In-well induced polarization data inversion method and system based on deep learning, and medium

Abstract

The invention provides an in-well induced polarization inversion method and system based on deep learning and a medium, and the method comprises the following steps: carrying out the modeling of a three-dimensional geologic body, and generating a geologic body model containing an abnormal body in a delimited range; randomly generating a plurality of geologic body models containing anomalous bodies with different positions and sizes, calculating secondary field abnormal values generated by the anomalous bodies in the geologic body models of the anomalous bodies with different positions and sizes, and obtaining in-well induced polarization data of fixed wells at corresponding positions of a plurality of groups of anomalous bodies; a network is constructed by using seven full-connection layers, and the constructed data set is trained to achieve the purpose of inversion. Compared with a traditional method, the method has the advantages that the precision and the speed are obviously improved, and the theoretical value of the method in the field of electrical prospecting data processing is also proved.

Description

Technical Field

[0001] The present invention relates to the field of electrical exploration in geophysical inversion technology and can be used for electrical exploration. More specifically, the present invention relates to a method, system and medium for inversion of well induced polarization data based on deep learning. Background Art

[0002] With the continuous advancement of industrial modernization, society's demand for mineral resources is also increasing. Although many prospecting methods have been introduced, geophysical methods still play a vital role in the field of oil and mineral exploration. Therefore, studying geophysical method inversion for prospecting is also crucial. During field exploration, a survey area is demarcated. By passing electricity underground, the ore body will produce a polarization effect on the surface after receiving the current, resulting in secondary field anomalies. Inversion requires inferring the distribution of underground space through these anomalies. Traditional geophysical inversion methods are based on analyzing secondary field anomaly curves and performing a series of numerical simulation inversions such as the slicing method. This method is computationally intensive and time-consuming. Later, with the development of computers, computers began to be used on a large scale for inversion. Commonly used methods include iterative algorithms such as the Newton method and the method of steepest descent. However, these methods are not immune to the shortcomings of being computationally intensive, time-consuming, and having low accuracy. Summary of the Invention

[0003] An object of the present invention is to solve at least the above problems and provide a method, system and medium for inversion of wellbore induced polarization data based on deep learning.

[0004] In a first aspect, an embodiment of the present application provides a deep learning-based induced polarization inversion method, comprising the following steps:

[0005] S1. Perform three-dimensional geological modeling to generate a geological model containing an anomaly within a defined range, which is used to simulate a geological model including a fixed well. Calculate the secondary field anomaly value generated by the anomaly after energization based on the position and size of the anomaly in the geological model, i.e., the induced polarization data of the fixed well at the corresponding position of the anomaly;

[0006] S2. randomly generate multiple geological body models containing anomalies of different positions and sizes, calculate the secondary field anomaly value generated by the anomaly in each geological body model of the anomaly of different positions and sizes according to the method in step S1, and obtain multiple sets of wellbore induced polarization data of fixed wells corresponding to the anomalies; and record the center position and size of each anomaly and the corresponding wellbore induced polarization data, use the center position and size parameters of each anomaly as the label data of the corresponding wellbore induced polarization data, and obtain a trainable labeled induced polarization data set, and repeat this step to construct a data set including multiple trainable labeled induced polarization data;

[0007] S3. Use seven fully connected layers to build a network and train the dataset constructed in step S2. After the network training is completed, an independent test dataset is input to evaluate the size and distribution of the error. When the test set error is within an acceptable range, the network training is considered complete. At this time, the input of the wellbore induced polarization data can output the corresponding predicted anomaly position and size parameters, achieving the purpose of inversion.

[0008] In the step S1, the three-dimensional geological body model is generated to generate a geological body model including an ellipsoidal anomaly within a defined range. The specific method for constructing the geological body model including the ellipsoidal anomaly is as follows:

[0009] According to the demarcated range, a corresponding rectangular parallelepiped geological model is constructed, the resistivity and polarizability of the rectangular parallelepiped geological model are set to a fixed value, and the rectangular parallelepiped geological model is gridded; a point A is arbitrarily determined as the center of the ellipsoid anomaly, and the axis length parameters a, b, and c of the ellipsoid anomaly are set. The position coordinates of the above-determined point A in the rectangular parallelepiped geological model are used as the center coordinates (h, k, l) of the ellipsoid anomaly, and the axis length parameters a, b, and c of the ellipsoid anomaly are set. According to the following spatial ellipsoid range calculation formula, all grid nodes that meet the formula requirements are calculated:

[0010]

[0011] in:

[0012] x, y, z represent the relative coordinate position of the node inside the geological body, h, k, l are the coordinates of the center of the ellipsoid; a, b, c are the axis lengths of the ellipsoid in the x, y, z directions;

[0013] All grid nodes in the model that meet the above formula are traversed. The small cuboids where these nodes are located can be combined to approximate an ellipsoid anomaly that meets the conditions. All small cuboids in the approximate ellipsoid are replaced with the resistivity and polarizability of the anomaly to construct a geological model that includes an ellipsoid anomaly that meets the above requirements.

