A deep learning-based resistivity model to velocity model method and system

Abstract

The present application relates to the technical field of oil and gas exploration, and particularly relates to a resistivity model to velocity model method and system based on deep learning. The method comprises generating random samples containing resistivity, formation depth and formation wave velocity through a variant of the Faust formula, training a neural network using the random samples, adjusting network parameters, constructing a BP network model capable of mapping each other between resistivity, formation depth and formation wave velocity, obtaining a measured data set to train the BP network model to obtain a resistivity to velocity model, obtaining resistivity values and corresponding formation depths at any regional position in the work area, and inputting the resistivity to velocity model to obtain a predicted acoustic wave velocity model. The present application can accurately output an acoustic wave velocity model of the full region of the underground surface layer, has higher flexibility, accuracy and adaptability compared to traditional methods, and can cope with deeper surface layer complex structures, thereby providing support for seismic data static correction and velocity modeling.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration technology, and in particular to a method and system for resistivity and velocity modeling based on deep learning. Background Technology

[0002] Electromagnetic exploration is a geophysical exploration method that utilizes the electromagnetic and electrochemical differences in subsurface media. By analyzing the distribution patterns and temporal characteristics of natural or artificial electromagnetic fields, it aims to locate different types of useful mineral deposits, identify geological structures, and solve geological problems. In oil and gas exploration, seismic exploration, with its higher precision, is primarily used. However, seismic processing is challenging in areas with complex surface structures and lithologies. Three-dimensional high-frequency electromagnetic methods can identify and characterize lithological bodies at higher resolution for shallow layers, aiding in three-dimensional seismic velocity modeling. Crucially, in constructing a three-dimensional seismic velocity model, the conversion from the resistivity model to the velocity model is critical. If this step fails, even with accurate resistivity inversion, the resulting velocity model will still contain significant errors, impacting final seismic data processing.

[0003] Current resistivity and velocity models employ the Faust formula, an empirical formula characterizing the statistical relationship between formation resistivity and acoustic velocity curves under depth constraints. However, the Faust formula is not applicable to all formations. This is because, in addition to being influenced by lithology, fluids within the formation have a greater impact on resistivity than acoustic waves. This limits the ability of the original Faust formula to construct a mapping relationship. When shallow anomalies reach depths of several thousand meters, it becomes difficult to fit a correct relationship, making curve reconstruction using the original Faust formula only applicable to formations where resistivity and acoustic velocity have a good statistical relationship. Summary of the Invention

[0004] The technical problem to be solved by the embodiments of the present invention is to provide a method and system for resistivity model to velocity model based on deep learning, so as to solve the problem that empirical formulas in the prior art are difficult to fit the relationship between resistivity model and velocity model for deep surface complex structures.

[0005] This invention discloses a deep learning-based resistivity-speed model method, comprising: Random samples containing resistivity, formation depth, and formation wave velocity are generated using a variation of the Faust formula. These random samples are then used to train a neural network, constructing a backpropagation (BP) network model that maps resistivity, formation depth, and formation wave velocity to each other. The variation of the Faust formula is as follows:

[0006] In the formula, Indicates the ground wave velocity; Represents resistivity; Indicates the depth of the strata. These represent the coefficient vectors of the corresponding terms; Obtain a measured dataset containing the resistivity at the well location, as well as the corresponding formation depth and formation wave velocity. Train the BP network model based on the measured dataset to obtain a resistivity-to-velocity model. Obtain the resistivity model and corresponding formation depth of any location within the work area obtained by electrical inversion, and input them into the resistivity rotational velocity model to obtain the predicted acoustic velocity model.

[0007] Optionally, generating a random sample containing resistivity, formation depth, and formation wave velocity using a variation of the Faust formula includes: Determine the range of geological surface resistivity, formation depth, and formation wave velocity; Select resistivity, formation depth, and formation wave velocity that meet a certain range, and establish the mapping relationship between resistivity, formation depth, and formation wave velocity through a variation of the Faust formula. The dataset containing resistivity, formation depth, and formation wave velocity after establishing the mapping relationship is randomly divided into the first training sample and the test sample.

[0008] Optionally, the step of training a neural network using random samples to construct a BP network model that allows for mutual mapping between resistivity, formation depth, and formation wave velocity includes: A neural network framework based on a backpropagation network is constructed, with the resistivity and formation depth in the first training sample as the input layer and the formation wave velocity in the first training sample as the output layer, and the neural network framework is trained. The trained neural network framework is tested using test samples, and the structure, model parameters, required number of samples, and input and output data formats of the neural network framework are determined based on the test results to obtain the BP network model.

