A deep-sea magnetometer based on three-component MEMS magnetic sensor

Abstract

The application discloses a deep-sea magnetometer based on a three-component MEMS magnetic sensor and relates to the deep-sea magnetometer, which comprises the following steps: S01, synchronously acquiring original three-axis magnetic signals of the three-component MEMS magnetic sensor, attitude and motion data of an inertial measurement unit and environmental parameters of a temperature and pressure sensor, and performing time alignment and initial filtering on the acquired data to obtain a processed data set; S02, based on a long short-term memory network model obtained through pre-training, real-time prediction and compensation of sensor scale factor errors and zero bias drift caused by deep-sea time-varying environment according to the time-series processed data set, and output of the three-axis magnetic data after preliminary correction. Through the introduction of the dynamic error correction based on the LSTM, the state augmentation adaptive filtering fusing the inertial information and the online compensation method combining the geomagnetic model and the sliding window regularization estimation, the data precision, the dynamic anti-interference ability and the long-term working stability of the three-component MEMS magnetic sensor are improved.

Description

Technical Field

[0001] This invention relates to deep-sea magnetometers, and more particularly to a deep-sea magnetometer based on a three-component MEMS magnetic sensor. Background Technology

[0002] Deep-sea magnetic measurement is a core technology in marine geophysical exploration, seabed resource surveys, underwater target detection, and geomagnetic navigation. Traditional marine magnetometers, such as proton precession magnetometers and optically pumped magnetometers, can provide high-precision total field scalar information, but their systems are usually bulky, consume a lot of power, and cannot directly obtain geomagnetic field vector information that is crucial for geological structure analysis and magnetic target identification.

[0003] The deep-sea working environment is characterized by high pressure, large temperature gradients, and complex background noise. MEMS sensors themselves lag behind high-end fluxgate or optical pump sensors in terms of sensitivity, noise level, and stability. Their raw output signal contains a mixture of various components, including the real geomagnetic field, inherent sensor errors, environmental interference, and interference from the carrier platform. Inherent sensor errors mainly include triaxial sensitivity inconsistency (scale factor error), inter-axis non-orthogonality, zero-point deviation (zero bias), and installation deviation. These parameters are not constant and drift with changes in ambient temperature and hydrostatic pressure. Traditional calibration and correction methods are mostly based on the assumption of ellipsoidal fitting under static or slow rotation in a uniform magnetic field. These methods are effective under controlled laboratory conditions but cannot adapt to the time-varying parameter problems caused by real-time temperature and pressure changes during dynamic deep-sea navigation, resulting in residual correction errors and affecting the accuracy of long-term measurements.

[0004] Besides sensor inherent errors, interference introduced by the vehicle's motion is another major noise source in deep-sea magnetic surveys. The complex motions, such as vibrations and swaying, generated by underwater vehicles during navigation and operation couple to the magnetic sensors through various physical mechanisms, forming motion-related induced noise. The spectrum of this noise often overlaps with the spectrum of the geomagnetic anomaly signal to be detected, causing conventional frequency domain filtering methods (such as low-pass filtering) to severely impair the useful magnetic anomaly signal while suppressing the noise. Although inertial measurement units (IMUs) can provide motion reference information, effectively fusing IMU data with magnetic signal models and designing adaptive, high-fidelity filtering algorithms to accurately separate the geomagnetic vector under non-stationary motion conditions remains a pressing problem. Summary of the Invention

[0005] To address the aforementioned technical problems, the present invention provides a deep-sea magnetometer based on a three-component MEMS magnetic sensor, comprising the following steps:

[0006] S01. Simultaneously acquire the raw triaxial magnetic signals of the three-component MEMS magnetic sensor, the attitude and motion data of the inertial measurement unit, and the environmental parameters of the temperature and pressure sensor, and perform time alignment and initial filtering on the acquired data to obtain the processed dataset.

[0007] S02. Based on the pre-trained long short-term memory network model, the sensor scaling factor error and zero bias drift caused by the time-varying environment in the deep sea are predicted and compensated in real time according to the processed dataset, and the preliminary corrected triaxial magnetic data are output.

[0008] S03. Using the pre-corrected triaxial magnetic data as the observation value, and combining it with the inertial measurement unit data, construct a state-amplified adaptive Kalman filter, and use the true value of the geomagnetic field and the magnetic noise caused by the carrier motion as the state for joint estimation, separate and suppress motion interference in real time, and output the purified geomagnetic vector estimate.

[0009] S04. Using the purified geomagnetic vector estimation sequence within the sliding time window, the corresponding carrier attitude information, and the background field data provided by the international geomagnetic reference field model, the carrier hard and soft magnetic interference parameters are solved online through a parameter estimation algorithm with regularization constraints. The geomagnetic vector at the current moment is compensated in real time, and finally high-precision balanced geomagnetic data or magnetic anomaly data are output.