[0014] In the S3 step, a network is constructed by using a fully connected layer, and the constructed data set is normalized using the Lure activation function, the Semigloss loss function and the Adam optimizer, and the average value and the root mean square are used, and then it is sent to the neural network; the input data of the neural network model is the in-well induced polarization data of the fixed well corresponding to the anomaly, and the output data is the center coordinates of the anomaly location and the axis lengths in the X, Y, and Z directions as reference labels for the output prediction values; the optimal number of training iterations is set, and the error of each round of training is back-propagated to the network, so that the network continuously optimizes the neural network model parameters to best fit the nonlinear mapping relationship between the in-well induced polarization data and the center coordinates and axis lengths of the underground anomaly. The training is stopped when the loss between the sample prediction value and the true value in the verification set decreases and tends to be stable after training; at this time, the network has learned the nonlinear mapping relationship between the input data and the output data, and the inverted anomaly position and axis length can be obtained after the induced polarization data is input into the network.

[0015] The distance between the ellipsoid anomaly and the boundary of the rectangular geological body model should be no less than 2 grids; the ellipsoid anomaly should not be exposed on the surface or other boundaries; the length of each axis of the ellipsoid anomaly should be no less than 1 grid length.

[0016] In a second aspect, an embodiment of the present application provides a deep learning-based induced polarization inversion system, comprising:

[0017] The 3D geological body modeling module is used to perform 3D geological body modeling and generate a geological body model containing anomalies within a defined range. This model is used to simulate a geological model including a fixed well. Based on the position and size of the anomaly in the geological body model, the secondary field anomaly value generated by the anomaly after power is applied is calculated, i.e., the induced polarization data of the fixed well at the corresponding position of the anomaly.

[0018] an induced polarization data set construction module, randomly generating a plurality of geological body models containing anomalies of different positions and sizes, and calculating the secondary field anomaly values generated by the anomalies in each geological body model of the anomalies of different positions and sizes according to the method in step S1, thereby obtaining a plurality of sets of in-well induced polarization data of fixed wells corresponding to the anomalies; and recording the center position and size of each anomaly and the corresponding in-well induced polarization data, and using the center position and size parameters of each anomaly as label data of the corresponding in-well induced polarization data to obtain a trainable labeled induced polarization data set, and repeating this step to construct a data set including a plurality of trainable labeled induced polarization data sets;

[0019] The borehole IP inversion module uses 7 fully connected layers to build a network and train the data set constructed in step S2. After the network training is completed, an independent test data set is input to evaluate the size and distribution of the error. When the test set error is within an acceptable range, the network training is considered complete. At this time, the borehole IP data can be input to output the corresponding predicted anomaly position and size parameters, achieving the inversion purpose.

[0020] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores program code, and when the program code is executed by a processor, the steps of the deep learning-based downhole induced polarization inversion method as described above are implemented.

[0021] Beneficial effects of the present invention:

[0022] (1) The present invention proposes a method for inverting underground anomalies from borehole induced polarization data based on deep learning, which improves the speed and efficiency of inversion compared with traditional borehole induced polarization inversion methods.

[0023] (2) The synthetic wellbore induced polarization data established by the present invention provides rich training resources for deep neural networks, solving the problem of insufficient wellbore induced polarization data corresponding to underground anomalies calculated manually.

[0024] (3) Other advantages, objectives, and features of the present invention will be partially reflected in the following description and partially understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flowchart of the invention's induced polarization data inversion anomaly;

[0026] Figure 2 This is a flow chart of the synthesis of induced polarization data in the well of the present invention;

[0027] Figure 3 This is a model diagram of the underground space without abnormal bodies in the present invention;

[0028] Figure 4 This is a location diagram of the underground space well of the present invention;

[0029] Figure 5 This is the distribution map of the ellipsoidal anomaly in the underground space of the present invention;

[0030] Figure 6 This is the induced polarization data of the ellipsoidal anomaly in the well of the present invention;

[0031] Figure 7 The borehole induced polarization data used to verify the present invention;

[0032] Figure 8This is a fitting diagram of the predicted values and true values of 50 randomly sampled data received in the well. DETAILED DESCRIPTION

[0033] This embodiment provides a method for inverting underground anomalies from well induced polarization data based on deep learning, which belongs to the field of electrical prospecting.

