A method and device for inverting airborne transient electromagnetic data resistivity based on genetic algorithm

Abstract

The application belongs to the technical field of geophysical signal processing and analysis, and specifically discloses an airborne transient electromagnetic data resistivity inversion method and device based on a genetic algorithm, which comprises the following steps: taking a resistivity model and a transmitting coil height as input, taking airborne transient electromagnetic observation response data as output, and training a deep learning network model to obtain a forward simulator; determining the optimal population number of the genetic algorithm according to the inflection point of the efficiency curve of the forward simulator, then generating an initial population by using a sliding smooth population sampling method, and forming a continuous resistivity model after processing by cubic spline interpolation; coupling the forward simulator to a parallel genetic algorithm parameter inversion framework to generate a target network model; and based on the inversion framework, performing resistivity inversion on the analyzed airborne transient electromagnetic data to obtain multiple groups of smooth resistivity models that meet data fitting differences. The application can significantly improve the reliability of transient electromagnetic data resistivity inversion.

Description

Technical Field

[0001] This application belongs to the field of geophysical signal processing and analysis technology, and more specifically, relates to a method and apparatus for resistivity inversion of airborne transient electromagnetic data based on genetic algorithms. Background Technology

[0002] Transient electromagnetic (TEM) is a geophysical method that uses an ungrounded or grounded source to emit a primary pulse magnetic field into the subsurface. During the intervals between the primary pulses, coils or grounded electrodes are used to observe the secondary induced eddy current field generated in the subsurface medium, thereby probing the medium. By analyzing the decay characteristics of this secondary induced eddy current field over time, the resistivity distribution of geological bodies can be studied, solving problems in geological exploration, resource detection, and other fields. Inverting the subsurface resistivity from the secondary induced eddy current field is a crucial step in transient electromagnetic exploration. Its purpose is to reconstruct model parameters (such as layer thickness and resistivity) that cannot be directly observed from the measured transient electromagnetic response data. However, this inversion process is usually highly nonlinear and has multiple solutions. Traditional gradient or local search algorithms often only provide a single optimal solution and struggle to comprehensively explore the full probability distribution of the model parameter space, limiting the reliability assessment of the inversion results and potentially even missing key geological targets.

[0003] Therefore, improving the reliability of resistivity inversion from transient electromagnetic data is a pressing problem that urgently needs to be addressed. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this application is to provide a method and apparatus for resistivity inversion of airborne transient electromagnetic data based on genetic algorithms, which can significantly improve the reliability of resistivity inversion of transient electromagnetic data.

[0005] To achieve the above objectives, in a first aspect, this application provides a method for resistivity inversion of airborne transient electromagnetic data based on a genetic algorithm, comprising the following steps:

[0006] S10, construct a synthetic dataset, and use the preset resistivity model and the height of the transmitting coil as input data, and the corresponding airborne transient electromagnetic observation response data as output data to train and optimize the deep learning network model to obtain a forward modeling simulator;

[0007] S20, plot the call efficiency curve of the forward simulator, determine the optimal population size of the genetic algorithm based on the inflection point of the curve, generate the initial population using the sliding smooth population sampling method, and form a continuous resistivity model after processing by cubic spline interpolation; wherein, each individual in the population consists of the resistivity values ​​and corresponding positions of several key nodes, and the resistivity values ​​and positions of each node are evenly distributed in the population.

[0008] S30, the forward modeling simulator is coupled to the parallel genetic algorithm parameter inversion framework to generate the target network model;

[0009] S40, based on the genetic algorithm parameter inversion framework, resistivity inversion is performed on the aerospace transient electromagnetic data to be analyzed to obtain multiple sets of smooth resistivity models that satisfy the data fitting difference.

[0010] This application provides a genetic algorithm-based resistivity inversion method for airborne transient electromagnetic data. By embedding a deep learning forward modeling simulator into an optimized genetic algorithm inversion system, it not only accelerates the forward modeling simulation but also promotes rapid global search and multiple solution coexistence for airborne transient electromagnetic data, effectively improving inversion efficiency and result reliability. The difference between this method and deep learning-based Bayesian inversion methods lies in the fact that this application automatically determines the population size inflection point using the efficiency curve output from the forward modeling simulator in the genetic algorithm and dynamically adjusts the crossover and mutation probabilities in each generation. Furthermore, this application proposes a "sliding smooth" population sampling strategy to ensure that the inversion results present a multi-level continuous distribution and can be based on multiple solution sets.