[0010] Preferably, step S02, which involves predicting and compensating for sensor scaling factor errors and zero-bias drift caused by the time-varying environment of the deep sea based on the processed dataset, and outputting preliminarily corrected triaxial magnetic data, includes:

[0011] S21. Construct a dual-branch input LSTM network model. The first input branch is used to receive continuous time series temperature and pressure environmental parameters, and the second input branch is used to receive triaxial raw magnetic signals in the same time series.

[0012] S22. The output of the LSTM network model is the sensor dynamic error parameter vector at the current moment. The sensor dynamic error parameter vector includes at least a three-axis scaling factor correction value and a three-axis zero bias correction value.

[0013] S23. Using the aforementioned dynamic error parameter vector, perform dynamic correction calculations:

[0014] ;

[0015] in, Let be the original triaxial magnetic signal vector at time t. This is the diagonal matrix of the triaxial scaling factor error predicted by the LSTM model at time t. The three-axis zero-bias vector predicted by the LSTM model at time t;

[0016] The original magnetic signal is inversely corrected in real time to obtain the dynamic error compensation value.

[0017] Preferably, the construction of the adaptive Kalman filter in step S3 includes the following steps:

[0018] S31. Define the state vector as follows: ,in: The true three-component state of the geomagnetic field in the carrier coordinate system. The three components of the magnetic noise caused by the carrier motion;

[0019] S32. Establish the dynamic model of the current inertial measurement unit as an empirical model related to the angular velocity and linear acceleration measured by the current inertial measurement unit, and add process noise;

[0020] S33. The output of the pre-corrected triaxial magnetic data is used as the observation, and its relationship with the state vector is the sum of the two.

[0021] S34. During the filtering iteration process, the Sage-Husa adaptive estimation algorithm is used to update the observed noise covariance matrix in real time to match the noise statistical characteristics of the carrier under different motion states.

[0022] Preferably, the empirical model related to the angular velocity and linear acceleration measured by the current inertial measurement unit in step S32 is obtained by pre-calibrating and fitting the inertial data and magnetic noise data under typical motion modes of the carrier, and the formula is as follows: ,in, Let be the angular velocity and linear acceleration of the carrier at time k, respectively. This is process noise.

[0023] Preferably, step S04, which involves using a parameter estimation algorithm with regularization constraints to solve for the hard and soft magnetic interference parameters of the carrier online and to compensate for the geomagnetic vector at the current moment in real time, ultimately outputting high-precision balanced geomagnetic data or magnetic anomaly data, includes:

[0024] S41. Set a sliding time window of length N, obtain the attitude matrix at each moment within the window, and transform the purified geomagnetic vector estimate at the corresponding moment from the carrier coordinate system to the navigation coordinate system.

[0025] S42. Based on the carrier's location and time, query the IGRF model to obtain the background geomagnetic field vector in the navigation coordinate system;

[0026] S43. Within the sliding window, establish a system of linear equations with the carrier hard magnetic interference vector, soft magnetic interference matrix and geomagnetic anomaly vector as unknown parameters, wherein the equations are composed of the transformed geomagnetic vector estimate, IGRF background field value and attitude matrix.

[0027] S44. Solve the linear equation system using the least squares method with Tikhonov regularization, wherein the regularization term is designed based on the symmetry and small trace characteristics of the soft magnetic interference matrix.

[0028] S45. Using the hard magnetic interference vector and soft magnetic interference matrix obtained from the solution, the estimated value of the geomagnetic vector in the carrier coordinate system at the current or next time moment is compensated. The calculation formula is as follows: ,in: The purified geomagnetic vector estimate output in step S03. For the estimated hard magnetic interference vector, For the estimated soft magnetic interference matrix, This is the final compensated geomagnetic field vector.

[0029] Preferably, step S44, which involves solving the linear equation system using the least squares method with Tikhonov regularization, includes:

[0030] S441. Construct the system of linear equations into a least squares problem: , where p is a parameter vector containing the elements of the hard magnetic interference vector and the soft magnetic interference matrix to be estimated, and A and b are constructed from the observation equations within the sliding window;

[0031] S442. Introduce the Tikhonov regularization term to construct a regularized least squares problem: , where Γ is the regularization matrix and λ is the regularization parameter;

[0032] S443. The regularization matrix is ​​designed based on the physical prior knowledge of the soft magnetic interference matrix and is used to constrain the symmetry of K and the magnitude of its elements in order to improve the numerical stability of the solution.

[0033] Preferably, the high-precision balanced geomagnetic data or magnetic anomaly data output in step S04 is adjusted, including the following steps:

[0034] S05. Analyze the angular velocity and linear acceleration data output by the inertial measurement unit in real time, and calculate the motion intensity index of the carrier;

[0035] S06. Based on the different preset intervals in which the exercise intensity index is located, dynamically adjust the preset amplitude of the process noise covariance matrix Q of the adaptive Kalman filter in step S03, and the length N of the sliding time window in step S04.

[0036] Specifically, when the motion intensity index is higher than the first threshold, the variance preset value of the corresponding motion noise state component in the Q matrix is ​​increased, and the window length N is shortened.

[0037] When the exercise intensity index is lower than the second threshold, the variance preset value is reduced and the window length N is increased.

