Aeromagnetic compensation method and device for multi-source sensor

Abstract

The embodiment of the invention discloses a multisource sensor aeromagnetic compensation method and device, and the method comprises the steps: obtaining the magnetic field values of a magnetometer at a plurality of historical moments and the interference field observation values of other sensors in a flying aircraft; taking the magnetic field value and the interference field observation value as input values, extracting spatial-temporal characteristic representation of the input values by using a graph neural network model, and predicting a magnetic field compensation value at the current moment according to the spatial-temporal characteristic representation; wherein the graph neural network model is obtained by training based on a measured value of a multi-source sensor and a magnetic field reference true value after interference elimination, and the multi-source sensor comprises a magnetometer and other sensors. According to the method, the magnetic field compensation precision is improved, generalization and robustness are enhanced, and cost and complexity are reduced.

Description

Technical Field

[0001] This invention relates to the field of neural network technology, and in particular to a method and apparatus for multi-source sensor aeromagnetic compensation. Background Technology

[0002] During flight, aircraft generate complex magnetic fields from their airframe structure, onboard electrical equipment, and cabin instruments. These magnetic interferences superimpose on the Earth's magnetic field signal, causing significant errors in the measurements taken by airborne magnetometers. Without effective compensation, the measured magnetic field data will deviate from the true value, affecting the accuracy of airborne magnetic surveys in geophysical exploration and navigation.

[0003] To weaken or eliminate the influence of airborne interfering magnetic fields on magnetic measurements and restore signals close to the true Earth's magnetic field, traditional aeromagnetic compensation methods mainly rely on hardware. This typically involves installing a tail boom at the tail of the aircraft, with a magnetometer positioned at its end. Because the tail boom is far from electrical equipment and metal structures within the cabin, its magnetic interference is relatively small, and the classic Tolles-Lawson method can yield relatively accurate magnetic measurement results. While this method offers a certain degree of reliability in engineering, the installation of the tail boom magnetometer is not only highly dependent on hardware but also affects the aircraft's structure and aerodynamic performance, thus impacting the accuracy of the magnetic measurement results. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-source sensor aeromagnetic compensation method and apparatus to solve the problems in the prior art.

[0005] To solve the above-mentioned technical problems, the embodiments of the present invention are implemented as follows: In a first aspect, the present invention provides a multi-source sensor aeromagnetic compensation method, comprising: Inside the aircraft in flight, acquire magnetic field values ​​from the magnetometer at multiple historical moments, as well as interference field observations from other sensors; Using the observed magnetic field value and interference field value as input values, a graph neural network model is used to extract the spatiotemporal feature representation of the input values, and the magnetic field compensation value at the current moment is predicted based on the spatiotemporal feature representation. The graph neural network model is trained based on measurements from multiple sources and the true reference value of the magnetic field after interference elimination. The multiple sources include magnetometers and other sensors.

[0006] Secondly, embodiments of the present invention provide a multi-source sensor aeromagnetic compensation device, comprising: The acquisition module is used to acquire the magnetic field values ​​of the magnetometer and the interference field observations of other sensors at multiple historical moments inside the aircraft in flight. The prediction module is used to take the observed magnetic field value and the interference field value as input values, use a graph neural network model to extract the spatiotemporal feature representation of the input values, and predict the magnetic field compensation value at the current moment based on the spatiotemporal feature representation. The graph neural network model is trained based on measurements from multiple sources and the true reference value of the magnetic field after interference elimination. The multiple sources include magnetometers and other sensors.

[0007] Thirdly, embodiments of the present invention provide an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the multi-source sensor aeromagnetic compensation method provided in the above embodiments.

[0008] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the multi-source sensor aeromagnetic compensation method provided in the above embodiments.

[0009] Fifthly, embodiments of the present invention provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the multi-source sensor aeromagnetic compensation method provided in the above embodiments.

[0010] As can be seen from the technical solutions provided by the above embodiments of the present invention, the embodiments of the present invention acquire the magnetic field values ​​of the magnetometer at multiple historical moments and the interference field observation values ​​of other sensors inside the aircraft in flight. Using the magnetic field values ​​and interference field observation values ​​as input values, a graph neural network model is used to extract the spatiotemporal feature representation of the input values. Based on the spatiotemporal feature representation, the magnetic field compensation value at the current moment is predicted. The graph neural network model is trained based on the measurement values ​​of multiple source sensors and the true reference value of the magnetic field after interference elimination. The multiple source sensors include the magnetometer and other sensors. The magnetic field compensation value predicted through the above process, compared with the acquired magnetic field value, eliminates the influence of the interfering magnetic field, resulting in a more accurate result and improved compensation accuracy. Prediction based on the graph neural network model enhances generalization and robustness, and provides stronger stability. It eliminates the need for a tail boom magnetometer, does not rely on hardware implementation, and reduces cost and complexity. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating a multi-source sensor aeromagnetic compensation method provided in an embodiment of the present invention. Figure 2 A schematic diagram of a graph neural network model provided in an embodiment of the present invention; Figure 3 A schematic diagram of the topology of the spatiotemporal attention network provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the graph neural network model establishment process provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of the multi-source sensor aeromagnetic compensation device provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of the multi-source sensor aeromagnetic compensation system provided in an embodiment of the present invention; Figure 7 This is a schematic diagram comparing the aeromagnetic compensation accuracy provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0013] This invention provides a multi-source sensor aeromagnetic compensation method, apparatus, and electronic device.