[0034] Due to the induced polarization effect, electrifying a geological body containing anomalies generates secondary field anomalies. These secondary field anomalies are detected by receiving electrodes in the well. Forward modeling involves calculating the secondary field anomalies generated by the electrified anomaly based on the known location and size of the anomaly. Inversion involves inferring the location and size of the anomaly from these secondary field anomalies. Forward modeling uses the finite difference method to calculate the secondary field anomaly values for each grid node. To simulate the situation of receiving electrodes in the well, all secondary field anomaly values at the grid nodes at the well location are extracted, thus simulating the distribution of receiving electrodes in the well. The dataset construction step involves generating a large amount of wellbore induced polarization data and the corresponding anomaly locations and sizes. This involves performing stochastic modeling to generate numerous geological models containing the anomalies.

[0035] The present invention infers the distribution and size of underground anomalies through the outliers of the secondary field in the well, that is, the in-well induced polarization data. Therefore, the in-well induced polarization data can be calculated, and the position and size of the corresponding anomaly distribution are used as label data for deep learning training. The trained network can output the position and size of the corresponding anomaly by inputting the in-well induced polarization data, thereby achieving the purpose of inversion. The purpose of constructing the data set is to perform random geological body modeling, and record the position and size of the anomaly modeled each time as the label data of the in-well induced polarization data calculated by this geological body model. After multiple modeling and calculations, a large amount of induced polarization data and label data will be generated, which are used to construct the data set.

[0036] The calculation formula of the secondary field potential value is:

[0037] Secondary field potential value = polarization field potential value - primary field potential value

[0038] The calculation formula of the secondary field outlier value is:

[0039] Secondary field anomaly value = secondary field potential value - background polarizability × polarization field potential value

[0040] Since forward modeling simulates the secondary field anomaly of a geological body under the induced polarization effect under real conditions, the background polarizability and anomalous body polarizability should be variable. However, this method is mainly used for secondary field anomaly inversion in a fixed area. In practice, these background polarizability, background conductivity, anomalous body conductivity polarizability, etc. should be known conditions obtained through other on-site methods. They are fixed values in the area and should not be used as inversion parameters. Therefore, similar fixed quantities not used as inversion variables and the values set in this article are: background polarizability 0.001, background conductivity 0.001, power supply current 3A, anomalous body polarizability 0.2, anomalous body conductivity 0.001. Finite difference is performed on these variables according to the given formula and grid division to obtain the results.

[0041] The present invention is further described below with reference to an embodiment. This embodiment provides a method for inverting underground anomalies from in-well induced polarization data based on deep learning, which belongs to the field of electrical prospecting.

[0042] First, the geological body containing the anomaly is modeled and the corresponding in-well induced polarization data is calculated. Secondly, the center position and size of the ellipsoidal anomaly are randomly set and forward calculations are performed. The center position and size parameters of the ellipsoidal anomaly are recorded as label data to obtain trainable data and construct a data set. Third, a network is constructed and trained using an existing data set. After training, the network's predictions are tested and the accuracy and error of the network's inversion are calculated. The flowchart is as follows: Figure 1 Shown, including:

[0043] S1 performs three-dimensional geological modeling and generates a three-dimensional model within the defined range; this three-dimensional model is used to simulate the underground distribution under real-world conditions. Assuming that the underground medium is evenly distributed and there is an anomaly, the three-dimensional model can be represented as a region with different resistivity and polarizability within a medium with the same resistivity and polarizability. By gridding the entire geological body at equal intervals and dividing it into several small rectangular blocks, all grids can be assigned the resistivity and polarizability of the underground medium, and the grids that meet the location of the anomaly can be replaced with the resistivity and polarizability of the anomaly, thus simulating the distribution of the underground anomaly. Figure 5 A geological model with an ellipsoidal anomaly constructed for this modeling.

