration technology and its large-scale application, it is very necessary to explore the technology of deep-domain seismic reservoir parameter prediction. At present, most technical methods focus on how to establish a synthetic seismic record model in the depth domain and clarify the relationship between wells and seismic. Among them, the majority of research focuses on obtaining depth domain wavelets and using convolution methods to produce depth domain synthetic records. However, these methods have two problems. First, the depth domain wavelet is different from the time domain wavelet. It changes with depth and is difficult to obtain. Second, the convolution theory is based on the time domain, and it is still questionable whether it is also applicable to the depth domain. Summary of the invention

[0003] The purpose of the present invention is to solve the problems existing in the above-mentioned prior art and provide a method, system, storage medium and device for predicting reservoir parameters in the depth domain. The invention aims to solve the problem that the mathematical model of deep domain seismic records has not yet been established, develop a nonlinear reservoir parameter prediction technology based on phase control constraints, and realize direct prediction of deep domain reservoir parameters.

[0004] One of the purposes of the present invention is to provide a method for predicting reservoir parameters in depth domain;

[0005] The second object of the present invention is to provide a depth domain reservoir parameter prediction system;

[0006] A third object of the present invention is to provide a computer readable storage medium;

[0007] A fourth object of the present invention is to provide a computer device.

[0008] The present invention is achieved through the following technical solutions:

[0009] The first aspect of the present invention provides a method for predicting reservoir parameters in the depth domain. The method constructs a reservoir parameter prediction learning model, takes the selected sensitive body attribute data as input, takes the seismic phase as a constraint, and uses the phase learning model to predict reservoir parameters under different waveform characteristics to obtain the prediction results of the three-dimensional depth domain parameter body.

[0010] A further improvement of the present invention is:

[0011] The method comprises:

[0012] Step S1, optimizing sensitive body attribute data;

[0013] Step S2, establishing seismic phases;

[0014] Step S3, constructing a reservoir parameter prediction learning model;

[0015] Step S4: predicting reservoir parameters in the depth domain.

[0016] A further improvement of the present invention is:

[0017] In step S1, the sensitive body attribute data is selected, and the specific operations include:

[0018] A method combining expert knowledge selection and correlation analysis is used to screen seismic attributes that are sensitive to seismic facies and reservoir parameters.

[0019] A further improvement of the present invention is:

[0020] The expert knowledge selection is to select the attributes that have the greatest influence on the seismic phase and reservoir parameters according to the expert knowledge;

[0021] The correlation analysis is to select based on the correlation between seismic attribute parameters and seismic phases and reservoir parameters. The correlation formula is:

[0022]

[0023] Where X is a seismic attribute parameter, Y is a seismic phase or reservoir parameter, Cov(X,Y) is the covariance of X and Y, D(X) and D(Y) are the variances of X and Y respectively, and P is the correlation coefficient.

[0024] A further improvement of the present invention is:

[0025] The step S2 of establishing the seismic phase includes: using the SOM neural network algorithm to establish the seismic phase;

[0026] The SOM neural network algorithm training steps are as follows:

[0027] Step S21: Network initialization

[0028] Initialize network weights ω ij (i=1,2,…,n;j=1,2,…m), n is the number of neurons in the input layer, m is the number of neurons in the competition layer, the weights are randomly initialized, and the maximum number of cycles T is set;

[0029] Step S22: Input training samples and normalize

[0030] Calculate the normalized input vector X = (x1, x2, ..., x n ) and the distance d between the competitive layer neuron j j :

[0031] Step S23: Find the winning neuron

[0032] Select the competition layer neuron c with the smallest distance to the input vector as the winning neuron;

[0033] Step S24: Adjust weights

[0034] Each time only the winning neuron and its neighborhood N are adjusted c The weight coefficient of the neuron contained in (t) is adjusted as follows: ij (t+1)=ω ij (t)+η(X i -ω ij ), where t is the number of iterations and η is the learning rate;

[0035] Step S25: Algorithm termination condition

[0036] Determine whether the set maximum number of iterations is met. If so, the algorithm ends; otherwise, return to step S22.

[0037] A further improvement of the present invention is:

[0038] In step S3, a reservoir parameter prediction learning model is constructed, and the specific operations include:

[0039] A nonlinear network prediction model is constructed using the long short-term memory network algorithm. By continuously adjusting and optimizing model parameters, the single-point prediction model is generalized and converged, and then a learning model corresponding to different earthquakes is obtained.