[0014] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0015] First, the terms used in this invention will be explained.

[0016] Calibration flights are crucial test flights used in aeromagnetic compensation to acquire the aircraft's magnetic interference characteristics. They typically involve flying the aircraft in different headings (e.g., north, east, south, west) and specific maneuvers (e.g., pitch, roll, yaw) to fully excite the aircraft's magnetic interference. Using the magnetometer data collected during calibration flights, a model can be established showing the relationship between the aircraft's magnetic interference and parameters such as flight attitude, and the compensation coefficients can be calculated accordingly. The results of calibration flights provide standardized reference data for subsequent applications of the TL method or other compensation methods.

[0017] The Tolles–Lawson (TL) method is a classic mathematical modeling approach in the field of airborne magnetic data compensation. Its basic idea is to construct a linear relationship model between aircraft magnetic interference and variables such as flight attitude by using magnetometer measurement data acquired during calibration flights, and then solve for a set of compensation coefficients. These compensation coefficients are then used in subsequent flights to correct the magnetometer observations, thereby eliminating or reducing the impact of airborne magnetic interference. Named after its proposers, Tolles and Lawson, this method has long been the standard method for compensating airborne magnetic measurement data.

[0018] Data filtering: A signal preprocessing technique used in magnetic compensation. It involves designing low-pass, high-pass, or band-pass filters to process flight magnetic measurement data, suppressing high-frequency noise or low-frequency drift, thereby preserving accurate magnetic anomaly information. For example, removing approximately 50,000 nT of geomagnetic background field from the raw observation data highlights magnetic anomaly signals on the order of several hundred nT, thus improving the accuracy of compensation and detection.

[0019] INS (Inertial Navigation System): An autonomous navigation system that does not rely on external information or radiate energy to the outside. Its operating environment includes not only the air and ground, but also underwater. The basic working principle of INS is based on Newton's laws of motion. By measuring the acceleration of the carrier in the inertial reference frame, integrating it over time, and transforming it into the navigation coordinate system, information such as velocity, yaw angle, and position in the navigation coordinate system can be obtained.

[0020] The multi-source sensor aeromagnetic compensation method and apparatus provided in this invention can be applied to airborne devices within aircraft, including but not limited to scenarios where magnetic field values ​​of magnetometers within aircraft are to be compensated. Through the above method and apparatus, magnetic field compensation accuracy can be improved, generalization and robustness enhanced, and cost and complexity reduced.

[0021] like Figure 1 As shown, this embodiment of the invention provides a multi-source sensor aeromagnetic compensation method, which may specifically include the following steps: S102: Inside the flying aircraft, acquire the magnetic field values ​​of the magnetometer at multiple historical moments, as well as the interference field observations of other sensors.

[0022] The historical time can include N times before the current time, and the value of N can be preset without any limitation.

[0023] S104: Using the observed magnetic field value and the disturbance field value as input values, a graph neural network model is used to extract the spatiotemporal feature representation of the input values, and the magnetic field compensation value at the current moment is predicted based on the spatiotemporal feature representation.

[0024] In actual flight, the trained graph neural network model can be deployed on airborne equipment, and magnetometers and other sensors can output data in real time as input to the graph neural network model, enabling online inference of magnetic field compensation values.

[0025] The current time can be any time. By repeating the above method multiple times, the magnetic field compensation values ​​at multiple times can be predicted, thereby obtaining a magnetic field compensation sequence. Specifically, a magnetic field compensation value or a magnetic field compensation sequence can be obtained as needed, and the embodiments of the present invention do not impose specific limitations on this.

[0026] The aforementioned graph neural network model is trained based on measurements from multiple sensors and the true reference value of the magnetic field after interference elimination. Specifically, the graph neural network model can be trained in a supervised manner using multi-source sensor data acquired during free flight and the corresponding tail boom magnetometer compensation true value. The root mean square error (RMSE) index is used as the loss function, and the parameters are iteratively optimized through gradient descent to solidify the graph neural network model parameters, thereby completing the training for subsequent real-time compensation applications.