[0044] S12: Use the finite difference method to calculate the secondary field anomaly at all grid points. Record the coordinates of the well location and extract the secondary field anomaly at all grid points perpendicular to the well location. This is the well IP data. The resulting well IP data is an 81x1 array. Figure 4 This is a schematic diagram of the scope of the geological model and the location of the well. Figure 5The corresponding position and size parameters of the ellipsoidal anomaly are (78, 53.640000, 9.403424, 11.912847, 38.048250). The forward calculation results of the wellbore IP data array are as follows: Figure 6 As shown, (78, 53.640000, 9.403424, 11.912847, 38.048250) is used as Figure 6 Shows the label data of the induced polarization data in the well.

[0045] S2 builds the dataset: According to step S1, randomly set an ellipsoidal anomaly that meets the conditions, and record the axis length of the spherical center position set in this step as the label data for the corresponding calculated well induced polarization data. Randomly generate a geological body model and calculate to generate a labeled trainable data. Repeat this step to build the dataset. The dataset used this time has a total of 50,000 labeled data as shown in the figure. The survey area range is 100mx100mx200m, the grid spacing is 2.5m, the well location is (50m, 75m), the y-direction position of the fixed anomaly is 50m, the current is 3A, of which the training set accounts for 80% and the validation set accounts for 20%. Before inputting the grid training, the input network data was normalized using the mean and standard deviation.

[0046] Selection and Training of the S3 Network: The network used in this study consists of a deep learning network consisting of seven fully connected layers, activated using the RELU activation function. A dropout of 0.1 was added after each fully connected layer to prevent overfitting, and the Adam optimizer was used. The number of input features equals the number of channels in the inversion input data, and the number of output features equals the five inversion parameters. After one round of training, the network is fed with data and the error is backpropagated. Training is terminated when the objective function, which is the root mean square error between the true and predicted values, reaches its optimal value. Training is terminated when the loss function stabilizes, indicating that the network has learned the relationship between the wellbore IP data and the location and size parameters of the anomaly.

[0047] The network used in this study consists of seven fully connected layers. The number of nodes in the input layer is determined by the number of features in the input data. This inversion is performed using borehole induced polarization data from 81 grid points, so the input layer has 81 nodes. The first hidden layer has 64 nodes, the second hidden layer has 128 nodes, the third hidden layer has 256 nodes, the fourth hidden layer has 512 nodes, the fifth hidden layer has 256 nodes, and the sixth hidden layer has 128 nodes. The number of nodes in the output layer is determined by the number of features in the output data. The inversion output is five inversion parameters, so the output layer has five nodes. The RELU activation function is used for activation, with a dropout of 0.1 added after each fully connected layer to prevent overfitting. The Adam optimizer is used for optimization. The number of input features is the number of channels in the inversion input data, and the number of output features is the five inverted parameters. After one round of training, the network is trained with data, and the error is backpropagated. Training is terminated when the objective function, which is the root mean square error between the true and predicted values, reaches its optimal value. When the loss function decreases to a stable level, the training is stopped and the network is considered to have learned the relationship between the wellbore induced polarization data and the position and size parameters of the anomaly.

[0048] S4 inputs new borehole IP data for inversion verification: At this point, the network has learned the nonlinear mapping relationship between borehole IP data and the position and size parameters of the anomaly. By inputting the borehole IP data, the network can output the position and size parameters of the anomaly center. Figure 7 For the well induced polarization data shown, the network output result is (31.323974, 103.278754, 26.508672, 51.708690). Since the y-direction position of the underground anomaly body is fixed at 50m, the inverted underground anomaly body position is (31.323974, 50.000000, 103.278754), and the axis lengths in the x, y, and z directions are (27.08297520.224374 52.118200) respectively. Since the inversion verification is performed using synthetic IP data, the location and size parameters of the underground anomaly are known. The true value of the sphere center position is (32.693242, 50.000000, 105.197839), and the true value of the axis length is (27.082975, 20.224374, 52.118200). It can be seen that the absolute error is within an acceptable range. The validation set used this time is 500 labeled well IP data. Figure 8 The inversion values of 50 randomly selected data points for this validation fit the true values. Table 1 shows the average error of this validation. The absolute errors in this inversion were all within the distance of two grid cells, indicating that the method is effective. This invention can be successfully applied to the field of electrical prospecting data processing and can be used in electrical prospecting.

[0049] At this point, the network has learned the nonlinear mapping relationship between the wellbore IP data and the position and size parameters of the anomaly. The network can output the position and size parameters of the anomaly center by inputting the wellbore IP data. Figure 7 For the well induced polarization data shown, the network output result is (31.323974, 103.278754, 26.508672, 18.46785, 51.708690). Since the y-direction position of the underground anomaly body is fixed at 50m, the inverted underground anomaly body position is (31.323974, 50.000000, 103.278754), and the axis lengths in the x, y, and z directions are (26.508672, 18.46785, 51.708690), respectively.

