Inversion method for borehole electromagnetic wave detection interface and implementation system thereof

Abstract

This invention discloses an inversion method for azimuth electromagnetic wave (AEM) detection interfaces during drilling, comprising: performing forward modeling simulations on different historical logging state points based on historical logging data, historical steering information, and geological information related to the target layer in the target reservoir area; calculating the amplitude ratio and phase difference data corresponding to each measuring point; training a preset neural network model based on this to generate a logging inversion model for realizing the inversion of azimuth electromagnetic wave logging data; acquiring the amplitude ratio and phase difference data of the current measured point obtained by the azimuth electromagnetic wave detection instrument during logging while drilling; and using the logging inversion model to output the corresponding logging inversion results for real-time geological steering. This invention eliminates the need to store a large forward model library, has high computational efficiency, can be used for real-time geological steering, and eliminates the cumbersome calibration process of azimuth electromagnetic waves, directly mapping the measurement signal to the geological signal.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation technology, and more specifically, to an inversion method and implementation system for the interface of azimuth electromagnetic wave detection while drilling. Background Technology

[0002] Accurate prediction of formation boundaries during drilling is crucial for real-time geological steering decisions, reservoir qualitative and even quantitative evaluation. Azimuth electromagnetic wave logging has advantages such as a large detection range and strong sensitivity to formation interface azimuth, and is widely used in real-time geological steering and reservoir evaluation. However, azimuth electromagnetic wave resistivity measures amplitude ratio and phase difference, which cannot directly reflect geological and geoelectric information, requiring inversion of measurement data.

[0003] In existing drilling electromagnetic resistivity detection inversion techniques, the main methods employed are forward modeling libraries or gradient-based inversion methods. Due to the numerous influencing factors in azimuth electromagnetic resistivity logging, forward modeling library methods require the creation of a large number of forward models, necessitating significant storage space. Furthermore, these methods cannot cover all influencing factors; for cases where the model library cannot be fully explored, linear interpolation algorithms are typically used, which have limited accuracy. Therefore, the application of this method is somewhat restricted. Additionally, gradient-based algorithms are computationally intensive and highly dependent on the selection of initial inversion values, resulting in low search efficiency. Consequently, the real-time performance of these algorithms needs improvement.

[0004] In summary, there is a need for a new drilling electromagnetic resistivity detection and inversion scheme that can solve one or more of the above-mentioned technical problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an inversion method for the azimuth electromagnetic wave detection interface during drilling, comprising: Step 1, performing forward modeling simulations on different historical logging state points based on historical logging data, historical steering information, and geological information related to the target layer in the target reservoir area, calculating the amplitude ratio and phase difference data corresponding to each measurement point, and training a preset neural network model based on this to generate a logging inversion model for realizing the inversion of azimuth electromagnetic wave logging data; Step 2, acquiring the amplitude ratio and phase difference data of the current measured point obtained by the azimuth electromagnetic wave detection instrument during logging while drilling, and using the logging inversion model to output the corresponding logging inversion results for real-time geological steering.

[0006] Preferably, step one includes: constructing a forward model of the target layer in the target reservoir area; establishing a single-point calculation method for directly calculating the amplitude ratio and phase difference of corresponding measuring points based on the feature information indicated by the forward model; generating multiple sets of resistivity parameters based on the specified formation resistivity range of the target layer and in conjunction with the forward model, and constructing several measuring points for each set of resistivity parameters, thereby using the single-point calculation method to calculate the amplitude ratio and phase difference data of all measuring points to form a forward calculation dataset; training a preset neural network model based on the forward calculation dataset to generate the well logging inversion model.

[0007] Preferably, in the step of establishing a single-point calculation method for directly calculating the amplitude ratio and phase difference of the corresponding measuring point based on the feature information indicated by the forward model, the Hankel transform method is used to construct the single-point calculation method, and the feature information includes formation information, wellbore trajectory information, and instrument parameters.

[0008] Preferably, the forward model is constructed as a two-layer forward model or a three-layer hierarchical forward model.

[0009] Preferably, the step of generating multiple sets of resistivity parameters and constructing several measurement points for each set of resistivity parameters within a specified formation resistivity range based on the target layer and in conjunction with the forward model includes: randomly generating multiple sets of resistivity parameters according to the specified formation resistivity range and the number of layers in the forward model, wherein the resistivity parameters include combinations of the resistivity of various formations in the forward model; when constructing a set of measurement points for the same set of resistivity parameters, randomly generating several combinations consisting of different instrument depths, different source distances, and different instrument transmission frequencies, and obtaining corresponding forward modeling features for each combination, wherein each combination corresponds to one measurement point.

