Aeromagnetic data compensation method and device based on physical constraints, equipment and medium

Abstract

This application discloses a method, apparatus, equipment, and medium for aeromagnetic data compensation based on physical constraints, relating to the field of geophysical exploration technology. The method includes preprocessing the flight parameters of the acquired flight platform to obtain a multi-dimensional original feature set; performing interference compensation calculations on the multi-dimensional original feature set to obtain an initial compensation feature set; training a pre-constructed residual multilayer perceptron neural network based on the multi-dimensional original feature set until the constraint fitting residual of the residual multilayer perceptron neural network matches a preset first-order autocorrelation physical constraint, thus obtaining a trained residual multilayer perceptron neural network; inputting the multi-dimensional original feature set into the trained residual multilayer perceptron neural network to obtain aeromagnetic compensation data output by the residual multilayer perceptron neural network; and compensating the initial compensation feature set based on the aeromagnetic compensation data to determine the final aeromagnetic measurement data. This application can improve the compensation accuracy of aeromagnetic data.

Description

Technical Field

[0001] This application relates to the field of geophysical exploration technology, and in particular to a method, apparatus, equipment and medium for aeromagnetic data compensation based on physical constraints. Background Technology

[0002] Airborne magnetic surveying, with its advantages of wide coverage, high efficiency, and strong accessibility, has become an important tool for large-scale mineral resource exploration and geological structure surveys. However, the ferromagnetic structure of the flight platform itself, as well as the changing magnetic fields generated by airborne electronic equipment, engines, and other components during flight, can introduce significant measurement errors into the geomagnetic field data collected by the magnetometer sensors mounted on the platform. This magnetic interference field caused by the flight platform itself often exceeds the magnitude of the weak magnetic anomaly signal generated by the target geological body, severely limiting the accuracy of airborne magnetic data and the reliability of geological interpretation. Therefore, effective compensation techniques must be used to correct the measurement data post-hocly.

[0003] Since its introduction in 1950, the Tolles-Lawson (TL) compensation model has been the most widely used method in the field of aeromagnetic compensation. However, in practice, it has been found that this model is essentially a linear model and is difficult to effectively describe the nonlinear and non-stationary magnetic disturbances generated by the maneuvering flight of the flight platform. This results in limited compensation accuracy of aeromagnetic data and fails to meet the actual needs of high-precision aeromagnetic measurements. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, device, and medium for aeromagnetic data compensation based on physical constraints, which can improve the compensation accuracy of aeromagnetic data.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a physically constrained aeromagnetic data compensation method, including: The flight parameters of the acquired flight platform are preprocessed to obtain a multi-dimensional original feature set; wherein, the multi-dimensional original feature set includes multi-dimensional basic features and three-axis maneuvering attitude information; Interference compensation calculation is performed on the original multi-dimensional feature set to obtain the initial compensated feature set; The pre-constructed residual multilayer perceptron neural network is trained based on the multidimensional original feature set until the constraint fitting residual of the residual multilayer perceptron neural network matches the preset first-order autocorrelation physical constraint of the residual, thus obtaining the trained residual multilayer perceptron neural network. The multi-dimensional original feature set is input into the trained residual multilayer perceptron neural network to obtain the aeromagnetic compensation data output by the residual multilayer perceptron neural network. The initial compensation feature set is compensated based on the aeromagnetic compensation data to determine the final aeromagnetic measurement data.

[0006] Optionally, the preprocessing of the collected flight parameters of the flight platform to obtain a multi-dimensional original feature set specifically includes: The flight parameters of the flight platform are collected; wherein the flight parameters include at least magnetic measurement data and three-axis maneuvering attitude data; the magnetic measurement data is collected by a magnetometer pre-set on the flight platform, and the three-axis maneuvering attitude data is collected by an inertial navigation device pre-set on the flight platform; Multidimensional fundamental features are extracted from the magnetic measurement data; wherein, the multidimensional fundamental features include at least constant magnetic field features, induced magnetic field features, and eddy current magnetic field features; Three-axis maneuvering attitude information is extracted from the three-axis maneuvering attitude data; wherein, the three-axis maneuvering attitude information includes at least roll maneuvering attitude information, pitch maneuvering attitude information, and yaw maneuvering attitude information; The multidimensional basic features and the three-axis maneuvering attitude information are determined as a multidimensional original feature set.

[0007] Optionally, the multidimensional basic features include 18 basic features, the three-axis maneuvering attitude information includes 3 supplementary features, and the multidimensional original feature set includes 21 features.

[0008] Optionally, the residual multilayer perceptron neural network is constructed as follows: Construct a lightweight, fully connected residual multilayer perceptron neural network; The residual multilayer perceptron neural network has 21 input layer channels; two fully connected network layers are sequentially set after the input layer, and ReLU activation functions are configured for each of the two fully connected network layers; and an output layer is set after the two fully connected network layers, with 1 output layer channel.

[0009] Optionally, the step of compensating the initial compensation feature set based on the aeromagnetic compensation data to determine the final aeromagnetic measurement data specifically includes: Extract the compensated magnetic field strength data from the initial compensation feature set; Subtracting the aeromagnetic compensation data from the compensated magnetic field strength data yields the final aeromagnetic measurement data.