[0038] The present invention has at least the following beneficial effects:

[0039] 1. By introducing deep learning models into the dynamic calibration of deep-sea MEMS sensors, and through the joint analysis of time-series environmental parameters and raw signals using long short-term memory networks, real-time and accurate prediction and compensation for drift of key sensor error parameters are achieved. This effectively overcomes the problem of traditional static calibration methods failing in the time-varying environment of the deep sea, significantly improving the accuracy and long-term stability of raw measurement data from the data source.

[0040] 2. By constructing a state-amplified adaptive Kalman filter, inertial measurement unit data is used as a key reference, and motion-induced magnetic noise, which is difficult to model, is used as a state for joint estimation and separation. This enables the algorithm to intelligently distinguish between geomagnetic signals and motion interference, especially in the challenging situation where their spectra overlap. It can still effectively suppress noise caused by carrier vibration and swaying, greatly enhancing the robustness and reliability of the system under dynamic navigation conditions.

[0041] 3. Utilizing historical data within a sliding time window, known geomagnetic field model background, and carrier attitude information, the system employs physical-constrained regularized estimation to calculate and compensate for hard and soft magnetic disturbances in real time. This not only reduces the stringent requirements of the operational process and improves the practicality and long-term availability of the deep-sea magnetometer, but also, through continuous adaptive updates, tracks the slow changes in disturbance parameters, ensuring the continuous output of high-precision magnetic anomaly or geomagnetic field vector data during long-endurance missions. This enhances the overall practical value and application potential of the system. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 The flowchart of a deep-sea magnetometer based on a three-component MEMS magnetic sensor is provided for Embodiment 1 of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0046] Example 1

[0047] This embodiment provides a deep-sea magnetometer based on a three-component MEMS magnetic sensor. The method includes the following steps: Figure 1 As shown:

[0048] S01. Simultaneously acquire the raw triaxial magnetic signals from the three-component MEMS magnetic sensor, the attitude and motion data from the inertial measurement unit, and the environmental parameters from the temperature and pressure sensors. Perform time alignment and initial filtering on the acquired data to obtain the processed dataset.

[0049] Specifically, by sending synchronous trigger pulses to the three-component MEMS magnetic sensor, inertial measurement unit (IMU), and temperature and pressure sensor, the system forces each sensor to perform data conversion and readout at the same physical moment, thereby ensuring data time consistency from the source and eliminating timing errors caused by internal sensor clock drift or independent sampling. After acquiring the raw data packets recorded by each sensor using its own clock, the system adds a standard timestamp generated by a high-precision master clock to each successfully received data packet, since there may be slight delays in data transmission between sensors or different communication protocols (such as SPI, I2C). Through interpolation or resampling algorithms, the system unifies the data from each channel with different inherent sampling rates (e.g., 200Hz for the IMU and 100Hz for the magnetic sensor) onto the same time series, forming a strictly synchronized multidimensional data frame. Finally, in the preprocessing stage, the system performs necessary preliminary cleaning on the raw data of each channel. For example, it performs moving average filtering on magnetic signals and IMU data to suppress high-frequency circuit noise, performs rationality checks on temperature and pressure data to remove outliers caused by communication errors, and converts all data to standard physical units (such as Tesla, degrees / second, degrees Celsius) to provide a clean, synchronous and standardized input data stream with uniform format for subsequent core algorithm modules.

[0050] S02. Based on the pre-trained long short-term memory network model, and according to the processed dataset, predict and compensate for sensor scaling factor errors and zero-bias drift caused by the time-varying environment in the deep sea in real time, and output the pre-corrected triaxial magnetic data, including the following steps:

[0051] S21. Construct a dual-branch input LSTM network model. The first input branch is used to receive continuous time series temperature and pressure environmental parameters, and the second input branch is used to receive triaxial raw magnetic signals in the same time series.

[0052] The output of the S22 LSTM network model is the sensor dynamic error parameter vector at the current moment. The sensor dynamic error parameter vector includes at least the three-axis scaling factor correction value and the three-axis zero bias correction value.

[0053] S23. Perform dynamic correction calculations using the dynamic error parameter vector:

[0054] ;

[0055] in, Let be the original triaxial magnetic signal vector at time t. This is the diagonal matrix of the triaxial scaling factor error predicted by the LSTM model at time t. The three-axis zero-bias vector predicted by the LSTM model at time t;

[0056] The original magnetic signal is inversely corrected in real time to obtain the dynamic error compensation value.