[0010] Preferably, before training the preset neural network model based on the forward calculation dataset to generate the well logging inversion model, the inversion method further includes: normalizing the forward calculation features, amplitude ratio data and phase difference data of each measuring point in the forward calculation dataset, wherein the forward calculation features include the distance from the drilling instrument to the boundary, the formation conductivity and the relative dip angle of adjacent formations.

[0011] Preferably, the step of training a preset neural network model based on the forward modeling dataset to generate the well logging inversion model includes: constructing multiple neural network models with different types of networks and different model parameters; training different neural network models using different learning rules based on the training dataset in the forward modeling dataset, and selecting the optimal model and the optimal learning rule.

[0012] Preferably, the optimal learning rule is a quasi-Newton method, and the optimal model contains three hidden layers with 25, 15 and 5 neurons in each hidden layer, respectively.

[0013] Preferably, the azimuth electromagnetic wave detection instrument includes a resistivity signal measurement unit and a geological signal measurement unit. The resistivity signal measurement unit includes at least one transmitting antenna and two receiving antennas, and the geological signal measurement unit includes at least one transmitting antenna and one receiving antenna.

[0014] On the other hand, the present invention provides a system for implementing an intelligent inversion method. The system includes: a logging inversion model generation device configured to perform forward modeling simulations on different historical logging state points based on historical logging data, historical steering information, and geological information related to the target layer of the target reservoir area, calculating the amplitude ratio and phase difference data corresponding to each logging point; and, based on this, training a preset neural network model to generate a logging inversion model for implementing azimuth electromagnetic wave logging data inversion; and a logging-while-drilling inversion device configured to acquire the amplitude ratio and phase difference data of the current measured point obtained by the azimuth electromagnetic wave detection instrument during logging-while-drilling, and, using the logging inversion model, output corresponding logging inversion results for real-time geological steering.

[0015] Compared with the prior art, one or more embodiments of the above solutions may have the following advantages or beneficial effects:

[0016] This invention proposes an inversion method and its implementation system based on artificial neural networks for azimuth electromagnetic wave detection interfaces during drilling. This scheme fully considers the influence of source distance, transmission frequency, formation thickness, relative dip angle, resistivity anisotropy, and instrument radius on the measurement response. It trains an artificial neural network model characterizing the well logging inversion process using a large amount of forward modeling data. The constructed neural network model enables the conversion of instrument measurement data into inversion parameters representing formation geological or physical characteristics, such as instrument distance from the boundary, formation conductivity, and relative dip angle, during well logging. Thus, after obtaining the well logging inversion model through model training using the artificial neural network inversion method, this invention eliminates the need to store a large forward model library, resulting in high computational efficiency. It can be used for real-time geological guidance and eliminates the cumbersome calibration process for azimuth electromagnetic waves, allowing direct correlation between measurement signals and geological signals.

[0017] While the invention will be described below in conjunction with some exemplary embodiments and methods of use, those skilled in the art will understand that it is not intended to limit the invention to these embodiments. Rather, it is intended to cover all alternatives, modifications, and equivalents that fall within the spirit and scope of the invention as defined in the appended claims.

[0018] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained by means of the structures particularly pointed out in the following description, claims, and drawings. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0020] Figure 1 This is a step diagram of the inversion method for the azimuth electromagnetic wave detection interface during drilling, according to an embodiment of this application.

[0021] Figure 2 This is a flowchart illustrating the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in an embodiment of this application.

[0022] Figure 3 This is a schematic diagram of the resistivity signal measurement unit in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in this application embodiment.

[0023] Figure 4 This is a schematic diagram of the geological signal measurement unit in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in this application embodiment.

[0024] Figure 5 This is an example diagram of a two-layer forward model in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in this application.

[0025] Figure 6 This is an example diagram of a three-layered forward model in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in this application.

[0026] Figure 7 This is a comparison chart showing the effect of amplitude ratio and phase difference data obtained by forward modeling calculation and amplitude ratio and phase difference data obtained by well logging inversion model in the inversion method for drilling azimuth electromagnetic wave detection interface according to an embodiment of this application.

[0027] Figure 8 This is a diagram showing the absolute error distribution between the amplitude ratio and phase difference data obtained by forward modeling and the amplitude ratio and phase difference data obtained by well logging inversion model in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in this application embodiment.

[0028] Figure 9This is a schematic diagram of the system for implementing the intelligent inversion method according to an embodiment of this application. Detailed Implementation

[0029] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings and examples, so that the process of how the present invention uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly. It should be noted that, as long as there is no conflict, the various embodiments and features in the various embodiments of the present invention can be combined with each other, and the resulting technical solutions are all within the protection scope of the present invention.