[0010] Optionally, after determining the final aeromagnetic measurement data, the method further includes: Perform standard deviation calculation on the original multi-dimensional feature set to obtain the standard deviation before compensation; The standard deviation of the aeromagnetic measurement data is calculated to obtain the compensated standard deviation; The compensation-improvement ratio is obtained by calculating the ratio of the standard deviation before compensation to the standard deviation after compensation. Based on the standard deviation before compensation, the standard deviation after compensation, and the compensation improvement ratio, an aeromagnetic data compensation evaluation result is generated.

[0011] Secondly, this application provides a physically constrained aeromagnetic data compensation device, comprising: The preprocessing unit is used to preprocess the flight parameters of the acquired flight platform to obtain a multi-dimensional original feature set; wherein, the multi-dimensional original feature set includes multi-dimensional basic features and three-axis maneuvering attitude information; The calculation unit is used to perform interference compensation calculation on the multi-dimensional original feature set to obtain the initial compensation feature set; The training unit is used to train the pre-constructed residual multilayer perceptron neural network based on the multidimensional original feature set until the constraint fitting residual of the residual multilayer perceptron neural network matches the preset residual first-order autocorrelation physical constraint, thereby obtaining the trained residual multilayer perceptron neural network. The input unit is used to input the multi-dimensional original feature set into the trained residual multilayer perceptron neural network to obtain the aeromagnetic compensation data output by the residual multilayer perceptron neural network. The compensation unit is used to compensate the initial compensation feature set based on the aeromagnetic compensation data to determine the final aeromagnetic measurement data.

[0012] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the physical constraint-based aeromagnetic data compensation method described in any one of the above.

[0013] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the physical constraint-based aeromagnetic data compensation method described above.

[0014] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the physical constraint-based aeromagnetic data compensation method described above.

[0015] In a sixth aspect, this application provides a chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run a program or instructions, and the processor implementing the steps of the physical constraint-based aeromagnetic data compensation method described above when executing the program or instructions.

[0016] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method, apparatus, device, and medium for aeromagnetic data compensation based on physical constraints. By constructing a multi-dimensional original feature set through multi-dimensional basic features and three-axis maneuvering attitude information, it can enrich the dimensions of network input features, strengthen the mapping relationship between flight attitude and magnetic interference, and improve the algorithm's cross-segment generalization ability. By introducing physical constraints of the first-order autocorrelation of residuals during the training process of the residual multilayer perceptron neural network, it can constrain the fitting residuals to approach pure white noise, avoiding the problems of accuracy degradation and overfitting of single magnetic field components during network training. By combining a two-level correction method of linear initial compensation and neural network nonlinear residual compensation, it can balance the efficient elimination of linear interference and the accurate fitting of complex nonlinear interference, effectively improving the overall compensation accuracy of aeromagnetic data. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a physically constrained aeromagnetic data compensation method provided in an embodiment of this application; Figure 2 A schematic diagram illustrating a compensation for flight attitude and path provided in an embodiment of this application; Figure 3 A schematic diagram of the functional modules of an aeromagnetic data compensation device based on physical constraints provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] In one exemplary embodiment, such as Figure 1 As shown, a physically constrained aeromagnetic data compensation method is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, it includes the following steps 101 to 105. Wherein: Step 101: Preprocess the collected flight parameters of the flight platform to obtain a multi-dimensional original feature set.

[0022] In this embodiment of the application, the multi-dimensional original feature set includes multi-dimensional basic features and three-axis maneuvering attitude information.

[0023] In this embodiment, before performing a compensation flight mission, the flight platform preferentially selects aircraft made of materials with low magnetic susceptibility, and performs omnidirectional demagnetization treatment on the fuselage before flight to eliminate the platform's inherent hard magnetic interference. During flight operations, multiple sensors mounted on the flight platform simultaneously collect data, specifically including: a scalar magnetometer collecting magnetic field strength data, a vector magnetometer collecting three-component magnetic field data, an inertial navigation system collecting attitude angle data of the flight platform, and the avionics system simultaneously recording flight status-related parameters. All the aforementioned sensors collect data at a uniform sampling frequency; for example, the sampling frequency can be set to 10Hz to ensure that the data from each channel is collected synchronously in the time dimension.

[0024] The collected raw data undergoes preprocessing, specifically including data synchronization and denoising. In the data synchronization stage, the scalar magnetometer data, vector magnetometer data, and attitude angle data output from the inertial navigation system are time-aligned according to a unified sampling timestamp, ensuring a one-to-one correspondence between the sensor data in the time dimension. This guarantees strict temporal matching of features during subsequent fusion calculations. In the denoising stage, the time-aligned sensor data is smoothed using a sliding window filtering method to remove sensor-specific electronic noise and external spike pulse interference, resulting in a smoothed total magnetic field strength time series and a three-component magnetic field strength time series.