[0040] A further improvement of the present invention is:

[0041] In step S4, the depth domain reservoir parameters are predicted, and the specific operations include:

[0042] The selected reservoir parameter sensitive body attribute data is used as the input of the reservoir parameter prediction learning model, and the seismic phase is used as a constraint. The corresponding learning model is used to predict reservoir parameters under different waveform characteristics, and finally a comprehensive prediction result of the three-dimensional depth domain parameter body is obtained.

[0043] A second aspect of the present invention provides a depth domain reservoir parameter prediction system, comprising:

[0044] Attribute optimization unit, used for sensitive body attribute data;

[0045] A seismic phase establishing unit, used for establishing seismic phases;

[0046] A model building unit, used for building a reservoir parameter prediction learning model;

[0047] The parameter prediction unit is connected with the attribute optimization unit, the seismic phase establishment unit and the model construction unit respectively, and is used to predict the reservoir parameters in the depth domain.

[0048] According to a third aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer executes the steps in the depth domain reservoir parameter prediction method.

[0049] A fourth aspect of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the depth domain reservoir parameter prediction method as described above.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention provides a method for predicting reservoir parameters in the depth domain, which is realized through four steps: selecting sensitive body attribute data, establishing seismic phases, building a reservoir parameter prediction learning model, and directly predicting reservoir parameters in the depth domain. First, the seismic attributes that are sensitive to seismic phases and reservoir parameters are screened by combining expert knowledge selection with correlation analysis to lay a data foundation for subsequent seismic phase and reservoir parameter analysis; secondly, the seismic phase is established by using a self-organizing competitive neural network algorithm; then, by introducing seismic phases, the corresponding learning model is used to predict reservoir parameters under different sedimentary characteristics to improve the prediction accuracy; finally, the corresponding network model is used to directly predict reservoir parameters in the depth domain under different waveform characteristics.

[0052] The present invention avoids the difficulty in obtaining depth domain wavelets in conventional depth domain parameter prediction methods, establishes a learning model by optimizing sensitive attributes of reservoir parameters, realizes direct parameter prediction under the guidance of geological characteristics, and effectively improves parameter prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The present invention provides a flow chart of a method for predicting reservoir parameters in depth domain;

[0054] Figure 2 It is the seismic phase classification result obtained by using the present invention;

[0055] Figure 3a It is the measured curve of the first well and the predicted curve result obtained by the learning model obtained by the present invention;

[0056] Figure 3b It is the measured curve of the second well and the predicted curve result obtained by the learning model obtained by the present invention;

[0057] Figure 3c It is the measured curve of the third well and the predicted curve result obtained by the learning model obtained by the present invention;

[0058] Figure 4 It is the prediction result of the depth domain seismic data obtained by using the present invention;

[0059] Figure 5 It is the depth domain longitudinal wave velocity prediction result obtained by using the present invention. DETAILED DESCRIPTION

[0060] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0061] The present invention provides a method for predicting reservoir parameters in the depth domain. By constructing a reservoir parameter prediction learning model, the selected sensitive body attribute data is used as input, and the seismic phase is used as a constraint. The corresponding learning model is used to predict reservoir parameters under different waveform characteristics to obtain the prediction results of the three-dimensional space depth domain parameter body.

[0062] The embodiments of the method of the present invention are as follows:

[0063] [Example 1]

[0064] like Figure 1 As shown, the prediction method includes:

[0065] Step S1: Optimize sensitive body attribute data:

[0066] The underground strata are distributed in three dimensions. In order to clarify the attributes of each point underground, three-dimensional volume attribute extraction is adopted, which can reveal seismic features that are difficult to distinguish from profile attributes and stratigraphic attributes, and can effectively reflect the lithology, oil and gas content, and fracture development of the reservoir. After obtaining a large number of seismic volume attributes, if multiple attributes are directly used for seismic phase and reservoir prediction research, the amount of calculation will be too large, and the relationship between seismic attributes and seismic phases and reservoir parameters is very complex. In different regions and different reservoirs, the sensitivity of seismic attributes to seismic phases and reservoir parameters is different; even in the same region and the same reservoir, the corresponding sensitive seismic attributes will vary with the difference of seismic phases and reservoir parameters.