[0030] Furthermore, the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0031] Accurate prediction of formation boundaries during drilling is crucial for real-time geological steering decisions, reservoir qualitative and even quantitative evaluation. Azimuth electromagnetic wave logging has advantages such as a large detection range and strong sensitivity to formation interface azimuth, and is widely used in real-time geological steering and reservoir evaluation. However, azimuth electromagnetic wave resistivity measures amplitude ratio and phase difference, which cannot directly reflect geological and geoelectric information, requiring inversion of measurement data.

[0032] In existing drilling electromagnetic resistivity detection inversion techniques, the main methods employed are forward modeling libraries or gradient-based inversion methods. Due to the numerous influencing factors in azimuth electromagnetic resistivity logging, forward modeling library methods require the creation of a large number of forward models, necessitating significant storage space. Furthermore, these methods cannot cover all influencing factors; for cases where the model library cannot be fully explored, linear interpolation algorithms are typically used, which have limited accuracy. Therefore, the application of this method is somewhat restricted. Additionally, gradient-based algorithms are computationally intensive and highly dependent on the selection of initial inversion values, resulting in low search efficiency. Consequently, the real-time performance of these algorithms needs improvement.

[0033] In summary, to address one or more of the aforementioned technical problems, this application provides an intelligent inversion method and its implementation system for azimuth electromagnetic wave detection interfaces during drilling. This method calculates corresponding forward modeling data based on historical logging data, historical steering data, and geological information of the target layer in the target reservoir area. The calculated forward modeling data is then used to train a pre-constructed intelligent neural network model, resulting in a logging inversion model. This model is then used to perform inversion processing on the amplitude ratio and phase difference data calculated from logging signals acquired by the azimuth electromagnetic wave detection instrument during logging, generating corresponding information related to geological steering and reservoir evaluation for real-time geological steering. Thus, the neural network inversion method of this invention can train the network based on existing measurement data and forward modeling data. The trained network has the ability to generalize to new data and can be used to predict new measurement data. After training, no training data storage is required, resulting in high computational efficiency and facilitating the utilization of large amounts of existing data. This offers certain advantages in real-time geological steering applications.

[0034] Figure 1 This is a step diagram illustrating the inversion method for the azimuth electromagnetic wave detection interface during drilling, according to an embodiment of this application. (Refer to the following...) Figure 1 The present invention describes the inversion method for the azimuth electromagnetic wave detection interface during drilling (hereinafter referred to as the "intelligent inversion method").

[0035] Step S110: Based on historical logging data, historical steering information and geological information related to the target layer of the target reservoir area, forward modeling is performed on different historical logging state points to calculate the amplitude ratio and phase difference data corresponding to each state point. Based on this, the preset neural network model is trained to generate a logging inversion model for realizing the inversion of azimuth electromagnetic wave logging data.

[0036] In step S110, firstly, all drilled and currently drilled wells covering the target layer in the target area to be studied are identified, and historical logging information, historical steering information, and geological information of these wells (historical wells) are acquired. Then, based on the logging operations performed on each historical well, the target logging parameters for each logging state point (the state corresponding to different depths reached by the logging-while-drilling instrument) are calculated using the historical logging information, historical steering information, and geological information of each historical well during each historical logging operation. These target logging parameters are the amplitude ratio and phase difference data obtained by the azimuth electromagnetic wave detection instrument at the current logging state point. Next, based on the amplitude ratio and phase difference data of each historical logging state point, and the historical logging information, historical steering information, and geological information of each historical well, a preset neural network model is trained. After training, a corresponding logging inversion model is generated. The logging inversion model is used for data inversion calculation processing functions required after the azimuth electromagnetic wave logging instrument obtains the electromagnetic wave received signals during the measurement-while-drilling process.

[0037] Therefore, after obtaining the well logging inversion model based on the artificial intelligence neural network model, the process proceeds to step S120 to apply the well logging inversion model in the logging-while-drilling process.

[0038] Step S120 acquires the real-time logging signals collected by the azimuth electromagnetic wave detection instrument during logging while drilling in the target reservoir area. The resulting (real-time) amplitude ratio and (real-time) phase difference data of the current (logging) measured point are then used to directly output the corresponding logging inversion results using the logging inversion model trained in step S110. These results are then applied to real-time geological steering. The logging inversion results predicted by the logging inversion model constructed based on an artificial intelligence neural network model include: the distance from the logging instrument to the boundary, formation conductivity, and the relative dip angle of adjacent formations.