[0027] The aforementioned multi-source sensors include magnetometers and other sensors. There may be one or more magnetometers, and the number of other sensors is not limited. Other sensors include, but are not limited to: current sensors, voltage sensors, strobe lights, avionics instruments, etc.

[0028] In this embodiment of the invention, step S104, which involves predicting the magnetic field compensation value at the current moment based on spatiotemporal characteristics, may include: The graph neural network model uses LSTM (Long Short-Term Memory) to calculate the spatiotemporal feature representation, and after layer normalization and fully connected layer processing, outputs the magnetic field compensation value at the current moment.

[0029] The LSTM calculation and layer normalization process can be repeated multiple times. For example, if a two-layer structure is used, the process can be repeated twice. The second processing takes the result of the first LSTM calculation and layer normalization process as input, and the result of the second processing is input into the fully connected layer for processing to obtain the final output of the magnetic field compensation value at the current moment.

[0030] In this embodiment of the invention, the step S104 above, which uses a graph neural network model to extract the spatiotemporal feature representation of the input value, may include: Graph neural network models include spatiotemporal attention networks, which include spatial attention networks and temporal attention networks. Use a spatial attention network to extract spatial augmentation features of the input values; Spatiotemporal feature representations are generated using a temporal attention network based on spatial augmented features.

[0031] In this embodiment of the invention, the spatiotemporal attention network may include one or more layers. When the spatiotemporal attention network includes multiple layers, each layer extracts spatiotemporal feature representations as output. The first layer takes the aforementioned input values ​​as input, and the remaining layers take the output of the previous layer as input.

[0032] In one implementation, the spatiotemporal attention network includes two layers, wherein the first layer extracts spatiotemporal feature representations of the input values, and the second layer extracts spatiotemporal feature representations again from the spatiotemporal feature representations as the output values ​​of the spatiotemporal attention network.

[0033] See Figure 2 In one implementation, the graph neural network model includes two layers of spatiotemporal attention networks, each comprising a spatial attention network and a temporal attention network. The first layer of spatiotemporal attention network: first, it establishes spatial topological dependencies between nodes using a spatial attention network (Spatial GATv2) to extract spatial augmentation features from the input values; then, it captures dynamic features that change over time using a temporal attention network (Temporal GATv2), generating a spatiotemporal feature representation based on the spatial augmentation features, thus obtaining the spatiotemporal feature representation of the first layer. The second layer of spatiotemporal attention network: the output of the first layer of spatiotemporal attention network is input into the Spatial GATv2 and Temporal GATv2 of the second layer of spatiotemporal attention network, further deepening the joint modeling of node spatial dependencies and temporal dynamics, thus obtaining the spatiotemporal feature representation of the second layer, which serves as the input for the task-aware prediction function.

[0034] The task-aware prediction function comprises a three-layer structure. The first and second layers are identical, both including LSTM and LN processing functions. The third layer is a fully connected (FC) layer (not shown in the figure). The spatiotemporal feature representation output by the aforementioned second-layer spatiotemporal attention network is first input into the first-layer LSTM to further capture deep dependencies across time slices. Layer normalization LN processing is then added after the LSTM output to stabilize the training process and prevent gradient vanishing or exploding. The result after the first-layer LN processing is then input into the second layer. After LSTM and LN processing, the ability to model long-term dependencies is enhanced, further refining and compensating for relevant temporal information. Finally, the task-aware prediction output is achieved through a fully connected layer, obtaining the magnetic field compensation index at the target time.

[0035] It should be noted that during the training of the graph neural network model, the collected and preprocessed data, along with the adjacency matrix, serve as the input to the spatiotemporal attention network, and the learned spatiotemporal feature representation serves as the output of the spatiotemporal attention network. This spatiotemporal feature representation is then processed by LSTM and LN, and finally processed by a fully connected layer to obtain the magnetic field compensation value, which serves as the output of the graph neural network model.

[0036] In this embodiment of the invention, the graph neural network model has a static topological structure, and the node connection relationship remains unchanged during the training and inference process of the graph neural network model, ensuring that time dependence can be modeled on a fixed spatial relationship.

[0037] In this embodiment of the invention, the spatiotemporal attention network may include magnetometer nodes and interference source nodes. There may be one or more magnetometer nodes and one or more interference source nodes, without specific limitation. For example, the spatiotemporal attention network includes 3 vector magnetometer nodes, 5-7 scalar magnetometer nodes, and several interference source nodes (such as sensors). The feature vector of each node may consist of various information to enrich the node representation and fuse synchronized spatial and state information. Specifically, the feature vector of the magnetometer node includes the real-time measured magnetic field value. The feature vector of the vector magnetometer node includes three-dimensional magnetic field components. The feature vector of the scalar magnetometer node is a one-dimensional magnetic field strength value. The feature vector of the interference source node includes interference field observations (such as current / voltage, barometric altitude, inertial measurements, etc.), typically a one-dimensional feature.