[0057] Specifically, during the offline training phase, a dual-branch Long Short-Term Memory (LSTM) network model architecture needs to be constructed. One branch of this architecture is dedicated to receiving the time-sequential sequence of temperature and pressure environmental parameters, while the other branch simultaneously receives the raw triaxial magnetic signal sequence within the same time window. The core of this design is to enable the LSTM network to simultaneously learn the deep coupling relationship and temporal dynamics between changes in the environmental physical field and the raw sensor output. The network model is trained in a controlled experimental environment by placing the MEMS sensor and a high-precision reference magnetic sensor (such as a fluxgate) on a non-magnetic turntable with precisely adjustable temperature and pressure. Full-state data covering the expected operating range (sensor output and reference true value under different combinations of temperature, pressure, and attitude) are collected. Supervised training is performed with the goal of minimizing the error between the network's predicted correction value and the reference true value, thereby enabling the network to intrinsically grasp the complex mapping law of error parameters (scaling factor and zero bias) as the environment and signal history change. During the online application phase, when the system is operating in real time in the deep sea, it simultaneously collects the current temperature and pressure data and raw magnetic signals, and continuously inputs them into the trained LSTM model in a sliding time window. Based on the learned patterns, the model dynamically infers the most likely sensor scaling factor error matrix and zero bias vector at the current moment.

[0058] S03. Using the pre-corrected triaxial magnetic data as the observation value, combined with the inertial measurement unit data, an adaptive Kalman filter with state amplification is constructed. The true value of the geomagnetic field and the magnetic noise caused by the carrier motion are used together as the state for joint estimation. Motion interference is separated and suppressed in real time, and the purified geomagnetic vector estimate is output.

[0059] Furthermore, the construction of the above-mentioned adaptive Kalman filter includes the following steps:

[0060] S31. Define the state vector as follows: ,in: The true three-component state of the geomagnetic field in the carrier coordinate system. The three components of the magnetic noise caused by the carrier motion;

[0061] S32. Establish the dynamic model of the current inertial measurement unit as an empirical model related to the angular velocity and linear acceleration measured by the current inertial measurement unit, and add process noise;

[0062] S33. The output triaxial magnetic data after preliminary correction is used as the observation, and its relationship with the state vector is the sum of the two.

[0063] S34. During the filtering iteration process, the Sage-Husa adaptive estimation algorithm is used to update the observed noise covariance matrix in real time to match the noise statistical characteristics of the carrier under different motion states.

[0064] Secondly, the empirical models related to the angular velocity and linear acceleration measured by the current inertial measurement unit in step S32 above are obtained by pre-calibrating and fitting the inertial data and magnetic noise data under typical motion modes of the carrier. The formula is as follows: ,in, Let be the angular velocity and linear acceleration of the carrier at time k, respectively. This is process noise.

[0065] Specifically, by having the carrier perform a series of actions involving typical angular and linear motions, high-precision reference magnetometer data, MEMS magnetometer data, and IMU angular velocity and linear acceleration data are recorded simultaneously. Then, the correlation between motion data and magnetic disturbance data is analyzed, and an empirical mapping model from inertial data to magnetic noise is established using system identification or curve fitting methods (such as polynomial fitting or neural networks). This model serves as a dynamic sub-model of the motion noise state v in the filter state equation.

[0066] During the real-time processing phase, the filter will output the true value of the geomagnetic field. and motion noise Together, they form the state vector. At each moment, the filter uses ω and a, measured in real time by the IMU, and applies the aforementioned empirical model. The algorithm predicts the evolution trend of motion noise and incorporates it into the state transition equation. Simultaneously, the magnetic data corrected in step S02 is used as the observation value, which is modeled as the superposition of the true geomagnetic state and the motion noise state. The most crucial adaptive mechanism is reflected in the continuous monitoring of the observation residual (i.e., the difference between the predicted and actual observation values) by the filter, and the online dynamic estimation and correction of the observation noise covariance matrix R using the Sage-Husa algorithm. This means that when the carrier moves violently and the model uncertainty increases, the algorithm automatically "trusts" the prediction of the state model; when the motion is gentle, it "trusts" the sensor observation values ​​more. Through this synergistic effect of state amplification and adaptive adjustment, the algorithm can, under the challenge of mixed motion noise and geomagnetic signal spectrum, separate high-precision geomagnetic field vector estimates in real time and dynamically, thus effectively suppressing co-band motion interference that is difficult to handle by traditional frequency domain filtering methods.

[0067] S04. Using the purified geomagnetic vector estimation sequence within the sliding time window, the corresponding carrier attitude information, and the background field data provided by the international geomagnetic reference field model, the hard and soft magnetic interference parameters of the carrier are solved online through a parameter estimation algorithm with regularization constraints. The geomagnetic vector at the current moment is compensated in real time, and finally high-precision balanced geomagnetic data or magnetic anomaly data are output.

[0068] Furthermore, the aforementioned parameter estimation algorithm with regularization constraints solves online for the hard and soft magnetic disturbance parameters of the carrier, and performs real-time compensation for the geomagnetic vector at the current moment, ultimately outputting high-precision equilibrium geomagnetic data or magnetic anomaly data, including:

[0069] S41. Set a sliding time window of length N, obtain the attitude matrix at each moment within the window, and transform the purified geomagnetic vector estimate at the corresponding moment from the carrier coordinate system to the navigation coordinate system.