[0039] Figure 2 This is a flowchart illustrating the inversion method for the azimuth electromagnetic wave detection interface during drilling, according to an embodiment of this application. Refer to the following... Figure 2 The intelligent inversion method described in the embodiments of the present invention will be explained in detail.

[0040] In step S110, firstly (step S1101, not shown), a forward model of the target layer in the target reservoir area to be studied is constructed. In step S1101, the forward model is constructed based on historical logging data, historical steering information, and geological information related to the target layer in the target reservoir area. In this embodiment, the forward model is constructed as a two-layer forward model or a three-layer stratified forward model.

[0041] Furthermore, the forward model constructed in this embodiment of the invention indicates a large amount of feature information. This feature information includes formation information, wellbore trajectory information, and instrument parameters. Specifically, the formation information is preferably formation resistivity-related data, including: upper and lower formation resistivity, upper and lower formation angle (relative dip angle), upper and lower formation distance (layer thickness), and upper and lower formation radial resistivity (resistivity anisotropy). The wellbore trajectory information is used to indicate the position of the azimuth electromagnetic wave detection instrument corresponding to different logging state points. The instrument parameters are the operating parameters of the azimuth electromagnetic wave detection instrument, specifically including the source distance between the transmitting coil (or antenna) and the receiving coil (or antenna) within the azimuth electromagnetic wave detection instrument, the transmission frequency, and other information.

[0042] Furthermore, in this embodiment of the invention, both historical logging operations performed on historical wells related to the target layer of the target reservoir area and measurement while drilling operations performed on ongoing wells related to the target layer of the target reservoir area are implemented using an azimuth electromagnetic wave detection instrument. The azimuth electromagnetic wave detection instrument includes a resistivity signal measurement unit and a geological signal measurement unit.

[0043] further, Figure 3 This is a schematic diagram of the resistivity signal measurement unit in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in an embodiment of this application. Figure 3 As shown, the resistivity signal measurement unit includes at least one transmitting antenna (S8) and two receiving antennas (S9, S10). When measuring the formation resistivity, the transmitting antenna (S8) transmits electromagnetic waves into the formation, and the two receiving antennas (S9, S10) receive the electromagnetic wave signals propagating through the formation. The formation resistivity is reflected by measuring the attenuation of the electromagnetic wave signals between the two receivers. Specifically, the transmitting antenna (S8) is preferably an axial transmitting coil, and the two receiving antennas (S9, S10) are axial receiving coils, respectively.

[0044] further, Figure 4 This is a schematic diagram of the geological signal measurement unit in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in an embodiment of this application. Figure 4 As shown, the geological signal measurement unit includes at least one transmitting antenna (S11) and one receiving antenna (S12). Specifically, the transmitting antenna (S8) is preferably an axial transmitting coil, and the receiving antenna (S12) is preferably a receiving coil whose coil plane forms a certain angle with the wellbore axial direction (the angle is less than 90 degrees, or greater than 90 degrees and less than 180 degrees).

[0045] Figure 5 This is an example diagram of a two-layer forward model in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in this application. Figure 5This is a schematic diagram illustrating the effect of constructing an example of a two-layer forward model of the target layer in a target reservoir region. Figure 5 As shown, R S Rt represents the resistivity of the surrounding rock, Rt represents the resistivity of the target layer, and DTB represents the vertical distance from the instrument center to the formation interface (the distance between the instrument and the boundary at a specific logging point). Additionally, Figure 5 The solid black portion in the image represents an azimuth electromagnetic wave detection instrument.

[0046] Figure 6 This is an example diagram of a three-layered forward model in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in this application. Figure 6 This is a schematic diagram illustrating the effect of constructing an example of a three-layered forward model of the target layer in a target reservoir region. Figure 6 As shown, R S1 and R S2 These represent the resistivity of the upper and lower surrounding rocks, respectively; Rt represents the resistivity of the target layer; and DTB1 and DTB2 represent the vertical distances from the instrument center to the upper and lower interfaces of the formation (the distances from the logging instrument to the upper and lower boundaries at a given logging point). Additionally, Figure 6 The solid black portion in the image represents an azimuth electromagnetic wave detection instrument.

[0047] After the forward model is constructed (step S1102, not shown), a single-point calculation method is established to directly calculate the amplitude ratio and phase difference of the corresponding measuring points based on the feature information indicated by the forward model constructed in step S110. In step S1102, the Hankel transform method is used to construct the single-point calculation method. Thus, when performing forward simulation on each historical measuring point using the single-point calculation method, in step S1103 (not shown), the amplitude ratio and phase difference data of the corresponding measuring point are directly calculated for each historical measuring point using the established single-point calculation method (related program) based on the historical logging data, historical steering data, and corresponding geological information of each measuring point.