[0025] After preprocessing, a multi-dimensional original feature set is extracted from the above data. This multi-dimensional original feature set contains two parts: The first part is the multi-dimensional basic features, which are extracted by performing coordinate transformation on the attitude angle data output by the inertial navigation system, converting the attitude angles in the body coordinate system into pitch, yaw, and roll angles in the geographic coordinate system. Based on the transformed pitch, yaw, and roll angles, the values ​​of each element of the direction cosine matrix are calculated. According to the TL compensation model, the interference magnetic field generated by the flight platform is decomposed into constant field interference, induced field interference, and eddy current field interference. Using the values ​​of each element of the direction cosine matrix, 18 interference basis functions are constructed. These 18 interference basis functions include 3 constant field basis functions, 6 induced field basis functions, and 9 eddy current field basis functions, which serve as the multi-dimensional basic features. The second part is the three-axis maneuvering attitude information, specifically the three real-time attitude data of pitch, yaw, and roll angles output by the inertial navigation system. These three data serve as supplementary features independent of the above 18 basis functions, reflecting the dynamic maneuvering state of the flight platform at each moment. Therefore, the multi-dimensional original feature set contains a total of 21 dimensions of feature information, of which the first 18 dimensions are the basis function features of the TL model and the last 3 dimensions are the maneuvering attitude features, which together constitute the basic input for subsequent compensation calculations.

[0026] As an optional implementation, step 101, which preprocesses the collected flight parameters of the flight platform to obtain a multi-dimensional original feature set, may include: The flight parameters of the flight platform are collected; wherein the flight parameters include at least magnetic measurement data and three-axis maneuvering attitude data; the magnetic measurement data is collected by a magnetometer pre-set on the flight platform, and the three-axis maneuvering attitude data is collected by an inertial navigation device pre-set on the flight platform; Multidimensional fundamental features are extracted from the magnetic measurement data; wherein, the multidimensional fundamental features include at least constant magnetic field features, induced magnetic field features, and eddy current magnetic field features; Three-axis maneuvering attitude information is extracted from the three-axis maneuvering attitude data; wherein, the three-axis maneuvering attitude information includes at least roll maneuvering attitude information, pitch maneuvering attitude information, and yaw maneuvering attitude information; The multidimensional basic features and the three-axis maneuvering attitude information are determined as a multidimensional original feature set.

[0027] This implementation method, by simultaneously collecting magnetic measurement data and three-axis maneuvering attitude data using a magnetometer and inertial navigation equipment, can completely obtain the two core data sources required for aeromagnetic compensation. By extracting three types of multi-dimensional basic features—constant, induced, and eddy current—from the magnetic measurement data, the linear magnetic interference of the flight platform can be fully characterized. By separately extracting the roll, pitch, and yaw three-axis maneuvering attitude information and fusing them to construct a multi-dimensional original feature set, attitude association priors can be provided for the neural network, enhancing the magnetic interference fitting ability and improving the aeromagnetic compensation correction effect.

[0028] In this embodiment of the application, the multidimensional basic features include 18 basic features, the three-axis maneuvering attitude information includes 3 supplementary features, and the multidimensional original feature set includes 21 features.

[0029] Step 102: Perform interference compensation calculation on the multi-dimensional original feature set to obtain the initial compensated feature set.

[0030] In this embodiment, based on the TL compensation model, the interfering magnetic field generated by the flight platform is decomposed into three parts: constant field interference, induced field interference, and eddy current field interference. The constant field interference is generated by the remanence and permanent magnetization of the ferromagnetic material of the flight platform itself, and its amplitude remains constant during flight and is independent of changes in flight attitude. The induced field interference is generated by the ferromagnetic structure of the flight platform being induced and magnetized by the geomagnetic field when the flight platform moves in the geomagnetic field, and its amplitude changes dynamically with changes in flight attitude. The eddy current field interference is generated by the eddy currents induced on the surface of the metal structure of the flight platform when it moves in the geomagnetic field, and its amplitude is related to the rate of change of flight attitude.

[0031] The TL compensation model can specifically be: In the formula, Indicates a constant magnetic field. Indicates the induced magnetic field. This represents an eddy current magnetic field.

[0032] in, Let be the direction cosines of the geomagnetic field vector along the three axes of the body coordinate system. Its rate of change, This represents the total intensity of the Earth's magnetic field. These are the 18 compensation coefficients to be determined.

[0033] Based on the physical mechanisms of the three types of interference mentioned above, 18 interference basis functions are constructed using a multi-dimensional original feature set. Three directional cosine values ​​of the directional cosine matrix are extracted from the multi-dimensional original feature set, and 18 basis functions are constructed based on these three directional cosine values, including 3 constant field basis functions, 6 induced field basis functions, and 9 eddy current field basis functions. These 18 basis functions are combined into an 18-dimensional feature vector, which serves as the input features for interference compensation calculation in the TL compensation model.

[0034] Ridge regression analysis was used to solve for the TL compensation coefficients. Ridge regression analysis prevents instability in coefficient calculation caused by matrix non-invertibility by adding an L2 regularization term to the least squares loss function, thus ensuring numerical stability of the solution process. Using the total magnetic field strength time series as the fitting target and 18 disturbance basis functions as input features, a ridge regression model was constructed and solved to obtain 18 TL compensation coefficients. These 18 compensation coefficients correspond to the amplitude coefficients of the disturbances in the constant field, induced field, and eddy current field, respectively.