[0067] The present invention adopts a method combining expert knowledge selection and correlation analysis to screen seismic attributes that are sensitive to seismic phases and reservoir parameters, laying a data foundation for subsequent seismic phase and reservoir parameter analysis.

[0068] Expert knowledge selection is to select those attributes that have the greatest impact on seismic facies and reservoir parameters based on the expert's knowledge, which requires considerable understanding of the target layer;

[0069] Correlation analysis uses the correlation between seismic attribute parameters and seismic phases and reservoir parameters to make selections. The correlation formula is:

[0070]

[0071] Where X is a seismic attribute parameter, Y is a seismic phase or reservoir parameter, Cov(X,Y) is the covariance of X and Y, D(X) and D(Y) are the variances of X and Y respectively, and P is the correlation coefficient.

[0072] Step S2, establishing seismic phase:

[0073] After the sensitive body attribute data is optimized, the seismic phase division is carried out. At present, the attribute-based phase division method is mainly based on K-means clustering, which can usually only find the local optimal solution, and the biggest disadvantage of the algorithm is that it is very sensitive to the selection of the initial cluster center point. The present invention uses the self-organizing competitive neural network SOM algorithm to establish the seismic phase. The learning of the network is self-organizing learning under unsupervised conditions. In the learning process, different neurons are sensitive to different input patterns through unsupervised competitive learning, so that specific neurons are only sensitive to a certain type of pattern, so that a certain input pattern can be detected.

[0074] The training steps of the SOM neural network algorithm are as follows:

[0075] Step S21: Network initialization

[0076] Initialize network weights ω ij (i=1,2,…,n;j=1,2,…m), n is the number of neurons in the input layer, m is the number of neurons in the competition layer, the weights can be randomly initialized, and the maximum number of cycles T is set;

[0077] Step S22: Input training samples and normalize

[0078] Calculate the normalized input vector X = (x1, x2, ..., x n ) and the distance d between the competitive layer neuron j j :

[0079] Step S23: Find the winning neuron

[0080] Select the competition layer neuron c with the smallest distance to the input vector as the winning neuron;

[0081] Step S24: Adjust weights

[0082] Each time only the winning neuron and its neighborhood N are adjusted c The weight coefficient of the neuron contained in (t) is adjusted by the formula ωij (t+1)=ω ij (t)+η(X i -ω ij ), where t is the number of iterations and η is the learning rate, which generally decreases linearly with the increase of the number of iterations. c The radius of (t) also gradually decreases with the increase of iteration number;

[0083] Step S25: Algorithm termination condition: Determine whether the set maximum number of iterations is met. If so, the algorithm ends; otherwise, return to step S22.

[0084] Step S3: Constructing a reservoir parameter prediction learning model:

[0085] Under different sedimentary environments, the prediction accuracy may be low if the same model is used. By introducing seismic phases and using different network models to predict reservoir parameters under different sedimentary characteristics, the prediction accuracy can be improved.

[0086] The present invention utilizes a long short-term memory network algorithm (LSTM), which has a good time series memory function. Since seismic data has certain time series signal characteristics, and the front and rear sampling points at different depths have a certain correlation, and since the stratum has a certain sedimentary gradient characteristic, the logging data also has a certain correlation in the vertical direction. Therefore, the long short-term memory network algorithm can be used in the direct prediction of reservoir parameters.

[0087] Combining the logging reservoir parameter data under a certain seismic phase and the selected reservoir parameter sensitive seismic multi-attribute data, a nonlinear network prediction model is constructed using the long short-term memory network algorithm. By continuously adjusting and optimizing the model parameters, the single-point prediction model is generalized and converged, and then the learning model corresponding to different earthquakes is obtained.

[0088] (4) Prediction of reservoir parameters in depth domain:

[0089] Taking the selected reservoir parameter sensitive body attribute data as input and the seismic phase as constraint, the corresponding network model is used to predict reservoir parameters under different waveform characteristics, and finally the comprehensive prediction results of the three-dimensional depth domain parameter body are obtained.

[0090] The present invention avoids the difficulty in obtaining depth domain wavelets in conventional depth domain parameter prediction methods, establishes a learning model by optimizing sensitive attributes of reservoir parameters, realizes direct parameter prediction under the guidance of geological characteristics, and effectively improves parameter prediction accuracy.