[0048] Furthermore, after establishing the single-point calculation method, the process proceeds to step S1103 (not shown). Specifically, step S1103 generates multiple sets of resistivity parameters based on the specified formation resistivity range of the current target layer and the forward model constructed in step S1101. For each set of resistivity parameters, several measurement points are constructed. Then, using the single-point calculation method established in step S1102, the amplitude ratio and phase difference data of all measurement points are calculated, thereby forming a forward model calculation dataset.

[0049] In this embodiment of the invention, the forward modeling dataset includes several measurement points, as well as forward modeling feature information, amplitude ratio data, and phase difference data corresponding to each measurement point. Further, the construction process of the measurement points is explained first. Specifically, in the first step, based on historical logging data, historical steering information, and geological information related to the target layer in the target reservoir area, the actual formation resistivity range of the current target layer is determined and recorded as the specified formation resistivity range. Then, based on the current specified formation resistivity range and the number of layers in the forward model constructed in step S1101, multiple sets of resistivity parameters are randomly generated. Each set of resistivity parameters includes a combination of formation resistivities from various formations in the current forward model. For example, if the specified formation resistivity range for the current target layer area is 1–50 Ω·m, then its randomly generated range can be set to 1–50 Ω·m. Because different combinations of resistivity are randomly selected, the forward modeling responses of resistivity signals and geological signals differ. For example, when the forward model has two layers, in one set of randomly generated resistivity parameters, the resistivities of the two layers are randomly generated as 20 Ω·m and 35 Ω·m, respectively; in another set of randomly generated resistivity parameters, the resistivities of the two layers are randomly generated as 10 Ω·m and 20 Ω·m, respectively; and so on. Furthermore, since the resistivity signals acquired by the azimuth electromagnetic wave detection instrument and the geological signal measurement responses differ, when randomly generating different sets of resistivity parameters, it is necessary to ensure that the resistivity values ​​of different layers within the same set of resistivity parameters are not the same.

[0050] After randomly generating multiple sets of resistivity parameters, each set is treated as a specific logging scenario. Then, for each specific logging scenario, corresponding measurement points (simulation points) are constructed. In other words, each randomly generated specific logging scenario is treated as a logging operation event for the surrounding rock layer and target layer indicated in the current forward model. Thus, the wellbore trajectory formed by each logging operation event can be divided into multiple measurement points. Further, in this embodiment, several combinations (of the first type) with different instrument depths, different source distances, and different instrument transmission frequencies are randomly generated, and corresponding forward modeling characteristics are obtained for each combination. Each combination corresponds to one measurement point.

[0051] In other words, each specific logging scenario corresponds to a set of measuring points. This set contains multiple measuring points specific to the current logging scenario, and the positions of these points form the wellbore trajectory characteristics for the current (to be simulated) logging operation event. Furthermore, for a given set of measuring points, which includes multiple measuring points, each point not only corresponds to specific resistivity parameters but also requires the random generation of different instrument depths within the wellbore trajectory range for the current logging operation event, different source distances that the drilling instrument can form for the current logging operation event, and different transmission frequencies that the drilling instrument can emit for the current logging operation event. Then, a first parameter sequence is formed based on different instrument depth data, a second parameter sequence is formed based on different source distance data, and a third parameter sequence is formed based on different transmission frequency data. Next, when generating a measuring point for the current set of resistivity parameters, any element data is randomly extracted from each of the three different parameter sequences to obtain the first type of combination information of the specified instrument depth location information, source distance information, and transmission frequency information for the current logging state point.

[0052] For example, if 500 sets of resistivity parameters are generated, and each set of well logging operation event model corresponds to 400 measurement points, then for the current forward modeling dataset, there are a total of 500 * 400 = 200,000 measurement points and 200,000 forward modeling data.

[0053] Furthermore, each measuring point also corresponds to a set of forward modeling features (information). In this embodiment of the invention, the forward modeling features include at least: the distance between the drilling instrument and the boundary, the formation conductivity, and the relative dip angle of adjacent formations. It should be noted that, in this embodiment of the invention, the forward modeling feature information for each measuring point can be read from the constructed forward model.

[0054] Thus, we obtain the information on all measurement points corresponding to each set of resistivity parameters, as well as the first type of combination information and forward modeling characteristics corresponding to each measurement point.