[0009] Optionally, the construction of the neural network framework based on the BP network includes: Based on the BP network, a neural network framework consisting of an input layer, an output layer, and four hidden layers is constructed. A hyperbolic tangent activation function is selected for each hidden layer, and a linear activation function is selected for the output layer.

[0010] Optionally, the step of training the BP network model based on the measured dataset to obtain the resistivity-transfer model includes: The measured dataset containing resistivity, formation depth, and formation wave velocity was randomly divided into a second training sample and a validation sample. The resistivity and formation depth in the second training sample are used as input data, and the formation wave velocity in the second training sample is used as output data. The BP network model is then trained to obtain the resistivity-velocity model. The resistivity in the second training sample and the validation sample, along with the corresponding formation depth, are input into the resistivity rotational velocity model to output the predicted formation wave velocity. The error of the prediction result is obtained by comparing the output predicted formation wave velocity with the corresponding formation wave velocities in the second training sample and the validation sample. The predictive performance of the resistivity-rotational model is verified based on the error between the prediction results of the measured training samples and the measured verification samples.

[0011] Optionally, obtaining the measured dataset including well location resistivity, corresponding formation depth, and formation wave velocity includes: Obtain vertical seismic profile data of the target well within the work area, and analyze the vertical seismic profile data to obtain the formation wave velocity from the shallow to the deep surface. Acquire three-dimensional high-frequency electromagnetic data of the target well within the work area, and obtain the resistivity corresponding to the location of the target well by inverting the three-dimensional high-frequency electromagnetic data; The resistivity obtained from the inversion and the formation wave velocity obtained from the analysis are interpolated to correspond to the same formation depth, resulting in a measured dataset that includes the resistivity at the well location, as well as the corresponding formation depth and formation wave velocity.

[0012] Optionally, the method includes a method for processing data after mapping the resistivity and the formation wave velocity to the same formation depth. Based on the logarithmic domain of the resistivity, the formation wave velocity and the formation depth are uniformly converted to the logarithmic domain; The resistivity, formation wave velocity, and formation depth in the logarithmic domain are normalized to obtain the measured dataset used to train the BP network model. The normalization function formula is as follows:

[0013] In the formula, This indicates the resistivity, formation depth, or formation wave velocity converted to the logarithmic domain. This represents the maximum value of resistivity, formation depth, or absolute value of formation wave velocity. This represents the normalized resistivity, formation depth, or formation wave velocity.

[0014] Optionally, the step of obtaining the resistivity model and corresponding formation depth of any location within the work area obtained by electrical inversion, and inputting them into the resistivity rotational velocity model to convert them into a predicted acoustic velocity model includes: A sideline with complex shallow structures was selected within the work area, and a resistivity model was obtained from the profile of the selected sideline through inversion. Based on the resistivity model, the resistivity corresponding to the preset formation depth is estimated by interpolation. The interpolated resistivity and the corresponding formation depth are used as input data into the resistivity-to-velocity model to obtain the acoustic velocity model of the selected side profile.

[0015] Optionally, this includes methods for processing the input and output data of the resistivity-transport model: The interpolated resistivity and the corresponding formation depth are converted into the logarithmic domain. The resistivity and the corresponding formation depth in the logarithmic domain are normalized to obtain the input data for the resistivity-to-velocity model. The predicted formation wave velocity output from the resistivity rotational velocity model is inversely normalized to obtain the acoustic velocity model for the selected side profile.

[0016] This invention also discloses a system for a resistivity-rotational model, employing the aforementioned resistivity-rotational model method. The system includes: The BP network model building module is used to generate random samples containing resistivity, formation depth and formation wave velocity through a variation of the Faust formula. The random samples are used to train the neural network and build a BP network model that can map resistivity, formation depth and formation wave velocity to each other. The resistivity-to-velocity model building module is used to obtain a measured dataset containing the resistivity at the well location, as well as the corresponding formation depth and formation wave velocity, and to train the BP network model based on the measured dataset to obtain the resistivity-to-velocity model. The acoustic velocity model conversion module is used to obtain the resistivity value and corresponding stratum depth at any location within the work area, and input them into the resistivity-to-velocity model to obtain the predicted acoustic velocity model.