[0038] In one implementation, attitude angle information can be incorporated as a dynamic feature into the feature vector of the magnetometer node. Attitude angle information includes the aircraft's global attitude, such as roll, pitch, and yaw angles. This approach can more comprehensively reflect the changes in attitude on the soft and hard ferromagnetic interferences of the aircraft, allowing the graph neural network model to use attitude information to interpret the correlation between changes in magnetic field readings.

[0039] It should be noted that the feature vector of each node in the topology can also incorporate time window statistics, such as the sliding window mean or first-order difference, which can enhance the description of temporal dynamics and provide a useful supplement to the temporal modeling of graph neural network models. Among them, the sliding window mean provides a smooth trend of recent observations, while the first-order difference can reflect the instantaneous rate of change of the measurement value.

[0040] In the graph neural network model of this invention, each node has a multi-dimensional feature description, which includes both the sensor's own readings and contextual information such as spatial location and flight status.

[0041] In this embodiment of the invention, the spatiotemporal attention network may include magnetometer nodes and interference source nodes. The connection relationships between nodes in the spatiotemporal attention network may include: There are edges connecting the magnetometer nodes; The magnetometer node and the interference source node are selectively connected based on statistical correlation or spatial proximity; There are no edge connections between the interference source nodes.

[0042] In one implementation, selective connections are made between the magnetometer node and the interference source node based on statistical correlation, including: Calculate the Pearson correlation coefficient between the magnetic field values ​​at the magnetometer nodes and the observed interference field values ​​at the interference source nodes; When the Pearson correlation coefficient is greater than or equal to the correlation threshold, there is an edge connection between the magnetometer node and the interference source node. When the Pearson correlation coefficient is less than the correlation threshold, there is no edge connection between the magnetometer node and the interference source node.

[0043] The correlation threshold can be preset as needed, for example, setting the correlation threshold to 0.6, based on the Pearson correlation coefficient. In this case, there is an edge connection between the magnetometer node and the interference source node.

[0044] In one implementation, the magnetometer node and the interference source node are selectively connected based on spatial proximity, including: Calculate the distance between the magnetometer node and the interference source node; When the distance is less than the distance threshold, there is an edge connection between the magnetometer node and the interference source node; When the distance is greater than or equal to the distance threshold, there is no edge connection between the magnetometer node and the interference source node.

[0045] The distance threshold can be preset as needed. For example, if the distance threshold is set to 2m, there will be an edge connection between the magnetometer node and the interference source node when the distance d between the magnetometer node and the interference source node is less than 2m.

[0046] See Figure 3 The diagram illustrates two magnetometer nodes and four interference source nodes. In practical applications, the number of magnetometer nodes and interference source nodes is not limited. The topology of the graph neural network model in this embodiment can be specifically established as follows: 1) Connections between magnetometer nodes: All magnetometer nodes are connected to each other by edges (forming an undirected complete subgraph). The advantage of this connection method is that if a magnetometer node at a certain location detects a change in interference, its neighboring magnetometer nodes can obtain this information through edge connections, thereby helping the network to determine the spatial distribution of interference sources and the global geomagnetic field level.

[0047] Edges between magnetometer nodes can also be weighted, specifically based on distance or historical correlation between the nodes. For example, edges between closer magnetometer nodes can be assigned larger weights, while edges between farther nodes can be assigned smaller weights. This allows physically adjacent magnetometer nodes to form denser subgroups, which better highlights the consistency of local disturbances during convolution in the graph neural network model. Additionally, a small number of long-distance connections can be added to ensure overall graph connectivity and global information propagation.

[0048] 2) Connection between magnetometer nodes and interference source nodes: Selective edge connections are made based on statistical correlation or spatial proximity to ensure that the connection relationship is reasonable, sparse and efficient.

[0049] Specifically, the selective linking based on statistical correlation is as follows: For any pair of magnetometer nodes and interference source nodes, calculate the Pearson correlation coefficient between the magnetic field value of the magnetometer node and the observed interference field value of the interference source node; if the Pearson correlation coefficient is greater than or equal to a set correlation threshold (e.g., ... If the Pearson correlation coefficient is less than the correlation threshold, then the correlation between the interference source and the magnetometer is considered weak, and no connection is needed between the magnetometer node and the interference source node, thus avoiding invalid information interfering with the graph learning process.

[0050] The selective edge connection based on spatial proximity is as follows: For any pair of magnetometer nodes and interference source nodes, the distance between the magnetometer node and the interference source node is calculated. If the distance is less than a set distance threshold (e.g., d < 2 m), an edge is established between the magnetometer node and the interference source node. If the distance is greater than or equal to the distance threshold, the distance is too far, and the interference effect is considered negligible; therefore, no connection is needed between the magnetometer node and the interference source node, i.e., no edge is established. Furthermore, weights can be assigned to the established edges, with greater weights for closer distances and smaller weights for farther distances, to reflect the decreasing trend of electromagnetic interference intensity with distance.