[0050] Since synthetic IP data is used for inversion verification, the position and size parameters of the underground anomaly are known. The true value of the sphere center position is (32.693242, 50.000000, 105.197839), and the true value of the axis length is (27.082975, 20.224374, 52.118200). The error calculation formula is:

[0051] Absolute error = |true value - network inversion value|

[0052] The absolute error can be calculated: (1.369268,0,1.919085,0.574303,1.756524,0.409510)

[0053] The average error formula is:

[0054]

[0055] The average error of a single piece of data can be calculated as: 1.00478167.

[0056] The grid spacing is 2.5m. It can be seen that the absolute error is within the two grid spacings, and the accuracy meets the requirements. The validation set used this time is 500 labeled borehole induced polarization data. The average value of the average error of the 500 inversions in this validation is calculated as follows:

[0057]

[0058] The error distribution is calculated by sorting the average errors of 500 data points, selecting the 25th and 75th percentile errors for review, and preventing the presence of outliers such as excessively large or small errors from affecting the overall error level of the assessment.

[0059] Figure 8The inversion values of 50 randomly selected data points for this validation fit the true values. Table 1 shows the average error of this validation. The absolute errors in this inversion were all within the distance of two grid cells, indicating that the method is effective. This invention can be successfully applied to the field of electrical prospecting data processing and can be used in electrical prospecting.

[0060] The embodiment of the present application provides a deep learning-based induced polarization inversion system, comprising:

[0061] The 3D geological body modeling module is used to perform 3D geological body modeling and generate a geological body model containing anomalies within a defined range. This model is used to simulate a geological model including a fixed well. Based on the position and size of the anomaly in the geological body model, the secondary field anomaly value generated by the anomaly after power is applied is calculated, i.e., the induced polarization data of the fixed well at the corresponding position of the anomaly.

[0062] an induced polarization data set construction module, randomly generating a plurality of geological body models containing anomalies of different positions and sizes, and calculating the secondary field anomaly values generated by the anomalies in each geological body model of the anomalies of different positions and sizes according to the method in step S1, thereby obtaining a plurality of sets of in-well induced polarization data of fixed wells corresponding to the anomalies; and recording the center position and size of each anomaly and the corresponding in-well induced polarization data, and using the center position and size parameters of each anomaly as label data of the corresponding in-well induced polarization data to obtain a trainable labeled induced polarization data set, and repeating this step to construct a data set including a plurality of trainable labeled induced polarization data sets;

[0063] The borehole IP inversion module uses 7 fully connected layers to build a network and train the data set constructed in step S2. After the network training is completed, an independent test data set is input to evaluate the size and distribution of the error. When the test set error is within an acceptable range, the network training is considered complete. At this time, the borehole IP data can be input to output the corresponding predicted anomaly position and size parameters, achieving the inversion purpose.

[0064] An embodiment of the present application provides a computer-readable storage medium storing program code. When the program code is executed by a processor, the steps of the deep learning-based downhole induced polarization inversion method as described above are implemented.

[0065] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0066] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0067] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0069] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0070] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0071] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0072] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

[0073] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

Claims

1. A deep learning-based induced polarization inversion method, characterized in that: The following steps are involved: S1. Perform three-dimensional geological modeling to generate a geological model containing an anomaly within a defined range, which is used to simulate a geological model including a fixed well. Calculate the secondary field anomaly value generated by the anomaly after energization based on the position and size of the anomaly in the geological model, i.e., the induced polarization data of the fixed well at the corresponding position of the anomaly; S2. randomly generate multiple geological body models containing anomalies of different positions and sizes, calculate the secondary field anomaly value generated by the anomaly in each geological body model of the anomaly of different positions and sizes according to the method in step S1, and obtain multiple sets of wellbore induced polarization data of fixed wells corresponding to the anomalies; and record the center position and size of each anomaly and the corresponding wellbore induced polarization data, use the center position and size parameters of each anomaly as the label data of the corresponding wellbore induced polarization data, and obtain a trainable labeled induced polarization data set, and repeat this step to construct a data set including multiple trainable labeled induced polarization data; S3. Use seven fully connected layers to build a network and train the dataset constructed in step S2. After the network training is completed, an independent test dataset is input to evaluate the size and distribution of the error. When the test set error is within an acceptable range, the network training is considered complete. At this time, the input of the wellbore induced polarization data can output the corresponding predicted anomaly position and size parameters, achieving the purpose of inversion.