[0070] S42. Based on the carrier's location and time, query the IGRF model to obtain the background geomagnetic field vector in the navigation coordinate system;

[0071] S43. Within the sliding window, establish a system of linear equations with the carrier hard magnetic interference vector, soft magnetic interference matrix and geomagnetic anomaly vector as unknown parameters. The equations are composed of the transformed geomagnetic vector estimate, IGRF background field value and attitude matrix.

[0072] S44. The least squares method with Tikhonov regularization is used to solve the linear equation system. The regularization term is designed based on the symmetry and small trace characteristics of the soft magnetic interference matrix.

[0073] S45. Using the hard magnetic interference vector and soft magnetic interference matrix obtained from the solution, the estimated value of the geomagnetic vector in the carrier coordinate system at the current or next time moment is compensated. The calculation formula is as follows: ,in: The purified geomagnetic vector estimate is output from step S03. For the estimated hard magnetic interference vector, For the estimated soft magnetic interference matrix, This is the final compensated geomagnetic field vector.

[0074] Specifically, the system maintains a first-in-first-out data buffer (sliding window) of length N, continuously storing the geomagnetic vector estimate output from step S03, which has had motion noise suppressed. And the carrier attitude matrix acquired synchronously and calculated by the inertial navigation system. For each data point within the window, the processor performs a coordinate transformation. To unify the coordinates to the navigation coordinate system (n-system), the influence of vehicle attitude changes on vector observation is eliminated. Simultaneously, based on the submersible's real-time position (latitude and longitude) and time, the background field vector corresponding to that point is calculated using the International Geomagnetic Reference Field (IGRF) model. Next, based on the classical model of magnetic disturbance, all the observation equations in the window are solved simultaneously to construct a model with the hard magnetic disturbance vector H. hard The soft magnetic disturbance matrix K and an average geomagnetic anomaly vector assumed to be slowly varying within a window. The system of overdetermined linear equations, Ap=b, has unknowns. To overcome ill-conditioned problems caused by insufficient attitude variation or data noise, Tikhonov regularization is introduced during the solution process. In this process, the regularization matrix Γ is specifically designed to constrain the symmetry and "small trace" physical properties of K, thereby guiding the solution towards a reasonable physical range. Finally, the optimal estimate obtained using this window solution is... and You can then use the formula The freshest of the present moment Real-time compensation is performed, and the final balanced data is output. This window slides and updates as new data arrives, thus enabling online tracking and continuous compensation of disturbance parameters.

[0075] Secondly, step S44 above uses the least squares method with Tikhonov regularization to solve the system of linear equations, including:

[0076] S441. Construct a system of linear equations into a least squares problem: , where p is a parameter vector containing the elements of the hard magnetic interference vector and the soft magnetic interference matrix to be estimated, and A and b are constructed from the observation equations within the sliding window;

[0077] S442. Introduce the Tikhonov regularization term to construct a regularized least squares problem: , where Γ is the regularization matrix and λ is the regularization parameter;

[0078] S443. The regularization matrix is ​​designed based on the physical prior knowledge of the soft magnetic interference matrix. It is used to constrain the symmetry of K and the magnitude of its elements in order to improve the numerical stability of the solution.

[0079] Specifically, firstly, the system sets a fixed-length sliding time window (e.g., including data from the past 60 seconds) and continuously caches the sequence of estimated geomagnetic vectors in the carrier coordinate system, obtained from the aforementioned steps and with motion noise suppressed, along with the synchronously acquired and calculated sequence of carrier attitude matrices within this window. In implementation, the system maintains a first-in, first-out (FIFO) data queue to ensure the timeliness and continuity of data within the window. For each moment within the window, the processor uses the corresponding attitude matrix to uniformly rotate the estimated geomagnetic vectors in the carrier coordinate system to the navigation coordinate system (usually the NE-C coordinate system). This coordinate transformation is crucial, aiming to align the observed data with a stable spatial reference benchmark, thereby enabling the subsequent use of a known geophysical model—the International Geomagnetic Reference Field (IGRF) model. Based on the real-time position and time information provided by the submersible navigation system, the system dynamically queries the IGRF model to calculate the theoretical background geomagnetic field vector at that location in the navigation coordinate system, serving as a clean reference benchmark.

[0080] Next, within the defined sliding window, the algorithm establishes a system of equations based on the classical mathematical model of magnetic interference. This model shows that the geomagnetic vector observed in the navigation frame can be expressed as the sum of the IGRF background field, the actual geomagnetic anomaly (detected target), and the projection of the carrier interference onto the navigation frame. The carrier interference projection can be correlated with the assumed constant hard magnetic interference vector and soft magnetic interference matrix in the carrier coordinate system through the attitude matrix. Therefore, a vector equation can be written for each time point within the window. Stacking the equations for all times within the window forms an overdetermined linear system of equations with the hard magnetic interference vector, the soft magnetic interference matrix, and the average geomagnetic anomaly vector within the window as unknowns. However, due to potentially insufficient attitude changes or the presence of noise, this system of equations is often ill-conditioned and unstable when solved directly.