[0011] As a further preferred option, step S10 specifically includes:

[0012] The resistivity model, transmitting coil height, and electromagnetic response data are normalized to generate normalized data.

[0013] Based on a preset magnetic field excitation intensity threshold, the normalized data is divided into low magnetic moment excitation data and high magnetic moment excitation data;

[0014] The deep learning network model is trained and optimized using the low magnetic moment excitation data and the high magnetic moment excitation data respectively to obtain the forward simulation model.

[0015] As a further preferred option, step S10 also includes:

[0016] The deep learning network model is trained iteratively by using the mean absolute error and squared error as loss functions until the iteration stopping condition is met, thereby obtaining the optimal model hyperparameters and forward simulation model.

[0017] As a further preferred embodiment, the hyperparameters of the deep learning network model are optimized through manual tuning or random search, and the hyperparameters include the number of network layers and the number of neurons.

[0018] As a further optimization, the optimal population size for inversion efficiency can be found by plotting the curve of forward modeling time versus the number of models participating in the forward modeling simulation.

[0019] As a further preferred embodiment, in the sliding smooth population sampling method, the first few values ​​of each population sample are used as the resistivity values ​​of key nodes, and the last few values ​​represent the position of the node in the profile.

[0020] As a further preferred option, in step S30, the genetic algorithm framework adopts the Python open-source code Geatpy2.

[0021] Secondly, this application provides a device for retrieving resistivity from airborne transient electromagnetic data based on a genetic algorithm, comprising:

[0022] The training module is used to construct a synthetic dataset, and takes a preset resistivity model and the height of the transmitting coil as input data, and the corresponding airborne transient electromagnetic observation response data as output data to train and optimize the deep learning network model to obtain a forward modeling simulator.

[0023] The population optimization module is used to plot the call efficiency curve of the forward simulator, determine the optimal population size of the genetic algorithm based on the inflection point of the curve, generate the initial population using a sliding smooth population sampling method, and form a continuous resistivity model after cubic spline interpolation. Each individual in the population consists of the resistivity values ​​and corresponding positions of several key nodes, and the resistivity values ​​and positions of each node are evenly distributed in the population.

[0024] A generation module is used to couple the forward modeling simulator to a parallel genetic algorithm parameter inversion framework to generate a target network model;

[0025] The inversion module is used to perform resistivity inversion on the aerospace transient electromagnetic data to be analyzed based on the genetic algorithm parameter inversion framework, and obtain multiple sets of smooth resistivity models that satisfy the data fitting difference.

[0026] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the airborne transient electromagnetic data resistivity inversion method based on genetic algorithm as described in any one of the above.

[0027] Fourthly, this application provides a computer-readable storage medium storing computer instructions thereon, which, when executed by a processor, implement the airborne transient electromagnetic data resistivity inversion method based on genetic algorithm as described in any one of the above.

[0028] It is understandable that the beneficial effects of the second, third and fourth aspects mentioned above can be found in the relevant descriptions in the first aspect above, and will not be repeated here. Attached Figure Description

[0029] Figure 1This is a flowchart of the resistivity inversion method for airborne transient electromagnetic data based on genetic algorithms provided in this application;

[0030] Figure 2 This is a flowchart of the resistivity inversion method for airborne transient electromagnetic data based on genetic algorithms provided in the embodiments of this application;

[0031] Figure 3 This is a schematic diagram illustrating the population size for determining optimal efficiency, provided in an embodiment of this application.

[0032] Figure 4 This is a schematic diagram of the sliding smooth population sample sampling strategy provided in the embodiments of this application;

[0033] Figure 5 This is a schematic diagram of the optimal population provided in the embodiments of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0035] It should be understood that, in the description of this application, the term "several" means at least one, such as one, two, etc., unless otherwise expressly and specifically defined.

[0036] like Figure 1 As shown, this application provides a method for resistivity inversion of airborne transient electromagnetic data based on genetic algorithms, including the following steps:

[0037] Step S10: Construct a synthetic dataset, and use the preset resistivity model and the height of the transmitting coil as input data, and the corresponding airborne transient electromagnetic observation response data as output data to train and optimize the deep learning network model, thereby obtaining a forward modeling simulator with millisecond-level call efficiency.

[0038] In step S10, the resistivity model characterizes the resistivity distribution of the underground medium, and the transmitting coil height characterizes the flight altitude of the airborne transient electromagnetic system.