[0051] For example, for any two nodes i and j in the magnetometer node set M and the interference source node set S, the Euclidean distance between the two nodes in the computer volume coordinate system is... Pearson correlation coefficient between the two nodes .like or If node i and node j are connected by an edge, then there is an edge connecting them. Distance threshold This is the correlation threshold.

[0052] 3) Connections between interference source nodes: Generally, different interference source nodes do not need to be directly connected because they represent independent devices or environmental factors.

[0053] This mechanism ensures that only interference sources that are physically or statistically significantly coupled with the magnetometer are connected. This approach not only reduces the complexity of the graph structure but also conforms to the electromagnetic theory principle that "interference attenuates with distance," thus improving the physical reliability and inference accuracy of the graph neural network model.

[0054] The aforementioned method of establishing the topology forms a multi-partition fully connected graph, with tightly interconnected subgraphs between magnetometer nodes and a complete bidirectional connection network between interference source nodes and magnetometer nodes. On the one hand, known physical distributions (distance, location) are used to guide the weighting and connection of edges, ensuring that the connection relationships are reasonable and reliable; on the other hand, the relatively dense connections ensure that information from any node can be transmitted to other nodes in the graph through several steps, improving the sufficiency of information propagation.

[0055] In this embodiment of the invention, the spatiotemporal attention network may include magnetometer nodes and interference source nodes. The connection relationship between nodes in the spatiotemporal attention network can be defined by an adjacency matrix, the calculation formula of which is as follows: ; ; ; Where A is the initial adjacency matrix. Let be the adjacency matrix between magnetometer nodes. This is the adjacency matrix between the magnetometer nodes and the interference source nodes. for The transpose of the matrix, This is the adjacency matrix between interference source nodes. This is the intermediate adjacency matrix. It is the identity matrix. It is an adjacency matrix. It is a diagonal matrix.

[0056] The set of nodes is denoted as M is the set of magnetometer nodes, S is the set of interference source nodes, and the initial adjacency matrix is... .

[0057] It should be noted that the calculation During the process, for each magnetometer node, k-nearest neighbors (based on spatial distance) and highly correlated edges (based on mutual information or Pearson's algorithm) can be established, and the union of the two can be obtained. This maintains the sparsity completeness of connectivity.

[0058] In this embodiment of the invention, the edges in the spatiotemporal attention network described above may also have weights, and the formula for calculating these weights is as follows: ; ; in, Let be the distance kernel between node i and node j. For the related kernels of node i and node j, Let be the Euclidean distance between node i and node j in the body coordinate system. Using distance as the scale, Let p be the Pearson correlation coefficient between node i and node j, and p be the exponentiation coefficient, where p=1 or p=2. Let be the weight of the edge between node i and node j. These are the weighting coefficients.

[0059] In this embodiment of the invention, based on the weights of the edges in the spatiotemporal attention network, the edges can be processed as follows: for all edges connected to each node in the spatiotemporal attention network, k edges are selected and retained according to their weights from largest to smallest, and the remaining edges are deleted, thereby suppressing redundant edges and noisy connections. Here, k can be preset as needed, and its specific value is not limited.

[0060] The method provided in this invention acquires magnetic field values ​​from a magnetometer and interference field observations from other sensors at multiple historical moments within an aircraft in flight. Using the magnetic field values ​​and interference field observations as input values, a graph neural network model is used to extract spatiotemporal feature representations of the input values. Based on these spatiotemporal feature representations, the magnetic field compensation value for the current moment is predicted. The graph neural network model is trained based on measurements from multiple sensors and a true reference value of the magnetic field after interference elimination. These multiple sensors include a magnetometer and other sensors. The magnetic field compensation value predicted through this process is more accurate than the acquired magnetic field value because it eliminates the influence of interfering magnetic fields, thus improving compensation precision. Prediction based on a graph neural network model enhances generalization and robustness, resulting in stronger stability. It eliminates the need for a tail boom magnetometer, is not hardware-dependent, and reduces cost and complexity.

[0061] In practical applications, this method does not rely on dedicated hardware such as tail boom magnetometers. Instead, compensation is achieved by fusing data from multiple sensors within the cabin, improving the method's versatility and applicability. Using a spatiotemporal graph neural network to model the sensor data not only explicitly depicts the spatial topological relationships and temporal dependencies between multiple sensors but also allows for differentiated processing of data characteristics from different sensor types. This deep fusion of multi-dimensional information across space, time, and sensor type significantly improves compensation accuracy and enhances the model's robustness and generalization ability under complex flight conditions.