[0081] To address this mathematically ill-posed problem, claim 6 introduces a least-squares estimation method with regularization constraints. In its implementation, the algorithm does not directly solve the original least-squares problem, but rather solves a problem of minimizing a penalty function. This penalty function adds a regularization term to the original data fitting error term. The regularization matrix is ​​specially designed to encode prior physical knowledge about the soft magnetic interference matrix: for example, the soft magnetic interference matrix is ​​typically nearly symmetric, and the absolute values ​​of its elements (corresponding to the inductive magnetization coefficients) should be small. The regularization parameter is used to balance the degree of data fitting with the prior rationality of the solution. By solving this regularization problem, the algorithm can stably obtain the optimal interference parameter estimates within the current time window, even during periods of insufficient carrier maneuverability.

[0082] Finally, the system applies the real-time estimated interference parameters to compensate for measurements at the current or next moment. The processor loads the latest estimated hard magnetic interference vector and soft magnetic interference matrix, and performs compensation calculations for the latest cleaned geomagnetic vector located in the carrier coordinate system. This calculation is mathematically equivalent to subtracting the estimated hard magnetic interference from the observed values ​​and using the estimated soft magnetic interference matrix to perform inverse correction on the result, thereby directly outputting a "balanced" geomagnetic field vector in the carrier coordinate system that has eliminated most of the fixed magnetic interference. This vector can be used directly for detection or further subtracted from the IGRF background field to calculate magnetic anomalies. The entire process is executed cyclically as the sliding window is continuously updated, achieving continuous online tracking and compensation of carrier magnetic interference, enabling the deep-sea magnetometer to maintain high accuracy during long-term operations.

[0083] Example 2

[0084] Based on the above embodiment one, the high-precision balanced geomagnetic data or magnetic anomaly data output in step S04 are adjusted, including the following steps:

[0085] S05. Analyze the angular velocity and linear acceleration data output by the inertial measurement unit in real time and calculate the motion intensity index of the carrier;

[0086] S06. Based on the different preset intervals in which the exercise intensity index is located, dynamically adjust the preset amplitude of the process noise covariance matrix Q of the adaptive Kalman filter in step S03, and the length N of the sliding time window in step S04.

[0087] Specifically, when the motion intensity index is higher than the first threshold, the variance preset value of the corresponding motion noise state component in the Q matrix is ​​increased, and the window length N is shortened.

[0088] When the exercise intensity index is below the second threshold, the variance preset value is reduced and the window length N is increased.

[0089] Specifically, the processor continuously reads the three-axis angular velocity vector ω and the three-axis acceleration vector a from the inertial measurement unit. Motion intensity index I m The calculation is usually performed in the form of a weighted sum of squares of its modulus, for example... , where α and β are normalized weighting coefficients used to balance the different contributions of rotational and linear motion to magnetic interference. This index is calculated as a scalar and smoothed using a moving average to filter out instantaneous jitter, thus obtaining a stable metric reflecting the intensity of the current carrier motion.

[0090] Subsequently, the system maps this quantification index to specific adjustments to downstream algorithm parameters based on a preset threshold strategy. This strategy typically defines at least two key thresholds (the first threshold Th...). high Second threshold Thlow And Th high >Th low The motion state is divided into intervals such as "high dynamic", "medium dynamic", and "low dynamic / stable". When I m >Th high This indicates that the carrier is undergoing severe maneuvering or strong disturbance, and the magnetic noise caused by the motion changes rapidly and has high uncertainty. At this time, the controller will perform two key operations: First, dynamically increase the diagonal element values ​​corresponding to the motion noise state component v in the process noise covariance matrix Q of the adaptive Kalman filter in step S03. This increase operation is equivalent to acknowledging that the motion model has greater uncertainty in the filter model, prompting the filter to be more inclined to trust the current observation values ​​(i.e., magnetic sensor data), thereby speeding up the tracking speed of abrupt signals and preventing filter lag. Second, synchronously shorten the sliding time window length N used for magnetic disturbance parameter estimation in step S04. A shorter window means that the algorithm only relies on the data of the most recent moment for estimation, and has a faster response capability to the time-varying characteristics of disturbance parameters (such as the instantaneous magnetic susceptibility change caused by severe vibration), avoiding the "tailing" effect of old data on the current estimation.

[0091] Conversely, when I m <Th low At this point, the carrier's motion is smooth, indicating an approximate steady state or straight-line cruising state. Under these conditions, motion noise is low and changes slowly, and magnetic interference parameters also tend to stabilize. At this time, the controller adopts the opposite strategy: reducing the preset variance of the motion noise component in the Kalman filter's Q matrix. This increases the confidence of the filter model, favoring stronger smoothing filtering, thus effectively suppressing the remaining background noise and achieving higher steady-state estimation accuracy. Simultaneously, the sliding window length N is significantly increased. A longer window contains more data points with varying attitudes, greatly enriching the observational information of the overdetermined equations used to solve for the interference parameters. With the aid of regularization constraints, more accurate and stable estimates of hard and soft magnetic interference parameters can be obtained, significantly improving the absolute accuracy of the equilibrium data.