[0017] Compared with existing technologies, the beneficial effects of the deep learning-based resistivity model-rotational speed model method and system provided in this invention are as follows: Random samples of resistivity, formation depth, and formation wave velocity were generated using a variation of the Faust formula and used to train a neural network. This allowed the BP network model to be adjusted to handle the mapping between resistivity, formation depth, and formation wave velocity, providing a feasible basis for subsequent BP network model training. The BP network model was then trained using measured data from well locations, enabling it to construct the mapping relationship between resistivity, formation depth, and formation wave velocity through deep learning, thus obtaining a resistivity-transformation velocity model. Once the trained resistivity-transformation velocity model could accurately predict formation wave velocity, resistivity values ​​and corresponding formation depths at any location within the work area were selected as inputs to the model. This allowed for the accurate output of a sonic velocity model for the entire subsurface area, exhibiting higher flexibility, accuracy, and adaptability, and enabling it to handle deeper, more complex surface structures, thus supporting static correction and velocity modeling of seismic data. Attached Figure Description

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 A schematic block diagram illustrating the steps of the resistivity model-rotational speed model method provided in this embodiment of the invention; Figure 2 The left-middle figure is a curve comparing the predicted and true values ​​of the second training sample provided in an embodiment of the present invention. Figure 2 The right-middle figure is a prediction error curve of the second training sample provided in an embodiment of the present invention; Figure 3 The left-middle figure is a curve comparing the predicted and actual values ​​of the verification sample provided in an embodiment of the present invention. Figure 3 The right-middle figure is a prediction error curve of the verification sample provided in an embodiment of the present invention; Figure 4 The left and middle graphs show a comparison between the calculated and actual values ​​of the traditional Faust empirical formula. Figure 4 The right-middle figure shows the calculation error curve of the traditional Faust empirical formula; Figure 5 Figure (a) is an inversion resistivity model diagram of a shallow lateral profile provided in an embodiment of the present invention. Figure 5 Figure (b) shows the velocity model calculated using the traditional Faust empirical formula. Figure 5 Figure (c) shows the acoustic velocity model of a shallow side profile provided in an embodiment of the present invention. Detailed Implementation

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0020] This invention discloses a deep learning-based resistivity-speed model method, comprising: S1. Generate random samples containing resistivity, formation depth, and formation wave velocity using a variation of the Faust formula. Train a neural network using these random samples to construct a BP network model that maps resistivity, formation depth, and formation wave velocity to each other. The variation of the Faust formula is:

[0021] In the formula, Indicates the ground wave velocity; Represents resistivity; Indicates the depth of the strata. These represent the coefficient vectors of the corresponding terms; S2. Obtain a measured dataset containing the resistivity at the well location, as well as the corresponding formation depth and formation wave velocity. Train the BP network model based on the measured dataset to obtain the resistivity-to-velocity model. S3. Obtain the resistivity model and corresponding formation depth of any region within the work area obtained by electrical inversion, and input them into the resistivity-to-velocity model to convert them into the predicted acoustic velocity model.

[0022] Through the implementation of the resistivity-to-velocity model method described above, BackPropagation (BP) is a widely used neural network training algorithm. The algorithm updates network weights and biases by calculating the gradient of the loss function with respect to network parameters. Foster's Equation (Faust) is an empirical formula characterizing the statistical relationship between formation resistivity and sonic waveforms under depth constraints. A variation of the Faust equation is used to generate random samples of resistivity, formation depth, and formation wave velocity, which are then used to train the neural network. This adjusts the network parameters of the BP network model, enabling it to handle the mapping between resistivity, formation depth, and formation wave velocity, providing a feasible basis for subsequent BP network model training. The BP network model is then trained using measured data from well locations, allowing it to construct the mapping relationship between resistivity, formation depth, and formation wave velocity through deep learning, thus obtaining the resistivity-to-velocity model. Once the trained resistivity-transformation velocity model can accurately predict formation wave velocity, by selecting the resistivity value and corresponding formation depth at any location within the work area as the input of the resistivity-transformation velocity model, it can accurately output the acoustic velocity model for the entire subsurface area. This model has higher flexibility, accuracy, and adaptability, and can cope with deeper complex surface structures, providing support for static correction and velocity modeling of seismic data.