[0035] The 18 TL compensation coefficients obtained from the solution are used to perform preliminary compensation calculations on the magnetic field strength data. The 18 interference basis functions are weighted and summed with their corresponding compensation coefficients to obtain the estimated interference magnetic field value. This estimated interference magnetic field value is then subtracted from the original magnetic field strength data to obtain the TL-compensated magnetic field strength data. Simultaneously, the difference between the TL-compensated magnetic field strength data and the original magnetic field strength data is calculated. This difference is used as the TL compensation residual sequence, which includes the nonlinear interference components that the TL linear model failed to fit.

[0036] The TL-compensated magnetic field strength data and the TL-compensated residual sequence are jointly defined as the initial compensation feature set. The TL-compensated magnetic field strength data serves as the initial compensation result, while the TL-compensated residual sequence serves as the fitting target for subsequent training of the residual multilayer perceptron neural network, used to train the neural network to learn and fit the implicit nonlinear disturbance patterns in the residuals.

[0037] Step 103: Train the pre-constructed residual multilayer perceptron neural network based on the multidimensional original feature set until the constraint fitting residual of the residual multilayer perceptron neural network matches the preset first-order autocorrelation physical constraint of the residual, and obtain the trained residual multilayer perceptron neural network.

[0038] In this embodiment, based on the physical principle that the aircraft interference field is dynamically driven by attitude changes, the residual of the aeromagnetic data after complete compensation should theoretically contain only white noise generated by the sensor itself, and should not retain any predictable temporal structure. Based on this physical principle, a first-order autocorrelation physical constraint on the residual is constructed. This physical constraint is used to penalize neural network fitting results that do not conform to the above physical principle.

[0039] Specifically, during the training process of the residual multilayer perceptron neural network, the difference between the residual fitted value output by the neural network and the TL-compensated residual sequence is used as the constraint fitting residual (i.e., the residual at the current time step), denoted as... The constraint fitting residual represents the error sequence that remains after neural network compensation. The first-order autocorrelation coefficient of this constraint fitting residual is calculated; this coefficient measures the degree of linear correlation between the constraint fitting residual at the current time step and the constraint fitting residual at the previous time step.

[0040] The specific calculation method for the first-order autocorrelation coefficient is as follows: First, calculate the mean of the constrained fitting residual sequence. The sum of the products of the residual deviations from the mean at adjacent time points is calculated using the standard deviation σ, and then divided by σ. 2 The first-order autocorrelation coefficient is obtained, and its absolute value is used as the physical constraint penalty term. When the constraint fitting residual is completely white noise, there is no linear correlation between the residual at the current time step and the residual at the previous time step, and the first-order autocorrelation coefficient should be zero. Therefore, the smaller the absolute value of the first-order autocorrelation coefficient, the closer the constraint fitting residual is to white noise, and the more sufficient the neural network compensation effect is; conversely, the larger the absolute value of the first-order autocorrelation coefficient, the more predictable temporal structure remains in the constraint fitting residual, and the less sufficient the neural network compensation effect is.

[0041] The absolute value of the first-order autocorrelation coefficient is used as a physical constraint penalty term and added to the loss function of the residual multilayer perceptron neural network, forming the total loss function together with the data fitting loss term. The residual multilayer perceptron neural network is iteratively trained using the backpropagation algorithm, aiming to minimize the total loss function including the physical constraint penalty term. This forces the neural network to focus not only on reducing the residual amplitude during training but also on whitening the residual temporal structure, thus making the compensation result output by the trained residual multilayer perceptron neural network physically more reasonable and achieving a more balanced compensation effect in cross-segment generalization scenarios.

[0042] As an optional implementation method, the residual multilayer perceptron neural network can be constructed as follows: Construct a lightweight, fully connected residual multilayer perceptron neural network; The residual multilayer perceptron neural network has 21 input layer channels; two fully connected network layers are sequentially set after the input layer, and ReLU activation functions are configured for each of the two fully connected network layers; and an output layer is set after the two fully connected network layers, with 1 output layer channel.

[0043] This implementation method, by building a lightweight, fully connected residual multilayer perceptron neural network, can reduce network computational overhead and improve data processing speed; by setting an input layer with 21 channels, it can fully receive all input data that fuses basic features and three-axis attitude information; by configuring ReLU activation functions in two fully connected network layers, it can effectively fit nonlinear magnetic interference in aeromagnetic data; and by setting a single-channel output layer, it can accurately output the residual magnetic interference compensation amount, ensuring the correction effect of aeromagnetic compensation.

[0044] In this embodiment of the application, the pitch angle, yaw angle and roll angle in the three-axis maneuver attitude information are normalized, and the three attitude angle data are mapped to a unified numerical range to obtain the normalized three attitude angle feature values. The normalized three attitude angle feature values ​​are used as independent feature dimensions and are concatenated with the 18 TL interference basis functions in the multidimensional basic features to form a 21-dimensional input feature vector. A residual multilayer perceptron neural network is constructed. The 21-dimensional input feature vector is used as the input feature of the residual multilayer perceptron neural network. The TL compensated residual sequence is used as the fitting target. Forward propagation calculation is performed on the residual multilayer perceptron neural network to obtain the residual fitting value output by the residual multilayer perceptron neural network. Construct a total loss function, which is the weighted sum of the data fitting loss term and the physical constraint penalty term. The data fitting loss term is the mean square error between the residual fitting value output by the residual multilayer perceptron neural network and the TL-compensated residual sequence. The physical constraint penalty term is the absolute value of the autocorrelation coefficient of the time series formed by the difference between the residual fitted value output by the residual multilayer perceptron neural network and the TL-compensated residual sequence. The calculation formula is: in, The residual at the current moment, The residual from the previous moment. Let σ be the mean of the residuals and σ be the standard deviation of the residuals. If the residuals are completely white noise, the autocorrelation coefficient should be zero. Therefore, the smaller this penalty term is, the closer the residuals are to white noise. x, y, and z represent the three orthogonal directions in the sensor measurement or aircraft coordinate system, i.e., the three-axis components.