[0055] Next, after obtaining the forward modeling characteristics of each measuring point, the resistivity parameters, first-type combination information, drilling instrument distance from the boundary data, formation conductivity data, and relative dip angle data of adjacent formations for the same measuring point are simultaneously substituted into the above-mentioned single-point calculation method. Using this single-point calculation method, the forward modeling data (including amplitude ratio data and phase difference data) corresponding to the current measuring point to be simulated is directly calculated. In other words, in this embodiment of the invention, the single-point calculation method is a method that uses the resistivity parameters of the current measuring point, instrument depth location information, source distance information, transmission frequency information, drilling instrument distance from the boundary data, formation conductivity data, and relative dip angle data of adjacent formations as inputs, and amplitude ratio data and phase difference data as outputs, achieving direct calculation between input and output. After calculating the amplitude ratio and phase difference data for all measuring points, a forward modeling dataset is formed based on the resistivity parameters, first-type combination information, forward modeling characteristics, amplitude ratio data, and phase difference data for each logging state point.

[0056] After the forward modeling dataset is constructed, the process proceeds to step S1104 (not shown) to train the well logging inversion model using the information in the forward modeling dataset.

[0057] Furthermore, to improve the accuracy of the well logging inversion model constructed in this embodiment of the invention, the data in the forward modeling dataset needs to be normalized before training the model. Step S1104 normalizes the forward modeling features, amplitude ratio data, and phase difference data of each measuring point in the forward modeling dataset. This ensures that the numerical range of all data is between -1 and 1, thus proceeding to step S1105 (not shown).

[0058] Step S1105: Based on the current forward calculation dataset, train the preset neural network model to generate the well logging inversion model required for intelligent inversion.

[0059] In step S1105, firstly, it is necessary to construct various neural network models composed of different types of neural networks and different model parameters. The different model parameters can be achieved by constructing different numbers of hidden layers and combinations of different numbers of neurons in each hidden layer.

[0060] Furthermore, before training the neural network model, this embodiment of the invention also needs to divide all measurement points in the forward modeling dataset into a training dataset and a test dataset according to a preset ratio. After forming multiple neural network models, this embodiment of the invention will train different neural network models using different learning rules based on the training dataset in the forward modeling dataset that has undergone data normalization, and select the optimal model and the optimal learning rule.

[0061] When training multiple pre-defined neural network models, the first-class combination information, amplitude ratio, and phase difference data of each measurement point are used as training inputs, and the forward modeling feature data (well logging inversion results) of each measurement point are used as training outputs (training labels). Different learning rules (e.g., rule A, rule B, and rule C) are applied to different types of pre-defined neural network models (e.g., M-type network models, N-type network models) using the training dataset. After each training reaches a pre-defined training accuracy standard, several specific preliminary neural network models are obtained, trained under a specific combination of learning rules and specific types of neural network models (e.g., a specific preliminary neural network model trained using rule A and N-type network models, a specific preliminary neural network model trained using rule C and M-type network models, etc.). Furthermore, if a specific preliminary neural network model trained by a certain training combination does not reach the pre-defined training accuracy, the model parameters of the current pre-defined neural network model need to be adjusted, and the adjusted pre-defined model needs to be trained again using the training dataset until the training accuracy standard is reached.

[0062] Then, the test dataset is used to test various specific preliminary neural network models that meet the training accuracy standard, thereby selecting the optimal learning rule and the optimal specific preliminary neural network model, and using the optimal specific preliminary neural network model as the well logging inversion model. Furthermore, in this embodiment of the invention, the optimal learning rule is a quasi-Newton method. The optimal model contains three hidden layers, with 25, 15, and 5 neurons in each hidden layer, respectively.

[0063] Therefore, a logging inversion model is generated that will ultimately be used in actual logging-while-drilling operations. In step S120, during logging-while-drilling in the target reservoir area, the amplitude ratio and phase difference data of the current measured point are calculated using the electromagnetic wave logging signals collected by the azimuth electromagnetic wave detection instrument during logging-while-drilling. Then, using the logging inversion model generated in step S110, the logging inversion results of the current real-time logging depth position are directly output, obtaining the distance of the logging instrument from the boundary, the formation conductivity, and the relative dip angle of adjacent formations. This information is then used to guide the geological steering task during logging-while-drilling operations in real time.

[0064] Example 1:

[0065] (1) Establish as follows Figure 5 The two-layer horizontal stratification model shown has a resistivity of Rs for the overlying stratum, a resistivity of Rt for the target stratum, and a distance of DTB from the instrument to the interface.

[0066] (2) Establish a two-layer model Hankel transform calculation method, calculate the measured voltage response for a given model parameter, and calculate the amplitude ratio and phase difference;

[0067] (3) Given the range of Rt resistivity values, randomly generate resistivity parameters within the range, and use the randomly generated resistivity parameters and well logging construction event model settings to calculate the measurement voltage response of the instrument at different depth points, and calculate its amplitude ratio and phase difference. For example, generate 500 sets of resistivity parameters, each set of model calculation points is 400 points, and there are a total of 500*400=200000 forward modeling data.