[0039] Step S20: Plot the call efficiency curve of the forward simulator, determine the optimal population size of the genetic algorithm based on the inflection point of the curve, generate the initial population using the sliding smooth population sampling method, and form a continuous resistivity model after processing by cubic spline interpolation.

[0040] Each individual in the population consists of the resistivity values ​​and corresponding positions of several key nodes, and the resistivity values ​​and positions of each node are evenly distributed throughout the population.

[0041] In step S20, this application combines the millisecond-level call efficiency of the deep learning forward simulator to draw an efficiency curve, and selects the population size corresponding to the inflection point of the curve as the optimal size, thereby maximizing the inversion efficiency while ensuring global search capability.

[0042] By optimizing the model generation method of conventional genetic algorithms, a population sampling method with multi-layer smooth model characteristics and covering a large range of model space is obtained: the first few values ​​of each population sample are used as the resistivity values ​​of key nodes, and the last few values ​​represent the position of the node in the profile, forming a "smooth" population sample. This method can cover most of the model space and can also achieve large-scale and efficient evaluation by relying on millisecond-level deep learning forward simulators.

[0043] By organically linking the above two steps, this application constructs an end-to-end efficient inversion system: First, a deep learning forward modeling simulator trained on a resistivity model and the height of the transmitting coil accurately reproduces the electromagnetic response at millisecond speeds; second, relying on the efficient invocation of this model, efficiency curves are plotted for different population sizes to lock in the optimal population size, ensuring that the genetic algorithm obtains the best search capability within a limited time; finally, a "sliding smooth" population sampling method is used to generate multi-layered continuous and widely covered initial solutions, enabling the genetic algorithm to efficiently converge to the optimal solution of the geological structure at the optimal scale with extremely low forward modeling cost, supplemented by elite preservation and diversity maintenance strategies. The synergy of these three aspects ensures both the speed and accuracy of the forward modeling simulation while maximizing the efficiency and reliability of global optimization.

[0044] Step S30: Couple the forward modeling simulator to the parallel genetic algorithm parameter inversion framework to generate the target network model.

[0045] It should be noted that genetic algorithms are a global optimization technique that simulates the natural evolution process. By performing gene encoding, selection, crossover and mutation operations on candidate solutions (individuals), a set of solutions with higher fitness is continuously evolved, which can effectively overcome the defect of traditional methods that are prone to getting trapped in local optima.

[0046] This application applies a genetic algorithm to transient electromagnetic inversion, which can search multiple resistivity models in parallel in the initial population, fully characterize the uncertainty, and evaluate the degree of fit between each model and the observation data through the fitness function, so as to realize the global search and multiple solutions for nonlinear, multi-peak optimization problems.

[0047] Based on this, this application proposes a method that combines a deep learning forward modeling simulator with an optimized genetic algorithm parameter inversion framework. This not only accelerates the forward modeling simulation but also promotes efficient exploration of the parameter space and acquisition of multiple solutions, thereby significantly improving the efficiency and reliability of transient electromagnetic data resistivity inversion.

[0048] S40, based on the genetic algorithm parameter inversion framework, resistivity inversion is performed on the aerospace transient electromagnetic data to be analyzed, and multiple sets of smooth resistivity models that satisfy the data fitting difference are obtained.

[0049] This application provides a genetic algorithm-based resistivity inversion method for airborne transient electromagnetic data. By embedding a deep learning forward modeling simulator into an optimized genetic algorithm inversion system, it not only accelerates the forward modeling simulation but also promotes rapid global search and multiple solution coexistence for airborne transient electromagnetic data, effectively improving inversion efficiency and result reliability. The difference between this method and deep learning-based Bayesian inversion methods lies in the fact that this application automatically determines the population size inflection point using the efficiency curve output from the forward modeling simulator in the genetic algorithm and dynamically adjusts the crossover and mutation probabilities in each generation. Furthermore, this application proposes a "sliding smooth" population sampling strategy to ensure that the inversion results present a multi-level continuous distribution and can be based on multiple solution sets.

[0050] The present application will now be described in detail with reference to specific embodiments.

[0051] This embodiment provides a method for resistivity inversion based on airborne transient electromagnetic data using a genetic algorithm, such as... Figure 2 As shown, the main steps include dataset preparation and preprocessing, forward network construction, genetic algorithm optimization, and the combination of the optimized genetic algorithm with deep learning forward simulation.