[0062] See Figure 4 In this embodiment of the invention, the process of establishing and training the graph neural network model may include: 1) Data collection: First, magnetometers, including scalar or vector magnetometers, are deployed at multiple locations on the aircraft as primary magnetic field measurement nodes. Simultaneously, status data from some onboard equipment is collected, including current and voltage parameters from electrical devices (such as power systems, onboard computers, strobe lights, and avionics instruments) as interference characteristic inputs. Outputs from environmental and navigation equipment (such as barometric altimeters) serve as auxiliary inputs. The inertial navigation system (INS) provides navigation calculations such as northbound velocity and vertical velocity.

[0063] Data was collected from all sensors deployed on the aircraft at multiple time points to form a multi-source sensor time-series dataset. During the data acquisition phase, it was ensured that the data from each sensor was recorded according to a unified time reference (e.g., a sampling frequency of 10Hz) and that the data fully covered the state parameters and magnetic field measurements of the aircraft during various flight maneuvers, providing sufficient data support for subsequent modeling.

[0064] 2) Data preprocessing A dedicated calibration flight mission (e.g., FOM calibration flight) is pre-arranged. During the calibration flight, the aircraft flies according to prescribed subjects in four headings: north, east, south, and west, performing standard maneuvers (such as ±10° roll, ±5° pitch, and ±5° sideslip) on each segment to fully stimulate airborne magnetic interference signals. Using the tail boom magnetometer data acquired during the calibration flight, a TL model is employed to perform standardized compensation calculations for the aircraft's interfering magnetic field. Specifically, this includes: decomposing and modeling the magnetic interference using the TL method, solving for the compensation coefficients of the aircraft's magnetic interference, and using these coefficients to correct the tail boom magnetometer measurements, obtaining the true reference value of the magnetic field after eliminating airframe interference. The tail boom magnetometer output data compensated by the TL method is used as the label (true value) for subsequent training of the graph neural network model.

[0065] The input data is then preprocessed. First, data cleaning is performed to remove missing, duplicate, and anomalous values ​​caused by instrument malfunctions or communication anomalies during flight. Next, data generation and transformation are performed, including converting geographic coordinates to Cartesian coordinates and checking the consistency between the transformation result and the original coordinates of the dataset (allowable error ±1.4cm). Velocity and specific force are obtained from the trajectory position using the finite difference method. The direction cosine matrix from the body coordinate system to the navigation coordinate system is calculated based on roll, pitch, and yaw angles. Corrected specific force data is obtained using the specific force measured by the inertial navigation system combined with the rotational drift angle. Data filtering is then performed, using a bandpass filter to suppress low-frequency drift and high-frequency noise while retaining valid magnetic anomaly signals. Finally, data time scale alignment is performed. The time delay is calculated using the spatial position difference between the sensor and the reference point, as well as the flight speed, and is approximated as constant throughout the flight to achieve strict correspondence between data from different sensors on the same time axis. Finally, data standardization is performed using the z-score scaling method to uniformly scale multi-source input features such as magnetic field, current, voltage, and velocity to the same dimension range, in order to ensure the stability and convergence of subsequent compensation model training.

[0066] 3) Graph construction First, establish the initial adjacency matrix. ; The adjacency matrix between magnetometer nodes can maintain a sparse completeness strategy: for each magnetometer, establish a k-nearest neighbor (based on spatial distance) and a highly correlated edge (based on mutual information or Pearson), and take the union of the two; : The adjacency matrix between magnetometer nodes and interference source nodes, with edges connected only if statistical correlation or spatial proximity is satisfied; The adjacency matrix between interference source nodes is not connected by default. If there is known coupling (shared power supply / shared ground / synchronous triggering), weak edges can be established to maintain the information path.

[0067] Then calculate the adjacency matrix. : ; ; in, This is the intermediate adjacency matrix. It is the identity matrix. It is an adjacency matrix. It is a diagonal matrix.

[0068] Additionally, weights can be assigned to the edges in the spatiotemporal attention network. The formula for calculating the weights is as follows: ; ; Here, node i and node j are any two nodes in the spatiotemporal attention network. Let be the distance kernel between node i and node j. For the related kernels of node i and node j, This is the Euclidean distance (or the equivalent distance after structural correction) between nodes i and j in the body coordinate system. Using distance as the scale, Let Pearson correlation coefficient be the correlation coefficient between node i and node j. Let be the weight of the edge between node i and node j. These are the weighting coefficients.