[0045] 'i' represents the axial index or component index, used to distinguish residual sequences in different directions. Because signals in aircraft magnetic field compensation or similar sensor modeling are typically triaxial, the residual is not a single scalar, but rather residual components in three directions. Therefore, it can be written as: i∈{x, y, z}, that is: When i=x, it represents the residual in the x-axis direction; When i=y, it represents the residual in the y-axis direction; When i=z, it represents the residual in the z-axis direction.

[0046] The residual multilayer perceptron neural network is iteratively trained using the backpropagation algorithm to minimize the total loss function. In each iteration, the network weight parameters and bias parameters of the residual multilayer perceptron neural network are updated synchronously until the total loss function converges and the absolute value of the first-order autocorrelation coefficient is lower than a preset threshold, thus obtaining the trained residual multilayer perceptron neural network.

[0047] Step 104: Input the multi-dimensional original feature set into the trained residual multilayer perceptron neural network to obtain the aeromagnetic compensation data output by the residual multilayer perceptron neural network.

[0048] Step 105: Based on the aeromagnetic compensation data, compensate the initial compensation feature set to determine the final aeromagnetic measurement data.

[0049] In this embodiment, the aeromagnetic compensation data output by the trained residual multilayer perceptron neural network is used to perform compensation calculations with the total magnetic field strength data after TL compensation in the initial compensation feature set. Specifically, the aeromagnetic compensation data is subtracted from the total magnetic field strength data after TL compensation. This aeromagnetic compensation data is the residual fitting value output by the residual multilayer perceptron neural network after performing nonlinear fitting on the TL compensation residual sequence, reflecting the estimated value of the nonlinear interference components that the TL linear model failed to fit. By subtracting the residual nonlinear interference from the initial compensation result through this subtraction operation, the final compensated total magnetic field strength data is obtained.

[0050] The final compensated total magnetic field strength data was determined as the final aeromagnetic measurement data.

[0051] After determining the final aeromagnetic measurement data, an optional step may be included: evaluating the compensation effect. Calculate the standard deviation of the final compensated total magnetic field strength data. This standard deviation reflects the degree to which the final compensated total magnetic field strength data deviates from the true geomagnetic field value. The smaller the standard deviation, the higher the compensation accuracy.

[0052] The improvement ratio of aeromagnetic data is calculated. This improvement ratio is the ratio of the standard deviation of the total magnetic field strength data before compensation to the standard deviation of the total magnetic field strength data after final compensation. The larger the improvement ratio, the more significant the compensation effect.

[0053] Optionally, the residual sequence of the final compensated total magnetic field strength data can be subjected to power spectral density analysis to calculate the distribution characteristics of the energy spectrum of the residual sequence in different frequency bands. When the energy spectrum shows a uniform distribution characteristic in the entire frequency band, it is determined that the residual sequence is close to white noise, thereby verifying the suppression effect of the physical constraint term on the residual temporal structure.

[0054] As an optional implementation, step 105, which compensates the initial compensation feature set based on the aeromagnetic compensation data to determine the final aeromagnetic measurement data, may include: Extract the compensated magnetic field strength data from the initial compensation feature set; Subtracting the aeromagnetic compensation data from the compensated magnetic field strength data yields the final aeromagnetic measurement data.

[0055] In this implementation method, by extracting the compensated magnetic field strength data from the initial compensation feature set, the basic data to be corrected after linear correction can be locked. By subtracting the aeromagnetic compensation data output by the neural network from the magnetic field strength data, the nonlinear residual magnetic interference that was not eliminated by the linear model can be canceled in one go, and the secondary correction can be completed simply and efficiently. This effectively eliminates the complex magnetic noise of the flight platform and improves the purity and compensation accuracy of the final aeromagnetic measurement data.

[0056] As an optional implementation, after step 105, the following steps may also be performed: Perform standard deviation calculation on the original multi-dimensional feature set to obtain the standard deviation before compensation; The standard deviation of the aeromagnetic measurement data is calculated to obtain the compensated standard deviation; The compensation-improvement ratio is obtained by calculating the ratio of the standard deviation before compensation to the standard deviation after compensation. Based on the standard deviation before compensation, the standard deviation after compensation, and the compensation improvement ratio, an aeromagnetic data compensation evaluation result is generated.

[0057] This implementation method allows for a direct distinction between the data error levels before and after compensation by calculating the standard deviations of the multi-dimensional original feature set and the final aeromagnetic measurement data. The compensation improvement ratio is obtained by comparing the two sets of standard deviations, which quantifies the interference suppression magnitude. By generating compensation evaluation results by comprehensively considering the three indicators, the aeromagnetic compensation effect can be objectively and quantitatively evaluated, facilitating rapid verification of the algorithm's correction capabilities.