[0052] The specific steps for dataset preparation and preprocessing provided in this embodiment are as follows:

[0053] Dataset preparation and preprocessing are crucial steps in the airborne transient electromagnetic inversion method based on genetic algorithms, directly impacting the accuracy of the inversion results. The resistivity model and flight altitude serve as output labels and efficiency. To ensure the quality and reliability of the inversion model, meticulous work is required in data construction, processing, standardization, and denoising. This process includes multiple steps such as collecting data from open-source databases, simulating electromagnetic responses using forward modeling formulas, and data standardization. In this embodiment, the synthetic dataset used for training originates from the research findings of Asif MR, Foged N, Bording T, et al., entitled "DL-RMD: a geophysically constrained electromagnetic resistivity model database (RMD) for deep learning (DL) applications," published in *Earth System Science Data*. This database provides geophysically constrained electromagnetic resistivity model data, widely used in deep learning model training. In this public database, all resistivity models have been smoothed, and the resistivity range has been adjusted to [1, 20,000 Ωm]. The corresponding airborne transient electromagnetic (ATEM) data for these models were generated using AarhusInv software, with flight altitudes ranging from [4, 200 m]. This synthetic dataset contains 20,001 samples, each including a resistivity model, ATEM data, and flight altitude information. In the training process of this embodiment, the resistivity model and flight altitude serve as input labels, while the ATEM data serves as output features, used to train and optimize the deep learning model.

[0054] The basic formula for forward modeling is, under the assumption of a one-dimensional model, for a magnetic moment of... The vertical magnetic dipole source is located on the ground. At that point, its particular solution of the first-order field satisfies the following differential equation:

[0055] (1)

[0056] In the horizontally layered model (i.e., the one-dimensional model), current can only flow horizontally. The potential function of a vertical magnetic dipole source. It can be written as follows:

[0057] (2)

[0058] The expression after Hankel transformation is as follows:

[0059] (3)

[0060] This embodiment employs a sinusoidal transform digital filtering algorithm to convert the frequency domain electromagnetic response to the time domain electromagnetic response, ultimately yielding a numerical simulation of the electromagnetic field response. Through this process, the open-source resistivity model and the forward simulation response are combined to form a dataset for preprocessing.

[0061] The preprocessing method used in this embodiment is as follows: the normalization formula for the resistivity model is:

[0062] (4)

[0063] The normalization formula for the height of the transmitting coil is:

[0064] (5)

[0065] The normalization formula for ATEM data is:

[0066]

[0067] (6)

[0068] This completes the construction of a normalized forward modeling dataset for network training.

[0069] The specific steps for constructing the forward modeling network provided in this embodiment are as follows:

[0070] After data preprocessing, deep learning network models are trained using low-magnetic-moment excitation data and high-magnetic-moment excitation data, respectively. Each set of data is used to train different network weights and parameters. During training, the model is optimized using the backpropagation algorithm, enabling it to accurately predict the relationship between electromagnetic response data and input features (resistivity model and transmitting coil height). Through multiple rounds of iterative training, the optimized network model is finally obtained. Due to the uncertainty of hyperparameter tuning, this embodiment does not provide fixed network structure information. Different tasks may require adjustments to the network structure, including hyperparameters such as the number of hidden layers and neurons, based on the characteristics of the data. Hyperparameter tuning typically employs methods such as grid search, random search, or Bayesian optimization to find the optimal network configuration. After the forward network is constructed, the optimal population size is tested to obtain the optimal efficiency. Figure 3 As shown, the efficiency advantage of forward simulation provided by deep learning models reaches its maximum when there are 2000 models, which is also the optimal population size for inversion.

[0071] The optimization steps of the genetic algorithm provided in this embodiment are as follows:

[0072] Genetic Algorithm (GA) is an optimization algorithm that simulates the biological evolution process in nature. Based on natural selection mechanisms such as selection, crossover, and mutation, it is used to solve optimization problems. Its basic idea is to gradually improve the fitness of individuals in the population by mimicking the genetic process of natural selection, ultimately finding the global optimum. Genetic algorithms have wide applications in optimization problems, especially when searching large-scale solution spaces. Through its unique population search mechanism, the genetic algorithm can effectively find potential optimal solutions. The fitness function of a genetic algorithm is typically calculated using the following formula:

[0073] (7)

[0074] Here, f(x) represents the fitness value of individual x, and the fitness function is used to measure how well an individual performs in the current environment. In electromagnetic inversion problems, the fitness value is usually represented by the error or goodness of fit of the electromagnetic response, that is, the difference between the predicted electromagnetic response and the actual observed data.