[0091] The embodiments of the method of the present invention are as follows:

[0092] [Example 2]

[0093] This embodiment takes a certain study area as an example and uses the method provided by the present invention to predict reservoir parameters in the depth domain.

[0094] As described in step S1, sensitive body attribute data is preferred: the underground strata are three-dimensionally distributed. In order to clarify the attributes of each point underground, three-dimensional body attribute extraction is adopted, which can reveal seismic features that are difficult to distinguish from profile attributes and layer attribute images, and can effectively reflect the lithology, oil and gas content, and fracture development of the reservoir. After obtaining a large number of seismic body attributes, if multiple attributes are directly used for seismic phase and reservoir prediction research, the amount of calculation will be too large, and the relationship between seismic attributes and seismic phases and reservoir parameters is very complicated. In different regions and different reservoirs, the sensitivity of seismic attributes to seismic phases and reservoir parameters is different; even in the same region and the same reservoir, the corresponding sensitive seismic attributes will vary with the difference of seismic phases and reservoir parameters.

[0095] This embodiment uses a method combining expert knowledge selection and correlation analysis to screen seismic attributes that are sensitive to seismic phases and reservoir parameters, laying a data foundation for subsequent seismic phase and reservoir parameter analysis.

[0096] Expert knowledge selection is to select those attributes that have the greatest impact on seismic facies and reservoir parameters based on the expert's knowledge, which requires considerable understanding of the target layer;

[0097] Correlation analysis uses the correlation between seismic attribute parameters and seismic phases and reservoir parameters to make selections. The correlation formula is:

[0098] Where X is a seismic attribute parameter, Y is a seismic phase or reservoir parameter, Cov(X,Y) is the covariance of X and Y, D(X) and D(Y) are the variances of X and Y respectively, and P is the correlation coefficient.

[0099] Through attribute sensitivity analysis, envelope attribute, relative wave impedance and instantaneous amplitude attribute are selected as seismic phase sensitive attributes.

[0100] As described in step S2, establish the seismic phase: after the sensitive body attribute data is optimized, carry out the seismic phase division. At present, the attribute-based phase division method is mainly based on K-means clustering, which can usually only find the local optimal solution, and the biggest disadvantage of the algorithm is that it is very sensitive to the selection of the initial cluster center point. This embodiment uses the self-organizing competitive neural network SOM algorithm to establish the seismic phase. The learning of this network is self-organizing learning under unsupervised conditions. During the learning process, different neurons are sensitive to different input patterns through unsupervised competitive learning, so that specific neurons are only sensitive to a certain type of pattern, so that a certain input pattern can be detected.

[0101] The training steps of the SOM neural network algorithm are as follows:

[0102] Step ①: Network initialization. Initialize network weights ω ij (i=1,2,…,n;j=1,2,…m), n is the number of neurons in the input layer, m is the number of neurons in the competition layer, the weights can be randomly initialized, and the maximum number of cycles T is set;

[0103] Step ②: Input training samples and normalize them. Calculate the normalized input vector X = (x1, x2, ..., x n ) and the distance d between the competitive layer neuron j j :

[0104]

[0105] Step 3: Find the winning neuron. Select the competitive layer neuron c with the smallest distance to the input vector as the winning neuron;

[0106] Step ④: Adjust weights. Each time, only the winning neuron and its neighborhood N are adjusted. c The weight coefficient of the neuron contained in (t) is adjusted as follows:

[0107] ω ij (t+1)=ω ij (t)+η(X i -ω ij )

[0108] Among them, t is the number of iterations, and η is the learning rate, which generally decreases linearly with the increase of the number of iterations.

[0109] In the iterative process, the neighborhood N c The radius of (t) also gradually decreases with the increase of iteration number;

[0110] Step ⑤: Algorithm termination condition. Determine whether the set maximum number of iterations is met. If so, the algorithm ends; otherwise, return to step ②.

[0111] like Figure 2 As shown in the figure, it is the classification result of seismic phases in the study area. According to the geological characteristics, the number of clusters is designed to be 3. It can be seen that the classification results develop turbidite fans, river sands and special geological bodies of different periods, which are consistent with the macro geological background and better characterize the characteristics of macro geological zoning.