[0069] 4) Model Building Based on the aforementioned graph topology and adjacency matrix, multi-source sensor nodes and their features are embedded into the graph structure. The first layer, a spatiotemporal graph attention network, inputs the values ​​into a spatial attention network (Spatial GATv2) to model the spatial topological dependencies between nodes, obtaining spatially enhanced features. These features are then input into a temporal attention network (Temporal GATv2) to capture dynamic features that change over time, resulting in the first layer's spatiotemporal feature representation. The second layer, a spatiotemporal graph attention network, inputs the spatiotemporal feature representation from the first layer back into the second layers of Spatial GATv2 and Temporal GATv2, further deepening the joint modeling of sensor spatial dependencies and temporal dynamics, resulting in the second layer's spatiotemporal feature representation, which provides input for the subsequent task perception and prediction part. The spatiotemporal feature representation of the second layer is input into the first layer of the Long Short-Term Memory (LSTM) network to further capture deep dependencies across time slices. Layer normalization (LN) is added after the LSTM output to stabilize the training process and avoid gradient vanishing or exploding. Then, it is input into the second layer of the LSTM to enhance the ability to model long-term dependencies and further refine the relevant temporal information for compensation. Finally, the task-aware prediction output is achieved through a fully connected layer (FC) to obtain the magnetic field compensation value at the target time.

[0070] 5) Model training and optimization The model performance is evaluated using a test set independent of the training data. Specifically, multi-source sensor data collected during actual flight is input into the trained graph neural network model to obtain a predicted compensation sequence, which is then compared with the result of compensation by the tail boom magnetometer using the TL method. The compensation effect of the model is evaluated using multiple indicators such as root mean square error (RMSE), standard deviation (STD), and improvement ratio (IR). Furthermore, to verify the robustness of the model under different flight states and disturbance conditions, flight data under different headings, altitudes, and loads are selected for testing. If the model exhibits a low error level under all flight conditions, it demonstrates that the method has good accuracy and generalization ability and can be used for real-time magnetic compensation tasks in actual flight.

[0071] The above describes a multi-source sensor aeromagnetic compensation method provided by embodiments of the present invention. Based on the same idea, embodiments of the present invention also provide a multi-source sensor aeromagnetic compensation device, such as... Figure 5 As shown, the multi-source sensor aeromagnetic compensation device includes the following modules.

[0072] The acquisition module 501 is used to acquire the magnetic field values ​​of the magnetometer and the interference field observation values ​​of other sensors at multiple historical moments inside the aircraft in flight.

[0073] The prediction module 502 is used to take the magnetic field value and the observed value of the interference field as input values, use a graph neural network model to extract the spatiotemporal feature representation of the input values, and predict the magnetic field compensation value at the current moment based on the spatiotemporal feature representation. The graph neural network model is trained based on measurements from multiple sources and the true reference value of the magnetic field after interference elimination. The multiple sources include magnetometers and other sensors.

[0074] In this embodiment of the invention, the prediction module 502 is used for: The spatiotemporal feature representation is calculated using a graph neural network model with a long short-term memory network. After layer normalization and fully connected layer processing, the magnetic field compensation value at the current moment is output.

[0075] In this embodiment of the invention, the graph neural network model includes a spatiotemporal attention network, which comprises a spatial attention network and a temporal attention network. The prediction module 502 is used for: Use a spatial attention network to extract spatial augmentation features of the input values; Spatiotemporal feature representations are generated using a temporal attention network based on spatial augmented features.

[0076] In this embodiment of the invention, the spatiotemporal attention network includes magnetometer nodes and interference source nodes, and the connection relationships between the nodes in the spatiotemporal attention network include: There are edges connecting the magnetometer nodes; The magnetometer node and the interference source node are selectively connected based on statistical correlation or spatial proximity; There are no edge connections between the interference source nodes.

[0077] In one implementation, selective connections are made between the magnetometer node and the interference source node based on statistical correlation, including: Calculate the Pearson correlation coefficient between the magnetic field values ​​at the magnetometer nodes and the observed interference field values ​​at the interference source nodes; When the Pearson correlation coefficient is greater than or equal to the correlation threshold, there is an edge connection between the magnetometer node and the interference source node. When the Pearson correlation coefficient is less than the correlation threshold, there is no edge connection between the magnetometer node and the interference source node.

[0078] In one implementation, the magnetometer node and the interference source node are selectively connected based on spatial proximity, including: Calculate the distance between the magnetometer node and the interference source node; When the distance is less than the distance threshold, there is an edge connection between the magnetometer node and the interference source node; When the distance is greater than or equal to the distance threshold, there is no edge connection between the magnetometer node and the interference source node.

[0079] In this embodiment of the invention, the spatiotemporal attention network includes magnetometer nodes and interference source nodes. The connection relationship between nodes in the spatiotemporal attention network is defined by an adjacency matrix, which is calculated using the following formula: ; ; ; Where A is the initial adjacency matrix. Let be the adjacency matrix between magnetometer nodes. This is the adjacency matrix between the magnetometer nodes and the interference source nodes. for The transpose of the matrix, This is the adjacency matrix between interference source nodes. This is the intermediate adjacency matrix. It is the identity matrix. It is an adjacency matrix. It is a diagonal matrix.