[0075] In genetic algorithms, the first step is to initialize a population. Each individual in the population represents a potential solution to the problem, typically represented by a vector or string. In resistivity inversion problems, an individual can be represented as a set of parameters of the resistivity model or characteristics of the electromagnetic response. The performance of each individual is evaluated using a fitness function, which measures the quality of the individual's solution. In genetic algorithms, the fitness function usually depends on the value of the objective function, which typically represents the goodness of fit between the model and the observed data. Individuals with smaller errors will have higher fitness values.

[0076] After fitness evaluation, individuals with higher fitness are selected as parents through selection operations, followed by crossover and mutation operations. Crossover mimics the process of gene recombination in nature, generating new offspring by exchanging gene portions of parent individuals, thus producing new potential solutions. Mutation simulates the process of gene mutation, randomly altering certain gene values ​​of individuals to increase population diversity and prevent the algorithm from getting trapped in local optima. Through continuous crossover and mutation operations, the individuals in the population are gradually optimized, ultimately finding the optimal solution.

[0077] However, while traditional genetic algorithms can effectively explore the solution space, their search efficiency and ability to find global solutions are often limited when facing high-dimensional and complex problems. To overcome this problem, this embodiment optimizes the genetic algorithm, particularly in population generation and evaluation.

[0078] To address the issues of population concentration and low search efficiency in traditional genetic algorithms, this embodiment proposes a "sliding smooth" population sampling method. This method combines a sliding window (e.g., ...) Figure 4As shown in the figure, the first six values ​​represent resistivity, and the last four represent the locations of these resistivity values. Based on this information, a smooth model of arbitrary layer thickness can be interpolated for use in forward and inverse simulations. The Latin hypercube sampling method ensures that the initial population broadly covers the solution space during the initial population generation stage. The resistivity model of each individual is represented by the resistivity values ​​and corresponding locations of several key nodes, and the resistivity values ​​and locations of each node are uniformly distributed throughout the population. This approach guarantees the uniformity of the initial population across the entire resistivity model space and effectively covers most of the potential solution region. This optimization not only improves population diversity but also avoids the over-concentration problem that may occur in traditional methods.

[0079] After population generation, a smoothing method based on cubic spline interpolation was employed. For each individual, the resistivity model was transformed into a continuous, smooth layered resistivity profile using interpolation, ensuring the model's physically reasonable continuity. Cubic spline interpolation effectively smooths the resistivity model, eliminating discontinuities caused by discrete individual resistivity values. The continuous model obtained after interpolation more accurately reflects the actual resistivity distribution of the subsurface medium and provides more reasonable input for subsequent electromagnetic response simulations.

[0080] Combination of optimized genetic algorithm with deep learning forward simulation

[0081] In this embodiment, the combination of the optimized genetic algorithm and the deep learning forward modeling simulator constitutes the core of resistivity inversion. The basic flow of the inversion process is as follows:

[0082] First, an initial population is generated using a "sliding smooth" population sampling method. Each individual in the population consists of resistivity values ​​and corresponding positions of several key nodes, which are processed by cubic spline interpolation to form a continuous resistivity model. Next, each population sample is rapidly evaluated using a deep learning forward model simulator to calculate its electromagnetic response data.

[0083] The difference between the electromagnetic response of each individual model and the actual observation data is evaluated using a fitness function, which reflects the goodness of fit between the model and the observation data. Based on the fitness evaluation results, a genetic algorithm performs selection, crossover, and mutation operations on the population, optimizing the resistivity model generation by generation. The population is updated at each generation, gradually converging towards the global optimum.

[0084] Figure 5The test results are presented, showing the closeness between the inversion results of the optimal individual and the target resistivity model. As can be seen from the figures, the resistivity structure of the optimal individual is very similar to the target model, proving that the optimized genetic algorithm can accurately reconstruct the subsurface resistivity profile. Furthermore, the results of the optimal population indicate that the genetic algorithm can not only provide a single optimal solution but also reveal uncertainties in the resistivity structure to a certain extent, providing multiple solutions that meet the data fitting requirements. This pluralistic approach provides an effective means for uncertainty assessment in practical applications.

[0085] By combining a deep learning forward modeling simulator with a genetic algorithm, leveraging the high computational efficiency of the simulator and the global search capability of the genetic algorithm, we successfully constructed a genetic algorithm framework based on deep learning forward modeling. This approach not only improves the computational efficiency of resistivity inversion but also effectively explores and optimizes the solution space of the resistivity model while maintaining model accuracy, ultimately providing an efficient and accurate solution for resistivity inversion.