[0058] In this embodiment of the application, the formula for calculating the standard deviation (STD) can be: in, n For the sample size, For the true value, This represents the data mean.

[0059] In this embodiment of the application, the formula for calculating the compensation improvement ratio (IR) can be: Please refer to the following: Figure 2 , Figure 2 A schematic diagram illustrating a compensation for flight attitude and path provided in an embodiment of this application; Figure 2 As shown, during aeromagnetic compensation flight operations, the aircraft maneuvers along a rectangular compensation flight path. Different segments of the path correspond to roll, pitch, and yaw maneuvers, respectively. During flight, the roll angle is controlled within ±10°, the pitch angle within ±5°, and the yaw angle within ±5°, sequentially completing these three-dimensional attitude maneuvers. At each maneuver phase, magnetic measurement data output from the magnetometer and attitude data output from the inertial navigation system are simultaneously collected, comprehensively covering various typical flight attitudes of the aircraft. This allows for the acquisition of magnetic interference samples generated by the flight platform under different attitude conditions. The collected sample data can be used to generate a multi-dimensional original feature set, providing sufficient attitude coverage samples for subsequent neural network training, which is beneficial for improving the model's fitting effect on magnetic interference under different flight attitudes.

[0060] For example, the dataset used is an open-source dataset. The dataset we used, numbered 1002, contains data from 5 scalar magnetometers, 3 vector magnetometers, inertial navigation systems, avionics systems, electrical systems, and radar. The sampling frequency is 10Hz, and the flight track is divided into three segments: DF1, DF2, and DF3, each with approximately 6000 sampling points. Segments DF1 and DF2 were used for training, and segment DF3 was used for testing. The cross-segment test results for the five methods used are shown in Table 1 below: Table 1. Segment test results of the five methods

[0061] Summary of cross-segment testing: 1. Comparison of methods under 18-dimensional features: When using only 18-dimensional TL basis functions (without INS pose information): The traditional TL method achieves a total field STD of 4.90 nT after correction, with an improvement ratio of 9.05.

[0062] TL exhibits extremely high suppression capability on the X component (STD_after=2.64nT, IR_STD=15.63), but is relatively weak on the Y component (IR=9.73).

[0063] The residual MLP (unconstrained) reduced the total field STD to 4.40 nT, improving the ratio to 10.08, and the Y component also showed significant improvement.

[0064] However, its X component is significantly worse than TL (STD=5.06nT vs 2.64nT), indicating that the unconstrained network sacrifices the accuracy of the X component in the learning process in exchange for minimizing the overall loss.

[0065] After introducing the autocorrelation constraint (λ=0.05), the network recovers to 4.58 nT on the X component, while the Y and Z components remain stable. Ultimately, the total field STD further decreases to 4.10 nT, achieving an improvement ratio of 10.81. This indicates that the autocorrelation constraint, as a regularization mechanism, can force the network to produce whiter residuals, thereby achieving a more balanced compensation effect across multiple components and avoiding excessive sacrifice in certain components by unconstrained networks.

[0066] 2. Comparison of methods under 21-dimensional features (including INS pose angle): When the input features are expanded to 21 dimensions (18-dimensional TL basis functions + 3-dimensional INS attitude angles): The TL method performs exactly the same as the 18-dimensional method because the TL model only uses the first 18 TL basis functions.

[0067] The total field STD of the residual MLP (unconstrained) is 4.18 nT, an improvement of 10.62, which is better than the 4.40 nT of the 18-dimensional model. This indicates that the INS attitude angle provides the network with additional attitude information decoupled from the geomagnetic field, which helps the network learn the disturbance field mapping more accurately.

[0068] After adding the autocorrelation constraint (λ=0.05), the total field STD further decreased to 4.03 nT, achieving an improvement ratio of 11.02, which is the best result among all methods. Compared with the effect of the autocorrelation constraint in 18 dimensions (4.10 nT), the further improvement under 21-dimensional features, although smaller, shows a good compatibility between INS information and physical constraints: INS provides a richer feature space, while the autocorrelation constraint helps the network to more effectively eliminate residual temporal structures in this space.

[0069] 3. The function of INS features: Under 21-dimensional features, regardless of whether autocorrelation constraints are applied, the total field STD of the residual MLP is lower than the corresponding 18-dimensional results (4.18 vs 4.40, 4.03 vs 4.10). This verifies the positive contribution of the INS attitude angle to cross-segment generalization. The INS attitude angle depends only on the aircraft's own motion and is independent of the geomagnetic environment, thus helping the network learn a more transferable "attitude-disturbance" mapping, which to some extent alleviates the input distribution drift problem caused by environmental changes in the direction cosine.

[0070] 4. The value of autocorrelation constraints: Under both feature dimensions, the autocorrelation constraint resulted in a reduction in the total field STD (18-dimensional: 4.40 → 4.10; 21-dimensional: 4.18 → 4.03), and the improvement persisted even in the 21-dimensional case. This indicates that the autocorrelation constraint is not only effective when features are insufficient, but can still provide additional regularization benefits after feature enhancement. By penalizing the first-order autocorrelation of the residuals, the autocorrelation constraint suppresses the residual temporal structured error in the residuals, making the compensated residuals closer to white noise, thereby improving the accuracy of dynamic compensation.