[0068] (4) In order to prevent abnormal data from affecting network training and to speed up network training, the 200,000 data points were normalized as a whole so that the numerical range of all forward data was within [-1,1].

[0069] (5) Data from different well logging construction event models obtained by forward modeling were used as input data for neural network training. DTB was used as label data for model training. The network was trained. After multiple experiments, the neural network learning rule was selected as the quasi-Newton method. The network contains three hidden layers with 25, 15, and 5 neurons in each layer, respectively. The network training accuracy was 0.001Ω·m.

[0070] (6) Using the same method of randomly generating parameters as in (3), 10,000 data points were randomly generated. A neural network was used to predict (invert) DTB. The comparison between the inversion calculation results and the forward calculation results is shown in the figure below. Figure 7 As shown ( Figure 7 This is a comparison chart showing the amplitude ratio and phase difference data obtained through forward modeling and the amplitude ratio and phase difference data obtained through well logging inversion model in the inversion method for azimuth electromagnetic wave detection interface in this application embodiment. The forward DTB is the instrument-interface distance calculated using forward modeling, and the inverted DTB is the DTB obtained through neural network inversion. It can be seen that the inversion results and the forward results are very close, as shown in the absolute error distribution chart below. Figure 8 As shown ( Figure 8 This is a distribution diagram of the absolute error between the amplitude ratio and phase difference data obtained by forward modeling and the amplitude ratio and phase difference data obtained by well logging inversion model in the inversion method for the azimuth electromagnetic wave detection interface during drilling, as described in this application embodiment. The absolute error is the difference between the forward DTB and the inverted DTB, and the frequency is the number of times the relative error value occurs. Figure 8 As shown, the absolute error is mainly concentrated within [-0.05, 0.05], with an average value of 0.01946, which meets the inversion accuracy requirements. Furthermore, the accuracy can be further increased when the number of hidden layers in the network is increased.

[0071] On the other hand, based on the above-mentioned intelligent inversion method, the present invention also provides a system for implementing the intelligent inversion method (hereinafter referred to as "intelligent inversion system"), which is used to implement the intelligent inversion method described above. Figure 9 This is a schematic diagram of the system for implementing the intelligent inversion method according to an embodiment of this application. Figure 9 As shown, the intelligent inversion system described in this embodiment of the invention includes: a well logging inversion model generation device 91 and a logging-while-drilling inversion device 92.

[0072] The well logging inversion model generation device 91 is implemented according to the method described in step S110 above. It is configured to perform forward modeling on different historical well logging state points based on historical well logging data, historical steering information and geological information related to the target layer of the target reservoir area, calculate the amplitude ratio and phase difference data corresponding to each measuring point, and train a preset neural network model based on this to generate a well logging inversion model for realizing the inversion of azimuth electromagnetic wave well logging data. The logging-while-drilling inversion device 92 is implemented according to the method described in step S120 above. It is configured to acquire the amplitude ratio and phase difference data of the current measured point obtained by the azimuth electromagnetic wave detection instrument during logging-while-drilling, and output the corresponding well logging inversion results using the above-mentioned well logging inversion model for real-time geological steering.

[0073] This invention provides an inversion method and implementation system based on an artificial neural network for azimuth electromagnetic wave detection interfaces during drilling. This scheme fully considers the influence of source distance, transmission frequency, formation thickness, relative dip angle, resistivity anisotropy, and instrument radius on the measurement response. It trains an artificial neural network model characterizing the well logging inversion process using a large amount of forward modeling data. The constructed neural network model enables the conversion of instrument measurement data into inversion parameters representing formation geological or physical characteristics, such as instrument distance from the boundary, formation conductivity, and relative dip angle, during well logging. Thus, after obtaining the well logging inversion model through model training using the artificial neural network inversion method, this invention eliminates the need to store a large forward model library, resulting in high computational efficiency. It can be used for real-time geological guidance and eliminates the cumbersome calibration process for azimuth electromagnetic waves, allowing direct mapping between measurement signals and geological signals.

[0074] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0075] It should be understood that the embodiments disclosed herein are not limited to the specific structures, processing steps, or materials disclosed herein, but should be extended to equivalent substitutions of these features as understood by those skilled in the art. It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting.