[0086] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for resistivity inversion from airborne transient electromagnetic data based on a genetic algorithm, characterized in that, Includes the following steps: S10, construct a synthetic dataset, and use the preset resistivity model and the height of the transmitting coil as input data, and the corresponding airborne transient electromagnetic observation response data as output data to train and optimize the deep learning network model to obtain a forward modeling simulator; S20, plot the call efficiency curve of the forward simulator, determine the optimal population size of the genetic algorithm based on the inflection point of the curve, generate the initial population using the sliding smooth population sampling method, and form a continuous resistivity model after processing by cubic spline interpolation; wherein, each individual in the population consists of the resistivity values ​​and corresponding positions of several key nodes, and the resistivity values ​​and positions of each node are evenly distributed in the population. S30, the forward modeling simulator is coupled to the parallel genetic algorithm parameter inversion framework to generate the target network model; S40, based on the genetic algorithm parameter inversion framework, resistivity inversion is performed on the aerospace transient electromagnetic data to be analyzed to obtain multiple sets of smooth resistivity models that satisfy the data fitting difference.

2. The resistivity inversion method for airborne transient electromagnetic data based on genetic algorithm as described in claim 1, characterized in that, Step S10 is as follows: The resistivity model, transmitting coil height, and electromagnetic response data are normalized to generate normalized data. Based on a preset magnetic field excitation intensity threshold, the normalized data is divided into low magnetic moment excitation data and high magnetic moment excitation data; The deep learning network model is trained and optimized using the low magnetic moment excitation data and the high magnetic moment excitation data respectively to obtain the forward simulation model.

3. The resistivity inversion method for airborne transient electromagnetic data based on genetic algorithm as described in claim 1, characterized in that, Step S10 also includes: The deep learning network model is trained iteratively by using the mean absolute error and squared error as loss functions until the iteration stopping condition is met, thereby obtaining the optimal model hyperparameters and forward simulation model.

4. The resistivity inversion method for airborne transient electromagnetic data based on genetic algorithm as described in claim 1, characterized in that, The hyperparameters of the deep learning network model are optimized through manual tuning or random search, and the hyperparameters include the number of network layers and the number of neurons.

5. The resistivity inversion method for airborne transient electromagnetic data based on genetic algorithm as described in claim 1, characterized in that, By plotting the curve of forward modeling time versus the number of models participating in the forward modeling simulation, we can find the population size that yields the best inversion efficiency.

6. The resistivity inversion method for airborne transient electromagnetic data based on genetic algorithm as described in claim 1, characterized in that, In the aforementioned smooth population sampling method, the first few values ​​of each population sample are used as the resistivity values ​​of key nodes, and the last few values ​​represent the position of the node in the profile.

7. The resistivity inversion method for airborne transient electromagnetic data based on genetic algorithm as described in claim 1, characterized in that, In step S30, the genetic algorithm framework uses the Python open-source code Geatpy2.

8. A device for resistivity inversion of airborne transient electromagnetic data based on genetic algorithm, characterized in that, include: The training module is used to construct a synthetic dataset, and takes a preset resistivity model and the height of the transmitting coil as input data, and the corresponding airborne transient electromagnetic observation response data as output data to train and optimize the deep learning network model to obtain a forward modeling simulator. The population optimization module is used to plot the call efficiency curve of the forward simulator, determine the optimal population size of the genetic algorithm based on the inflection point of the curve, generate the initial population using a sliding smooth population sampling method, and form a continuous resistivity model after cubic spline interpolation. Each individual in the population consists of the resistivity values ​​and corresponding positions of several key nodes, and the resistivity values ​​and positions of each node are evenly distributed in the population. A generation module is used to couple the forward modeling simulator to a parallel genetic algorithm parameter inversion framework to generate a target network model; The inversion module is used to perform resistivity inversion on the aerospace transient electromagnetic data to be analyzed based on the genetic algorithm parameter inversion framework, and obtain multiple sets of smooth resistivity models that satisfy the data fitting difference.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the resistivity inversion method for aerospace transient electromagnetic data based on genetic algorithms as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instruction is executed by the processor, it implements the resistivity inversion method for aerospace transient electromagnetic data based on genetic algorithms as described in any one of claims 1 to 7.

Images