2. The deep learning-based induced polarization inversion method according to claim 1, characterized in that: In the step S1, the three-dimensional geological body model is generated to generate a geological body model including an ellipsoidal anomaly within a defined range. The specific method for constructing the geological body model including the ellipsoidal anomaly is as follows: According to the demarcated range, a corresponding rectangular parallelepiped geological model is constructed, the resistivity and polarizability of the rectangular parallelepiped geological model are set to a fixed value, and the rectangular parallelepiped geological model is gridded; a point A is arbitrarily determined as the center of the ellipsoid anomaly, and the axis length parameters a, b, and c of the ellipsoid anomaly are set. The position coordinates of the above-determined point A in the rectangular parallelepiped geological model are used as the center coordinates (h, k, l) of the ellipsoid anomaly, and the axis length parameters a, b, and c of the ellipsoid anomaly are set. According to the following spatial ellipsoid range calculation formula, all grid nodes that meet the formula requirements are calculated: in: x, y, z represent the relative coordinate position of the node inside the geological body, h, k, l are the coordinates of the center of the ellipsoid; a, b, c are the axis lengths of the ellipsoid in the x, y, z directions; All grid nodes in the model that meet the above formula are traversed. The small cuboids where these nodes are located can be combined to approximate an ellipsoid anomaly that meets the conditions. All small cuboids in the approximate ellipsoid are replaced with the resistivity and polarizability of the anomaly to construct a geological model that includes an ellipsoid anomaly that meets the above requirements.

3. The deep learning-based induced polarization inversion method according to claim 2, characterized in that: In the S3 step, a network is constructed by using a fully connected layer, and the constructed data set is normalized using the Lure activation function, the Semigloss loss function and the Adam optimizer, and the average value and the root mean square are used, and then it is sent to the neural network; the input data of the neural network model is the in-well induced polarization data of the fixed well corresponding to the anomaly, and the output data is the center coordinates of the anomaly location and the axis lengths in the X, Y, and Z directions as reference labels for the output prediction values; the optimal number of training iterations is set, and the error of each round of training is back-propagated to the network, so that the network continuously optimizes the neural network model parameters to best fit the nonlinear mapping relationship between the in-well induced polarization data and the center coordinates and axis lengths of the underground anomaly. The training is stopped when the loss between the sample prediction value and the true value in the verification set decreases and tends to be stable after training; at this time, the network has learned the nonlinear mapping relationship between the input data and the output data, and the inverted anomaly position and axis length can be obtained after the induced polarization data is input into the network.

4. The deep learning-based induced polarization inversion method according to claim 2, characterized in that: The distance between the ellipsoid anomaly and the boundary of the rectangular geological body model should be no less than 2 grids; the ellipsoid anomaly should not be exposed on the surface or other boundaries; the length of each axis of the ellipsoid anomaly should be no less than 1 grid length.

5. A deep learning-based IP inversion system, characterized by: include, The 3D geological body modeling module is used to perform 3D geological body modeling and generate a geological body model containing anomalies within a defined range. This model is used to simulate a geological model including a fixed well. Based on the position and size of the anomaly in the geological body model, the secondary field anomaly value generated by the anomaly after power is applied is calculated, i.e., the induced polarization data of the fixed well at the corresponding position of the anomaly. an induced polarization data set construction module, randomly generating a plurality of geological body models containing anomalies of different positions and sizes, and calculating the secondary field anomaly values generated by the anomalies in each geological body model of the anomalies of different positions and sizes according to the method in step S1, thereby obtaining a plurality of sets of in-well induced polarization data of fixed wells corresponding to the anomalies; and recording the center position and size of each anomaly and the corresponding in-well induced polarization data, and using the center position and size parameters of each anomaly as label data of the corresponding in-well induced polarization data to obtain a trainable labeled induced polarization data set, and repeating this step to construct a data set including a plurality of trainable labeled induced polarization data sets; The borehole IP inversion module uses 7 fully connected layers to build a network and train the data set constructed in step S2. After the network training is completed, an independent test data set is input to evaluate the size and distribution of the error. When the test set error is within an acceptable range, the network training is considered complete. At this time, the borehole IP data can be input to output the corresponding predicted anomaly position and size parameters, achieving the inversion purpose.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program code, and when the program code is executed by a processor, the steps of the deep learning-based downhole induced polarization inversion method according to any one of claims 1 to 4 are implemented.

Images