[0071] 5. Comparison between intra-segment self-consistency and cross-segment generalization: In intra-segment self-consistency tests, the neural network comprehensively outperformed the training-test (TL) network (IR of 23.08 for a 21-dimensional MLP on DF3 vs. 15.73 for TL). However, in cross-segment tests, the total field IR of the same method was only 11.02 (for a 21-dimensional autocorrelation MLP), a significant narrowing of the improvement relative to TL's 9.05. This contrast clearly reveals the core bottleneck of cross-segment generalization: the distribution drift between the training and testing segments. The success of intra-segment self-consistency demonstrates the network's nonlinear fitting ability, while the performance degradation in cross-segment tests points to mapping failure caused by environmental changes. This points the way for future work—more effective domain adaptation or environment-invariant feature learning strategies are needed to bridge this gap.

[0072] Optimal result: 21-dimensional MLP with autocorrelation constraint → Total STD = 4.03nT, IR = 11.02, an improvement of 21.8% from IR = 9.05 for TL to IR = 11.02 for 21D with autocorrelation.

[0073] 6. Test conclusions of the example: ① The residual MLP outperforms the TL linear model in all aspects of intra-segment self-consistency, verifying the effectiveness of deep learning in aeromagnetic compensation; ② Feature enhancement (21-dimensional INS pose angle) significantly improves the compensation effect in cross-segment generalization; ③ Autocorrelation physical constraints, as a regularization method, effectively improve component balance in cross-segment scenarios; ④ The intra-segment IR is much higher than the cross-segment IR → The main challenge is the distribution drift between the training segment and the test segment; ⑤ Optimal solution: 21-dimensional MLP + autocorrelation constraint → Total STD = 4.03nT, IR = 11.02.

[0074] Implementing steps 101 to 105 above, a multi-dimensional original feature set is constructed by combining multi-dimensional basic features and three-axis maneuvering attitude information. This enriches the dimensions of the network input features, strengthens the mapping relationship between flight attitude and magnetic interference, and improves the algorithm's cross-segment generalization ability. By introducing the physical constraint of the first-order autocorrelation of the residuals during the training process of the residual multilayer perceptron neural network, the fitting residuals can be constrained to approach pure white noise, avoiding the problems of accuracy degradation and overfitting of a single magnetic field component during network training. By combining a two-level correction method of linear initial compensation and neural network nonlinear residual compensation, both efficient elimination of linear interference and accurate fitting of complex nonlinear interference can be achieved, effectively improving the overall compensation accuracy of aeromagnetic data.

[0075] Based on the same inventive concept, this application also provides a physical constraint-based aeromagnetic data compensation device for implementing the aforementioned physical constraint-based aeromagnetic data compensation method. The solution provided by this device is similar to the implementation described in the above method. Therefore, the specific limitations of one or more physical constraint-based aeromagnetic data compensation device embodiments provided below can be found in the limitations of the physical constraint-based aeromagnetic data compensation method described above, and will not be repeated here.

[0076] In one exemplary embodiment, such as Figure 3 As shown, a physically constrained aeromagnetic data compensation device is provided, comprising: The preprocessing unit 301 is used to preprocess the flight parameters of the acquired flight platform to obtain a multi-dimensional original feature set; wherein, the multi-dimensional original feature set includes multi-dimensional basic features and three-axis maneuvering attitude information; The calculation unit 302 is used to perform interference compensation calculation on the multi-dimensional original feature set to obtain the initial compensation feature set; Training unit 303 is used to train a pre-constructed residual multilayer perceptron neural network based on the multidimensional original feature set until the constraint fitting residual of the residual multilayer perceptron neural network matches the preset residual first-order autocorrelation physical constraint, thereby obtaining the trained residual multilayer perceptron neural network. Input unit 304 is used to input the multi-dimensional original feature set into the trained residual multilayer perceptron neural network to obtain the aeromagnetic compensation data output by the residual multilayer perceptron neural network. The compensation unit 305 is used to compensate the initial compensation feature set based on the aeromagnetic compensation data to determine the final aeromagnetic measurement data.

[0077] By implementing the above methods, a multi-dimensional original feature set can be constructed using multi-dimensional basic features and three-axis maneuvering attitude information. This enriches the dimensions of the network input features, strengthens the mapping relationship between flight attitude and magnetic interference, and improves the algorithm's generalization ability across flight segments. By introducing the physical constraint of the first-order autocorrelation of the residuals during the training process of the residual multilayer perceptron neural network, the fitting residuals can be constrained to approach pure white noise, avoiding the problems of accuracy degradation and overfitting of single magnetic field components during network training. By combining a two-stage correction method of linear initial compensation and neural network nonlinear residual compensation, both efficient elimination of linear interference and accurate fitting of complex nonlinear interference can be achieved, effectively improving the overall compensation accuracy of aeromagnetic data.

[0078] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores aeromagnetic data compensation data based on physical constraints. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a physical constraint-based aeromagnetic data compensation method.