[0092] Example 3

[0093] This invention provides a non-transitory computer-readable storage medium storing at least one instruction or at least one program segment, which is loaded and executed by a processor to implement the following steps:

[0094] The raw triaxial magnetic signals of the three-component MEMS magnetic sensor, the attitude and motion data of the inertial measurement unit, and the environmental parameters of the temperature and pressure sensor are acquired simultaneously. The acquired data are then time-aligned and initially filtered to obtain the processed dataset.

[0095] Based on the pre-trained long short-term memory network model, and according to the time-series processed dataset, the sensor scaling factor error and zero bias drift caused by the time-varying environment in the deep sea are predicted and compensated in real time, and the preliminary corrected triaxial magnetic data are output.

[0096] Using the pre-corrected triaxial magnetic data as observations and combining it with inertial measurement unit data, an adaptive Kalman filter with state amplification is constructed. The true value of the geomagnetic field and the magnetic noise caused by the carrier motion are used together as the state for joint estimation. Motion interference is separated and suppressed in real time, and the purified geomagnetic vector estimate is output.

[0097] Using the purified geomagnetic vector estimation sequence within the sliding time window, the corresponding carrier attitude information, and the background field data provided by the international geomagnetic reference field model, the hard and soft magnetic disturbance parameters of the carrier are solved online through a parameter estimation algorithm with regularization constraints. The geomagnetic vector at the current moment is compensated in real time, and finally high-precision equilibrium geomagnetic data or magnetic anomaly data are output.

[0098] 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 of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0099] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0100] Example 4

[0101] This invention provides an electronic device, including a processor and a memory, wherein the memory stores at least one instruction or at least one program segment, and the at least one instruction or the at least one program segment is loaded and executed by the processor to implement the following steps:

[0102] The raw triaxial magnetic signals of the three-component MEMS magnetic sensor, the attitude and motion data of the inertial measurement unit, and the environmental parameters of the temperature and pressure sensor are acquired simultaneously. The acquired data are then time-aligned and initially filtered to obtain the processed dataset.

[0103] Based on the pre-trained long short-term memory network model, and according to the time-series processed dataset, the sensor scaling factor error and zero bias drift caused by the time-varying environment in the deep sea are predicted and compensated in real time, and the preliminary corrected triaxial magnetic data are output.

[0104] Using the pre-corrected triaxial magnetic data as observations and combining it with inertial measurement unit data, an adaptive Kalman filter with state amplification is constructed. The true value of the geomagnetic field and the magnetic noise caused by the carrier motion are used together as the state for joint estimation. Motion interference is separated and suppressed in real time, and the purified geomagnetic vector estimate is output.

[0105] Using the purified geomagnetic vector estimation sequence within the sliding time window, the corresponding carrier attitude information, and the background field data provided by the international geomagnetic reference field model, the hard and soft magnetic disturbance parameters of the carrier are solved online through a parameter estimation algorithm with regularization constraints. The geomagnetic vector at the current moment is compensated in real time, and finally high-precision equilibrium geomagnetic data or magnetic anomaly data are output.

[0106] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A deep-sea magnetometer based on a three-component MEMS magnetic sensor, characterized in that, Includes the following steps: S01. Simultaneously acquire the raw triaxial magnetic signals of the three-component MEMS magnetic sensor, the attitude and motion data of the inertial measurement unit, and the environmental parameters of the temperature and pressure sensor, and perform time alignment and initial filtering on the acquired data to obtain the processed dataset. S02. Based on the pre-trained long short-term memory network model, the sensor scaling factor error and zero bias drift caused by the time-varying environment in the deep sea are predicted and compensated in real time according to the processed dataset, and the preliminary corrected triaxial magnetic data are output. S03. Using the pre-corrected triaxial magnetic data as the observation value, and combining it with the inertial measurement unit data, construct a state-amplified adaptive Kalman filter, and use the true value of the geomagnetic field and the magnetic noise caused by the carrier motion as the state for joint estimation, separate and suppress motion interference in real time, and output the purified geomagnetic vector estimate. S04. Using the purified geomagnetic vector estimation sequence within the sliding time window, the corresponding carrier attitude information, and the background field data provided by the international geomagnetic reference field model, the carrier hard and soft magnetic interference parameters are solved online through a parameter estimation algorithm with regularization constraints. The geomagnetic vector at the current moment is compensated in real time, and finally high-precision balanced geomagnetic data or magnetic anomaly data are output.

2. A deep-sea magnetometer based on a three-component MEMS magnetic sensor according to claim 1, characterized in that, Step S02, which involves using the processed dataset based on the time series to predict and compensate for sensor scaling factor errors and zero-bias drift caused by the time-varying environment of the deep sea, and outputting pre-corrected triaxial magnetic data, includes: S21. Construct a dual-branch input LSTM network model. The first input branch is used to receive continuous time series temperature and pressure environmental parameters, and the second input branch is used to receive triaxial raw magnetic signals in the same time series. S22. The output of the LSTM network model is the sensor dynamic error parameter vector at the current moment. The sensor dynamic error parameter vector includes at least a three-axis scaling factor correction value and a three-axis zero bias correction value. S23. Using the aforementioned dynamic error parameter vector, perform dynamic correction calculations: ； in, Let be the original triaxial magnetic signal vector at time t. This is the diagonal matrix of the triaxial scaling factor error predicted by the LSTM model at time t. The three-axis zero-bias vector predicted by the LSTM model at time t; The original magnetic signal is inversely corrected in real time to obtain the dynamic error compensation value.