[0023] The variation of the Faust formula mentioned above is obtained by taking the logarithm of both sides of the traditional Faust empirical formula. , and The first-order approximate linear relationship is obtained by taking the logarithm of a higher-order polynomial due to the complexity of shallow relationships. The traditional Faust empirical formula is:

[0024] Where k represents an empirical constant related to rock physical properties; , This represents an empirical constant related to the strata.

[0025] Furthermore, random samples containing resistivity, formation depth, and formation wave velocity are generated through variations of the Faust formula, including: Determine the range of geological surface resistivity, formation depth, and formation wave velocity; Select resistivity, formation depth and formation wave velocity that meet the specified range, and establish the mapping relationship between resistivity, formation depth and formation wave velocity through a variation of the Faust formula. The dataset containing resistivity, formation depth, and formation wave velocity after establishing the mapping relationship is randomly divided into the first training sample and the test sample.

[0026] Furthermore, a BP network model is constructed by training a neural network using random samples, which allows for mutual mapping between resistivity, formation depth, and formation wave velocity, including: A neural network framework based on a backpropagation network was constructed, with the resistivity and formation depth in the first training sample as the input layer and the formation wave velocity in the first training sample as the output layer, and the neural network framework was trained. The trained neural network framework is tested using test samples, and the structure, model parameters, required number of samples, and input and output data formats of the neural network framework are determined based on the test results to obtain the BP network model.

[0027] Through the implementation of the above-described resistivity-to-velocity model method, a certain number of samples are randomly generated within a defined range of resistivity, formation depth, and formation wave velocity within the geological surface using a variation of the Faust formula. These samples are then used as the first training samples and the second as the test samples. A BP-based neural network framework is trained using the first training samples, and the structure and model parameters of the neural network framework are adjusted using the test samples. This allows for the determination of the number of samples required to construct the mapping relationship between resistivity, formation depth, and formation wave velocity using the neural network, as well as the impact of the sample data format on the prediction results. Since existing resistivity-to-velocity models are based on traditional empirical formulas, this embodiment of the invention, as the first to utilize a neural network, aims to determine whether the current network architecture can accurately predict once the resistivity model and the acoustic velocity model satisfy the variation of the Faust formula. Furthermore, it aims to adjust the parameters and input / output data formats of the BP network model, providing a basis for the feasibility of training the BP network model using measured data. Therefore, the BP network model at this stage cannot be considered the final prediction model.

[0028] Furthermore, a neural network framework based on BP networks is constructed, including: Based on the BP network, a neural network framework consisting of an input layer, an output layer, and four hidden layers is constructed. Choose a hyperbolic tangent activation function for each hidden layer and a linear activation function for the output layer.

[0029] In the implementation of the resistivity-speed model method described above, the number of neurons in the input layer should be consistent with the number of input features, while the number of neurons in the output layer depends on the specific task, such as the number of predicted values ​​in a regression problem. Four hidden layers are defined, and the number of neurons in each hidden layer can be set according to the complexity of the problem. For the hidden layers, the hyperbolic tangent (Tanh) activation function is chosen, which can output values ​​in the range [-1, 1], contributing to the stability and convergence of the model. For the output layer, a linear activation function (identity function) is chosen, which can maintain consistency between the output value and the predicted value in regression problems. Preferably, the training parameters of the BP network model are set to 1000 iterations and an error threshold. The learning rate is 0.01.

[0030] Furthermore, a resistivity-transfer model is obtained by training the BP network model based on the measured dataset, including: The measured dataset containing resistivity, formation depth, and formation wave velocity was randomly divided into a second training sample and a validation sample. The resistivity and formation depth in the second training sample are used as input data, and the formation wave velocity in the second training sample is used as output data. The BP network model is then trained to obtain the resistivity-velocity model. The resistivity in the second training sample and the validation sample, along with the corresponding formation depth, are input into the resistivity rotational velocity model to output the predicted formation wave velocity. The error of the prediction result is obtained by comparing the output predicted formation wave velocity with the corresponding formation wave velocities in the second training sample and the validation sample. The predictive performance of the resistivity-transfer rate model was verified based on the error between the prediction results of the actual training samples and the actual verification samples.