[0079] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0080] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0081] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0082] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0083] In one exemplary embodiment, a chip is provided, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the steps in the above method embodiments and achieve the same technical effect, and will not be described again here to avoid repetition.

[0084] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0085] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0086] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0087] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0088] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0089] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for compensating aeromagnetic data based on physical constraints, characterized in that, The aeromagnetic data compensation method based on physical constraints includes: The flight parameters of the acquired flight platform are preprocessed to obtain a multi-dimensional original feature set; wherein, the multi-dimensional original feature set includes multi-dimensional basic features and three-axis maneuvering attitude information; Interference compensation calculation is performed on the original multi-dimensional feature set to obtain the initial compensated feature set; The pre-constructed residual multilayer perceptron neural network is trained based on the multidimensional original feature set until the constraint fitting residual of the residual multilayer perceptron neural network matches the preset first-order autocorrelation physical constraint of the residual, thus obtaining the trained residual multilayer perceptron neural network. The multi-dimensional original feature set is input into the trained residual multilayer perceptron neural network to obtain the aeromagnetic compensation data output by the residual multilayer perceptron neural network. The initial compensation feature set is compensated based on the aeromagnetic compensation data to determine the final aeromagnetic measurement data.

2. The aeromagnetic data compensation method based on physical constraints according to claim 1, characterized in that, The preprocessing of the collected flight parameters of the flight platform yields a multi-dimensional original feature set, specifically including: The flight parameters of the flight platform are collected; wherein the flight parameters include at least magnetic measurement data and three-axis maneuvering attitude data; the magnetic measurement data is collected by a magnetometer pre-set on the flight platform, and the three-axis maneuvering attitude data is collected by an inertial navigation device pre-set on the flight platform; Multidimensional fundamental features are extracted from the magnetic measurement data; wherein, the multidimensional fundamental features include at least constant magnetic field features, induced magnetic field features, and eddy current magnetic field features; Three-axis maneuvering attitude information is extracted from the three-axis maneuvering attitude data; wherein, the three-axis maneuvering attitude information includes at least roll maneuvering attitude information, pitch maneuvering attitude information, and yaw maneuvering attitude information; The multidimensional basic features and the three-axis maneuvering attitude information are determined as a multidimensional original feature set.

3. The aeromagnetic data compensation method based on physical constraints according to claim 2, characterized in that, The multidimensional basic features include 18 basic features, the three-axis maneuvering attitude information includes 3 supplementary features, and the multidimensional original feature set includes 21 features.

4. The aeromagnetic data compensation method based on physical constraints according to claim 3, characterized in that, The specific method for constructing the residual multilayer perceptron neural network is as follows: Construct a lightweight, fully connected residual multilayer perceptron neural network; The residual multilayer perceptron neural network has 21 input layer channels; two fully connected network layers are sequentially set after the input layer, and ReLU activation functions are configured for each of the two fully connected network layers; and an output layer is set after the two fully connected network layers, with 1 output layer channel.

5. The aeromagnetic data compensation method based on physical constraints according to claim 4, characterized in that, The step of compensating the initial compensation feature set based on the aeromagnetic compensation data to determine the final aeromagnetic measurement data specifically includes: Extract the compensated magnetic field strength data from the initial compensation feature set; Subtracting the aeromagnetic compensation data from the compensated magnetic field strength data yields the final aeromagnetic measurement data.

6. The aeromagnetic data compensation method based on physical constraints according to claim 5, characterized in that, After determining the final aeromagnetic measurement data, the method further includes: The standard deviation is calculated on the original multi-dimensional feature set to obtain the standard deviation before compensation; The standard deviation of the aeromagnetic measurement data is calculated to obtain the compensated standard deviation; The compensation-improvement ratio is obtained by calculating the ratio of the standard deviation before compensation to the standard deviation after compensation. Based on the standard deviation before compensation, the standard deviation after compensation, and the compensation improvement ratio, an aeromagnetic data compensation evaluation result is generated.

7. A physically constrained aeromagnetic data compensation device, characterized in that, The aeromagnetic data compensation device based on physical constraints includes: The preprocessing unit is used to preprocess the flight parameters of the acquired flight platform to obtain a multi-dimensional original feature set; wherein, the multi-dimensional original feature set includes multi-dimensional basic features and three-axis maneuvering attitude information; The calculation unit is used to perform interference compensation calculation on the multi-dimensional original feature set to obtain the initial compensation feature set; The training unit is used to train the pre-constructed residual multilayer perceptron neural network based on the multidimensional original feature set until the constraint fitting residual of the residual multilayer perceptron neural network matches the preset residual first-order autocorrelation physical constraint, thereby obtaining the trained residual multilayer perceptron neural network. The input unit is used to input the multi-dimensional original feature set into the trained residual multilayer perceptron neural network to obtain the aeromagnetic compensation data output by the residual multilayer perceptron neural network. The compensation unit is used to compensate the initial compensation feature set based on the aeromagnetic compensation data to determine the final aeromagnetic measurement data.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the physically constrained aeromagnetic data compensation method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the aeromagnetic data compensation method based on physical constraints as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the aeromagnetic data compensation method based on physical constraints as described in any one of claims 1-6.

Images