3. A deep-sea magnetometer based on a three-component MEMS magnetic sensor according to claim 1, characterized in that, The construction of the adaptive Kalman filter in step S3 includes the following steps: S31. Define the state vector as follows: ,in: The true three-component state of the geomagnetic field in the carrier coordinate system. The three components of the magnetic noise caused by the carrier motion; S32. Establish the dynamic model of the current inertial measurement unit as an empirical model related to the angular velocity and linear acceleration measured by the current inertial measurement unit, and add process noise; S33. The output of the pre-corrected triaxial magnetic data is used as the observation, and its relationship with the state vector is the sum of the two. S34. During the filtering iteration process, the Sage-Husa adaptive estimation algorithm is used to update the observed noise covariance matrix in real time to match the noise statistical characteristics of the carrier under different motion states.

4. A deep-sea magnetometer based on a three-component MEMS magnetic sensor according to claim 3, characterized in that, The empirical model related to the angular velocity and linear acceleration measured by the current inertial measurement unit in step S32 is obtained by pre-calibrating and fitting the inertial data and magnetic noise data under typical motion modes of the carrier. The formula is as follows: ,in, Let be the angular velocity and linear acceleration of the carrier at time k, respectively. This is process noise.

5. A deep-sea magnetometer based on a three-component MEMS magnetic sensor according to claim 1, characterized in that, Step S04, which involves using a parameter estimation algorithm with regularization constraints to solve for the hard and soft magnetic interference parameters of the carrier online and to compensate for the geomagnetic vector at the current moment in real time, ultimately outputs high-precision balanced geomagnetic data or magnetic anomaly data, includes: S41. Set a sliding time window of length N, obtain the attitude matrix at each moment within the window, and transform the purified geomagnetic vector estimate at the corresponding moment from the carrier coordinate system to the navigation coordinate system. S42. Based on the carrier's location and time, query the IGRF model to obtain the background geomagnetic field vector in the navigation coordinate system; S43. Within the sliding window, establish a system of linear equations with the carrier hard magnetic interference vector, soft magnetic interference matrix and geomagnetic anomaly vector as unknown parameters, wherein the equations are composed of the transformed geomagnetic vector estimate, IGRF background field value and attitude matrix. S44. Solve the linear equation system using the least squares method with Tikhonov regularization, wherein the regularization term is designed based on the symmetry and small trace characteristics of the soft magnetic interference matrix. S45. Using the hard magnetic interference vector and soft magnetic interference matrix obtained from the solution, the estimated value of the geomagnetic vector in the carrier coordinate system at the current or next time moment is compensated. The calculation formula is as follows: ,in: The purified geomagnetic vector estimate output in step S03. For the estimated hard magnetic interference vector, For the estimated soft magnetic interference matrix, This is the final compensated geomagnetic field vector.

6. A deep-sea magnetometer based on a three-component MEMS magnetic sensor according to claim 5, characterized in that, The step S44, which involves solving the linear equation system using the least squares method with Tikhonov regularization, includes: S441. Construct the system of linear equations into a least squares problem: , where p is a parameter vector containing the elements of the hard magnetic interference vector and the soft magnetic interference matrix to be estimated, and A and b are constructed from the observation equations within the sliding window; S442. Introduce the Tikhonov regularization term to construct a regularized least squares problem: , where Γ is the regularization matrix and λ is the regularization parameter; S443. The regularization matrix is ​​designed based on the physical prior knowledge of the soft magnetic interference matrix and is used to constrain the symmetry of K and the magnitude of its elements in order to improve the numerical stability of the solution.

7. A deep-sea magnetometer based on a three-component MEMS magnetic sensor according to claim 1, characterized in that, The high-precision balanced geomagnetic data or magnetic anomaly data output in step S04 are adjusted by the following steps: S05. Analyze the angular velocity and linear acceleration data output by the inertial measurement unit in real time, and calculate the motion intensity index of the carrier; S06. Based on the different preset intervals in which the exercise intensity index is located, dynamically adjust the preset amplitude of the process noise covariance matrix Q of the adaptive Kalman filter in step S03, and the length N of the sliding time window in step S04. Specifically, when the motion intensity index is higher than the first threshold, the variance preset value of the corresponding motion noise state component in the Q matrix is ​​increased, and the window length N is shortened. When the exercise intensity index is lower than the second threshold, the variance preset value is reduced and the window length N is increased.

8. 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 steps of the deep-sea magnetometer based on a three-component MEMS magnetic sensor as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the deep-sea magnetometer based on a three-component MEMS magnetic sensor as described in any one of claims 1 to 7.

Images