[0031] Through the implementation of the above resistivity model-velocity model method, taking the three-dimensional resistivity inversion data of a three-dimensional high-frequency electromagnetic project in a complex structural zone of an oilfield as an example: The predicted formation wave velocity is compared with the corresponding formation wave velocities in the second training sample and the validation sample to obtain the error between the predicted value and the true value. Figure 2 The left figure in the image shows a curve comparing the predicted and true values ​​of the second training sample. Figure 2 The right figure in the figure is the prediction error curve of the second training sample. It can be seen that the prediction error percentage of the second training sample is within 2%. Figure 3 The left figure in the image is a curve comparing the predicted and actual values ​​of the validation sample. Figure 3 The right figure shows the prediction error curve for the validation sample. It can be seen that the prediction error percentage for the validation sample is within 4%. Furthermore, the formation velocity was also calculated using the traditional Faust empirical formula. Figure 4 The left figure in the graph is a curve comparing the calculated values ​​with the actual values ​​using empirical formulas. Figure 4 The right figure shows the error curve of the empirical formula calculation. It can be seen that the formation velocity error obtained by the empirical formula is within 25%. Therefore, the resistivity-velocity model method of this invention has significant advantages over the traditional empirical formula method, yielding more accurate results.

[0032] Furthermore, a measured dataset containing well location resistivity, as well as the corresponding formation depth and formation wave velocity, was obtained, including: Obtain vertical seismic profile data of the target well within the work area, and analyze the vertical seismic profile data to obtain the formation wave velocity from the shallow to the deep surface. Acquire three-dimensional high-frequency electromagnetic data of the target well within the work area, and obtain the resistivity corresponding to the location of the target well by inverting the three-dimensional high-frequency electromagnetic data; The resistivity obtained from the inversion and the formation wave velocity obtained from the analysis are interpolated to correspond to the same formation depth, resulting in a measured dataset that includes the resistivity at the well location, as well as the corresponding formation depth and formation wave velocity.

[0033] Furthermore, this includes a method for data processing after mapping resistivity and formation wave velocity to the same formation depth: Based on the logarithmic domain of resistivity, formation wave velocity and formation depth are uniformly converted to the logarithmic domain; The resistivity, formation wave velocity, and formation depth in the logarithmic domain are normalized to obtain the measured dataset used to train the BP network model. The normalization function formula is as follows:

[0034] In the formula, This indicates the resistivity, formation depth, or formation wave velocity converted to the logarithmic domain. This represents the maximum value of resistivity, formation depth, or absolute value of formation wave velocity. This represents the normalized resistivity, formation depth, or formation wave velocity.

[0035] Through the implementation of the above-described resistivity-velocity model method, when training the BP network model using measured data, it is considered that the inverted resistivity and the formation wave velocity data obtained from Vertical Seismic Profile (VSP) do not correspond at the same bottom depth. Therefore, it is necessary to interpolate the inverted resistivity and formation wave velocity according to the required accuracy to ensure a one-to-one correspondence between resistivity and formation wave velocity at the same depth. Secondly, the value of resistivity is usually in the logarithmic domain. Therefore, to simplify the learning difficulty of the BP network model, this embodiment of the invention chooses to take the logarithm of both formation depth and formation wave velocity, so that the data learned by the BP network model are all in the logarithmic domain. Thus, a portion of the processed data is used as the second training sample, and the other portion is used as the validation sample.

[0036] Furthermore, the resistivity model and corresponding formation depth of any location within the work area obtained through electrical resistivity inversion are acquired and input into the resistivity-transformation model to obtain the predicted acoustic velocity model, including: A sideline with complex shallow structures was selected within the work area, and a resistivity model was obtained from the profile of the selected sideline through inversion. Based on the resistivity model, the resistivity corresponding to the preset formation depth is estimated by interpolation. The interpolated resistivity and corresponding formation depth are used as input data into the resistivity-to-velocity model, which is then converted to obtain the acoustic velocity model of the selected side profile.

[0037] Furthermore, this includes methods for processing the input and output data of the resistivity-transport model: The interpolated resistivity and corresponding formation depth are converted into the logarithmic domain. The resistivity and corresponding formation depth in the logarithmic domain are normalized to obtain the input data for the resistivity-to-velocity model. The predicted formation wave velocity output from the resistivity-transfer model is inversely normalized to obtain the acoustic velocity model for the selected side profile.