[0076] The phrase "an embodiment" or "an embodiment" used in this specification means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the invention. Therefore, the phrase "an embodiment" or "an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0077] While the embodiments disclosed in this invention are as described above, the content is merely for the purpose of facilitating understanding of the invention and is not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and changes in form and detail of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection of this invention shall still be determined by the scope defined in the appended claims.

Claims

1. An inversion method for a borehole azimuthal electromagnetic wave detecting interface, comprising: Step one, in a model construction stage, according to the historical logging data, historical steering information and geological information related to the target reservoir area target layer, forward modeling is performed on different historical logging state points, the amplitude ratio and phase difference data corresponding to each logging point are calculated, based on which, a preset neural network model is trained to generate a logging inversion model for realizing azimuthal electromagnetic wave logging data inversion; Step two, the amplitude ratio and phase difference data of the current measured point obtained by the azimuthal electromagnetic wave detection instrument during logging while drilling are obtained, and the logging inversion model is used to directly output the logging inversion result corresponding to the current measured point for real-time geosteering, the logging inversion result includes the distance of the drilling instrument from the boundary, the formation conductivity and the relative dip angle of the adjacent formation, wherein the step one comprises: building a forward model about the target reservoir area target layer; establishing a single-point calculation method for directly calculating the amplitude ratio and phase difference of the corresponding logging point according to the characteristic information indicated by the forward model; based on the specified formation resistivity range of the target layer, combining the number of layers of the forward model, a plurality of groups of resistivity parameters are randomly generated, and then a plurality of logging points are constructed for each group of resistivity parameters, so as to calculate the amplitude ratio and phase difference data of all logging points by using the single-point calculation method to form a forward calculation data set, wherein when the logging point set is constructed for the same group of resistivity parameters, a plurality of combinations composed of different instrument depths, different source distances and different instrument transmission frequencies are randomly generated, and the corresponding forward calculation characteristics representing the steering indication result are obtained for each combination, wherein each combination corresponds to a logging point, and the forward calculation characteristics include the distance of the drilling instrument from the boundary, the formation conductivity and the relative dip angle of the adjacent formation; according to the forward calculation data set, and taking the amplitude ratio and phase difference data in the forward calculation data set as input and the corresponding forward calculation characteristics as output, a preset neural network model is trained to generate the logging inversion model.

2. The inversion method of claim 1, wherein, In the step of establishing a single-point calculation method for directly calculating the amplitude ratio and phase difference of the corresponding logging point according to the characteristic information indicated by the forward model, the Hankel transform method is used to construct the single-point calculation method, and the characteristic information includes formation information, well trajectory information and instrument parameters.

3. The inversion method of claim 1, wherein, The forward model is constructed as a two-layer forward model or a three-layer layered forward model.

4. The inversion method of claim 1, wherein, The resistivity parameters include combinations of formation resistivities in the forward model.

5. The inversion method of claim 4, wherein, Before the step of training a preset neural network model according to the forward calculation data set to generate the logging inversion model, the inversion method further comprises: normalizing the forward calculation characteristics, amplitude ratio data and phase difference data of each logging point in the forward calculation data set, respectively.

6. The inversion method according to any one of claims 1 to 5, characterized in that, In the step of training a preset neural network model according to the forward calculation data set to generate the logging inversion model, comprising: building a plurality of neural network models with different types of networks and different model parameters; According to the training data set in the forward calculation data set, different neural network models are trained by using different learning rules, and an optimal model and an optimal learning rule are selected.

7. The inversion method of claim 6, wherein, The optimal learning rule is a quasi-Newton method, and the optimal model comprises three hidden layers, and the number of neurons of each hidden layer is 25, 15 and 5 respectively.

8. The inversion method according to any one of claims 1-5, wherein, The azimuthal electromagnetic wave detection instrument comprises a resistivity signal measurement unit and a geological signal measurement unit, the resistivity signal measurement unit comprises at least one transmitting antenna and two receiving antennas, and the geological signal measurement unit comprises at least one transmitting antenna and one receiving antenna.

9. A system for implementing an intelligent inversion method, characterized in that, The system is used for implementing the inversion method as claimed in any one of claims 1-8, and the system comprises: The logging inversion model generation device is configured to perform forward simulation on different historical logging state points according to historical logging data, historical steering information and geological information related to target reservoir region target layers, calculate amplitude ratio and phase difference data corresponding to each logging point, and train a preset neural network model based on the data to generate a logging inversion model for implementing azimuthal electromagnetic wave logging data inversion; The logging while drilling inversion device is configured to obtain amplitude ratio and phase difference data of a current measured point obtained by the azimuthal electromagnetic wave detection instrument during logging while drilling, output corresponding logging inversion results by using the logging inversion model, and use the results for real-time geological steering.

Images