[0112] As described in step S3, a reservoir parameter prediction learning model is constructed: according to different sedimentary environments, the same model may be used for prediction, which may result in low prediction accuracy. By introducing seismic phases, different networks are used to predict reservoir parameters under different sedimentary characteristics to improve prediction accuracy. This embodiment uses a long short-term memory network algorithm, which has a good time series memory function. Since seismic data has certain time series signal characteristics, and the front and rear sampling points at different depths have certain correlations, and since the stratum has certain sedimentary gradient characteristics, the logging data also has certain correlations in the vertical direction, so the long short-term memory network algorithm can be used in direct prediction of reservoir parameters.

[0113] Combining the logging reservoir parameter data under a certain seismic phase and the selected reservoir parameter sensitive seismic multi-attribute data, the long short-term memory network algorithm is used to build a nonlinear network prediction model. By continuously adjusting and optimizing the model parameters, the single-point prediction model is generalized and converged, and then the learning model corresponding to different earthquakes is obtained. According to the sensitive attribute analysis of the study area and the P-wave velocity, the input data that contributes more to the P-wave prediction is selected, mainly including: seismic traces, envelope attributes, thin layer factors, relative wave impedance, Hilbert attributes, instantaneous frequency, main frequency and instantaneous phase attributes, and the seismic data of a certain type of seismic phase is selected as a constraint to carry out the construction of the P-wave velocity learning network model.

[0114] in, Figure 3a and Figure 3b The black curve in the figure is the measured curve, and the measured curve is used as input to establish the learning model. Figure 3a and Figure 3b The red curve in the figure is the result obtained by using the established learning model (i.e., the prediction result). Figure 3c The black curve in the middle is the measured curve, and the red curve is the predicted result obtained by learning based on the learning model established above. It can be seen that the predicted results are highly consistent with the wells.

[0115] As described in step S4, predicting the reservoir parameters in the depth domain: taking the selected reservoir parameter sensitive body attribute data as input, taking the seismic phase as constraint, using the corresponding network model to predict the reservoir parameters under different waveform characteristics, and finally obtaining the comprehensive prediction results of the three-dimensional space depth domain parameter body. Figure 4 and Figure 5 Shown are the depth domain seismic data and P-wave velocity prediction results obtained using the present invention. From the overall trend, the reservoir parameter prediction profile based on the present invention better retains the overall characteristics and local details of the seismic data, and the degree of consistency with the well is also good.

[0116] The present invention also provides a depth domain reservoir parameter prediction system, and the embodiments of the system are as follows:

[0117] [Example 3]

[0118] The system comprises:

[0119] Attribute optimization unit, used to optimize sensitive attribute data;

[0120] A seismic phase establishing unit, used for establishing seismic phases;

[0121] A model building unit, used for building a reservoir parameter prediction learning model;

[0122] The parameter prediction unit is connected with the attribute optimization unit, the seismic phase establishment unit and the model construction unit respectively, and is used to predict the reservoir parameters in the depth domain.

[0123] The seismic phase establishing unit uses a SOM neural network algorithm to establish the seismic phase;

[0124] The training steps of the SOM neural network algorithm are as follows:

[0125] Step 1: Network initialization

[0126] Initialize network weights ω ij (i=1,2,…,n;j=1,2,…m), n is the number of neurons in the input layer, m is the number of neurons in the competition layer, the weights can be randomly initialized, and the maximum number of cycles T is set;

[0127] Step 2: Input training samples and normalize

[0128] Calculate the normalized input vector X = (x1, x2, ..., x n ) and the distance d between the competitive layer neuron j j :

[0129] Step 3: Find the winning neuron

[0130] Select the competition layer neuron c with the smallest distance to the input vector as the winning neuron;

[0131] Step 4: Adjust weights

[0132] Each time only the winning neuron and its neighborhood N are adjusted c The weight coefficient of the neuron contained in (t) is adjusted as follows: ij (t+1)=ω ij (t)+η(X i -ω ij ), where t is the number of iterations and η is the learning rate;

[0133] Step 5: Algorithm termination condition

[0134] Determine whether the set maximum number of iterations is met. If so, the algorithm ends; otherwise, return to step 2.

[0135] The present invention also provides a computer-readable storage medium, and an embodiment of the computer-readable storage medium is as follows:

[0136] [Example 4]

[0137] The computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer executes the steps in the above-mentioned depth domain reservoir parameter prediction method.

[0138] The present invention also provides a computer device, and an embodiment of the computer device is as follows:

[0139] [Example 5]

[0140] The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the depth domain reservoir parameter prediction method as described above.