[0038] Through the implementation of the above resistivity model-velocity model method, taking the three-dimensional resistivity inversion data of a three-dimensional high-frequency electromagnetic project in a complex structural zone of an oilfield as an example: An inversion resistivity model is selected from a shallow, structurally complex lateral profile within the three-dimensional work area, such as... Figure 5 As shown in Figure (a), the profile contains multiple high-resistivity and low-resistivity anomalies, reaching a vertical depth of 2000 meters. Resistivity values ​​at the required depths are obtained through interpolation. These resistivity values ​​and formation depths are then converted to the logarithmic domain and normalized. The normalized resistivity and corresponding formation depths are used as input data and substituted into the resistivity-to-velocity model to obtain the predicted formation velocity. This is then inversely normalized to obtain the acoustic velocity model of the profile, as shown below. Figure 5 As shown in Figure (c). Furthermore, the velocity model is calculated using the traditional Faust empirical formula, as follows: Figure 5 As shown in Figure (b), compared with the traditional Faust empirical formula, it can be seen that the neural network prediction results of the embodiment of the present invention are more accurate in shallow resistivity anomaly areas. Therefore, using the acoustic velocity model obtained from the embodiment of the present invention for static correction and velocity modeling of seismic acquisition data can achieve better results.

[0039] This invention also discloses a system for a resistivity-rotational model, employing the aforementioned resistivity-rotational model method, comprising: The BP network model building module is used to generate random samples containing resistivity, formation depth and formation wave velocity through a variation of the Faust formula. The random samples are used to train the neural network and build a BP network model that can map resistivity, formation depth and formation wave velocity to each other. The resistivity-to-velocity model building module is used to obtain a measured dataset containing the resistivity at the well location, as well as the corresponding formation depth and formation wave velocity. The BP network model is trained based on the measured dataset to obtain the resistivity-to-velocity model. The acoustic velocity model conversion module is used to obtain the resistivity value and corresponding stratum depth at any location within the work area, input them into the resistivity-to-velocity model, and obtain the predicted acoustic velocity model.

[0040] Based on the above-described resistivity model-speed model method, this invention also discloses a computer-readable storage medium, which, when executed by a processor, implements the steps of the above-described resistivity model-speed model method.

[0041] Based on the above-described resistivity model-speed model method, the present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described resistivity model-speed model method.

[0042] This invention is described based on flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to specific embodiments. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the flowcharts and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0043] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0044] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0045] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Those skilled in the art can modify the technical solutions described in the above embodiments, or make equivalent substitutions for some of the technical features; and all such modifications and substitutions should fall within the protection scope of the appended claims of the present invention.

Claims

1. A method for resistivity-speed model based on deep learning, characterized in that, The resistivity model-speed model method includes: Random samples containing resistivity, formation depth, and formation wave velocity are generated using a variation of the Faust formula. These random samples are then used to train a neural network, constructing a backpropagation (BP) network model that maps resistivity, formation depth, and formation wave velocity to each other. The variation of the Faust formula is as follows: In the formula, Indicates the ground wave velocity; Represents resistivity; Indicates the depth of the strata. These represent the coefficient vectors of the corresponding terms; Obtain a measured dataset containing the resistivity at the well location, as well as the corresponding formation depth and formation wave velocity. Train the BP network model based on the measured dataset to obtain a resistivity-to-velocity model. Obtain the resistivity model and corresponding formation depth of any location within the work area obtained by electrical inversion, and input them into the resistivity rotational velocity model to obtain the predicted acoustic velocity model.

2. The deep learning-based resistivity model-speed model method according to claim 1, characterized in that, The generation of random samples containing resistivity, formation depth, and formation wave velocity through a variation of the Faust formula includes: Determine the range of geological surface resistivity, formation depth, and formation wave velocity; Select resistivity, formation depth, and formation wave velocity that meet a certain range, and establish the mapping relationship between resistivity, formation depth, and formation wave velocity through a variation of the Faust formula. The dataset containing resistivity, formation depth, and formation wave velocity after establishing the mapping relationship is randomly divided into the first training sample and the test sample.

3. The deep learning-based resistivity model-speed model method according to claim 2, characterized in that, The method of training a neural network using random samples to construct a BP network model that allows for mutual mapping between resistivity, formation depth, and formation wave velocity includes: A neural network framework based on a backpropagation network is constructed, with resistivity and formation depth in the first training sample as the input layer and formation wave velocity in the first training sample as the output layer, and the neural network framework is trained. The trained neural network framework is tested using test samples, and the structure, model parameters, required number of samples, and input and output data formats of the neural network framework are determined based on the test results to obtain the BP network model.

4. The deep learning-based resistivity model-speed model method according to claim 3, characterized in that, The construction of the neural network framework based on the BP network includes: Based on the BP network, a neural network framework consisting of an input layer, an output layer, and four hidden layers is constructed. A hyperbolic tangent activation function is selected for each hidden layer, and a linear activation function is selected for the output layer.

5. The deep learning-based resistivity model-speed model method according to claim 1, characterized in that, The step of training the BP network model based on the measured dataset to obtain the resistivity-transfer model includes: The measured dataset containing resistivity, formation depth, and formation wave velocity was randomly divided into a second training sample and a validation sample. The resistivity and formation depth in the second training sample are used as input data, and the formation wave velocity in the second training sample is used as output data. The BP network model is then trained to obtain the resistivity-velocity model. The resistivity in the second training sample and the validation sample, along with the corresponding formation depth, are input into the resistivity rotational velocity model to output the predicted formation wave velocity. The error of the prediction result is obtained by comparing the output predicted formation wave velocity with the corresponding formation wave velocities in the second training sample and the validation sample. The predictive performance of the resistivity-rotational model is verified based on the error between the prediction results of the measured training samples and the measured verification samples.

6. The deep learning-based resistivity-speed model method according to claim 5, characterized in that, The acquisition of the measured dataset, which includes well location resistivity, corresponding formation depth, and formation wave velocity, includes: Obtain vertical seismic profile data of the target well within the work area, and analyze the vertical seismic profile data to obtain the formation wave velocity from the shallow to the deep surface. Acquire three-dimensional high-frequency electromagnetic data of the target well within the work area, and obtain the resistivity corresponding to the location of the target well by inverting the three-dimensional high-frequency electromagnetic data; The resistivity obtained from the inversion and the formation wave velocity obtained from the analysis are interpolated to correspond to the same formation depth, resulting in a measured dataset that includes the resistivity at the well location, as well as the corresponding formation depth and formation wave velocity.

7. The deep learning-based resistivity model-speed model method according to claim 6, characterized in that, This includes a method for processing data after mapping the resistivity and formation wave velocity to the same formation depth: Based on the logarithmic domain of the resistivity, the formation wave velocity and the formation depth are uniformly converted to the logarithmic domain; The resistivity, formation wave velocity, and formation depth in the logarithmic domain are normalized to obtain the measured dataset used to train the BP network model. The normalization function formula is as follows: In the formula, This indicates the resistivity, formation depth, or formation wave velocity converted to the logarithmic domain. This represents the maximum value of resistivity, formation depth, or absolute value of formation wave velocity. This represents the normalized resistivity, formation depth, or formation wave velocity.

8. The deep learning-based resistivity model-speed model method according to claim 1, characterized in that, The process of obtaining the resistivity model and corresponding formation depth of any location within the work area through electrical inversion, and inputting them into the resistivity rotational velocity model to convert them into a predicted acoustic velocity model includes: A sideline with complex shallow structures was selected within the work area, and a resistivity model was obtained from the profile of the selected sideline through inversion. Based on the resistivity model, the resistivity corresponding to the preset formation depth is estimated by interpolation. The interpolated resistivity and the corresponding formation depth are used as input data into the resistivity-to-velocity model to obtain the acoustic velocity model of the selected side profile.

9. The deep learning-based resistivity model-speed model method according to claim 8, characterized in that, This includes methods for processing the input and output data of the resistivity-speed model: The interpolated resistivity and the corresponding formation depth are converted into the logarithmic domain. The resistivity and the corresponding formation depth in the logarithmic domain are normalized to obtain the input data for the resistivity-to-velocity model. The predicted formation wave velocity output from the resistivity rotational velocity model is inversely normalized to obtain the acoustic velocity model for the selected side profile.

10. A system for resistivity-rotational modeling, employing the resistivity-rotational modeling method according to any one of claims 1-9, characterized in that, include: The BP network model building module is used to generate random samples containing resistivity, formation depth and formation wave velocity through a variation of the Faust formula. The random samples are used to train the neural network and build a BP network model that can map resistivity, formation depth and formation wave velocity to each other. The resistivity-to-velocity model building module is used to obtain a measured dataset containing the resistivity at the well location, as well as the corresponding formation depth and formation wave velocity, and to train the BP network model based on the measured dataset to obtain the resistivity-to-velocity model. The acoustic velocity model conversion module is used to obtain the resistivity value and corresponding stratum depth at any location within the work area, and input them into the resistivity-to-velocity model to obtain the predicted acoustic velocity model.

Images