[0141] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0142] The above technical solution is only one implementation mode of the present invention. For those skilled in the art, it is easy to make various types of improvements or modifications based on the principles disclosed in the present invention, and it is not limited to the technical solution described in the above specific embodiments of the present invention. Therefore, the above description is only preferred and does not have a restrictive meaning.

Claims

1. A method for predicting reservoir parameters in depth domain, characterized in that: The method constructs a reservoir parameter prediction learning model, takes the selected sensitive body attribute data as input, takes the seismic phase as a constraint, uses the corresponding learning model to predict reservoir parameters under different waveform characteristics, and obtains the prediction results of the three-dimensional space depth domain parameter body.

2. The method for predicting reservoir parameters in depth domain according to claim 1, characterized in that: The method comprises: Step S1, optimizing sensitive body attribute data; Step S2, establishing seismic phases; Step S3, constructing a reservoir parameter prediction learning model; Step S4: predicting reservoir parameters in the depth domain.

3. The method for predicting reservoir parameters in depth domain according to claim 2, characterized in that: In step S1, the sensitive body attribute data is selected, and the specific operations include: A method combining expert knowledge selection and correlation analysis is used to screen seismic attributes that are sensitive to seismic facies and reservoir parameters.

4. The method for predicting reservoir parameters in depth domain according to claim 3, characterized in that: The expert knowledge selection is to select the attributes that have the greatest influence on the seismic phase and reservoir parameters according to the expert knowledge; The correlation analysis is to select based on the correlation between seismic attribute parameters and seismic phases and reservoir parameters. The correlation formula is: Where X is a seismic attribute parameter, Y is a seismic phase or reservoir parameter, Cov(X,Y) is the covariance of X and Y, D(X) and D(Y) are the variances of X and Y respectively, and P is the correlation coefficient.

5. The method for predicting reservoir parameters in depth domain according to claim 2, characterized in that: The step S2 of establishing the seismic phase includes: using the SOM neural network algorithm to establish the seismic phase; The SOM neural network algorithm training steps are as follows: Step S21: Network initialization Initialize network weights ω ij (i=1,2,…,n;j=1,2,…m), n is the number of neurons in the input layer, m is the number of neurons in the competition layer, the weights are randomly initialized, and the maximum number of cycles T is set; Step S22: Input training samples and normalize Calculate the normalized input vector X = (x1, x2, ..., x n ) and the distance d between the competitive layer neuron j j : Step S23: Find the winning neuron Select the competition layer neuron c with the smallest distance to the input vector as the winning neuron; Step S24: Adjust weights Each time only the winning neuron and its neighborhood N are adjusted c The weight coefficient of the neuron contained in (t) is adjusted as follows: ij (t+1)=ω ij (t)+η(X i -ω ij ), where t is the number of iterations and η is the learning rate; Step S25: Algorithm termination condition Determine whether the set maximum number of iterations is met. If so, the algorithm ends; otherwise, return to step S22.

6. The method for predicting reservoir parameters in depth domain according to claim 2, characterized in that: In step S3, a reservoir parameter prediction learning model is constructed, and the specific operations include: A nonlinear network prediction model is constructed using the long short-term memory network algorithm. By continuously adjusting and optimizing model parameters, the single-point prediction model is generalized and converged, and then a learning model corresponding to different earthquakes is obtained.

7. The method for predicting reservoir parameters in depth domain according to claim 2, characterized in that: In step S4, the depth domain reservoir parameters are predicted, and the specific operations include: The selected reservoir parameter sensitive body attribute data is used as the input of the reservoir parameter prediction learning model, and the seismic phase is used as a constraint. The corresponding learning model is used to predict reservoir parameters under different waveform characteristics, and finally a comprehensive prediction result of the three-dimensional depth domain parameter body is obtained.

8. A depth domain reservoir parameter prediction system, characterized in that: include: Attribute optimization unit, used to optimize sensitive attribute data; A seismic phase establishing unit, used for establishing seismic phases; A model building unit, used for building a reservoir parameter prediction learning model; The parameter prediction unit is connected with the attribute optimization unit, the seismic phase establishment unit and the model construction unit respectively, and is used to predict the reservoir parameters in the depth domain.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer executes the steps in the depth domain reservoir parameter prediction method according to any one of claims 1 to 7.

10. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the depth domain reservoir parameter prediction method according to any one of claims 1 to 7.

Images