[0080] In this embodiment of the invention, the edges in the spatiotemporal attention network also have weights, and the formula for calculating these weights is as follows: ; ; Here, node i and node j are any two nodes in the spatiotemporal attention network. Let be the distance kernel between node i and node j. For the related kernels of node i and node j, Let be the Euclidean distance between node i and node j in the body coordinate system. Using distance as the scale, Let p be the Pearson correlation coefficient between node i and node j, and p be the exponentiation coefficient, where p=1 or p=2. Let be the weight of the edge between node i and node j. These are the weighting coefficients.

[0081] In this embodiment of the invention, the spatiotemporal attention network may include one or more layers. When the spatiotemporal attention network includes multiple layers, each layer extracts spatiotemporal feature representations as output. The first layer takes the aforementioned input values ​​as input, and the remaining layers take the output of the previous layer as input.

[0082] In one implementation, the spatiotemporal attention network includes two layers, wherein the first layer extracts spatiotemporal feature representations of the input values, and the second layer extracts spatiotemporal feature representations again from the spatiotemporal feature representations as the output values ​​of the spatiotemporal attention network.

[0083] The apparatus provided in this embodiment can execute the method provided in the above method embodiment. For details, please refer to the description in the method embodiment, which will not be repeated here.

[0084] The device provided in this embodiment acquires magnetic field values ​​from a magnetometer and interference field observations from other sensors at multiple historical moments within an aircraft in flight. Using the magnetic field values ​​and interference field observations as input values, a graph neural network model is used to extract the spatiotemporal feature representation of the input values. Based on this spatiotemporal feature representation, the magnetic field compensation value for the current moment is predicted. The graph neural network model is trained based on measurements from multiple sensors and a true reference value of the magnetic field after interference elimination. These multiple sensors include a magnetometer and other sensors. The magnetic field compensation value predicted through this process is more accurate than the acquired magnetic field value because it eliminates the influence of interfering magnetic fields, thus improving compensation precision. Prediction based on a graph neural network model enhances generalization and robustness, resulting in stronger stability. It eliminates the need for a tail boom magnetometer, is not hardware-dependent, and reduces cost and complexity.

[0085] See Figure 6This invention also provides a multi-source sensor aeromagnetic compensation system, which automates the entire process of aeromagnetic interference compensation model from offline training to online deployment. The system includes a data acquisition layer, a model building layer, and an online inference layer. The data acquisition layer includes a data acquisition module and a data storage and management module. The data acquisition module acquires raw data from scalar / vector magnetometers, multi-source interference sensors, and auxiliary information sources. The data storage and management module is connected to the data acquisition module and is used to cache, format, store, and manage the data, providing historical and real-time data streams for model training and online inference, respectively.

[0086] The model building layer receives raw training data from the data acquisition layer and sequentially performs feature extraction and model building through the data preprocessing module, graph construction module, and neural network construction module. Finally, the model training and optimization module completes the supervised training and parameter optimization of the model. The trained model is then output via the model deployment interface.

[0087] The core of the online inference layer is the onboard computing unit, which receives real-time sensor data streams and a deployed graph neural network model. Internally, the onboard computing unit sequentially performs preprocessing and graph construction, neural network inference, and finally provides high-precision magnetic interference compensation signals to the online compensation application through an output interface, completing the real-time compensation task.

[0088] See Figure 7 A comparison of magnetic field values ​​before and after aeromagnetic compensation using the above method was conducted. It can be seen that the method, by explicitly modeling spatial topological relationships and differentially modeling heterogeneous features, can more accurately capture the correlation between sensors, thereby improving the overall accuracy of magnetic compensation. Utilizing graph structures and differential processing methods, the GraphSensor network model can still operate stably under different flight conditions and sensor combinations, improving its adaptability to new environments and tasks. Through graph structure modeling and spatiotemporal dependency modeling, the excessive impact of single-point anomalies on overall prediction is avoided, ensuring that the compensation results remain stable even when a single sensor exhibits noise or anomalies. Compensation is achieved entirely through algorithmic modeling, rather than relying on new hardware, avoiding platform hardware modifications and reducing the cost and difficulty of system upgrades and deployments.

[0089] In summary, this invention utilizes a graph neural network structure incorporating spatiotemporal graph attention to perform spatiotemporal joint modeling of the graph topology constructed by the magnetometer and multi-source interference sensors. This explicitly characterizes the spatial topological relationships and temporal dependencies between sensors, enabling compensation for magnetic interference during flight. It overcomes the limitation of traditional CNNs and RNNs in explicitly expressing unstructured multi-sensor relationships, allowing the compensation model to more realistically reflect the interference propagation mechanism, thereby improving magnetic compensation accuracy and robustness under complex flight conditions. The graph topology structure is constructed based on correlation and spatial proximity constraints. Using the Pearson correlation coefficient and the spatial positional relationship of the sensors, the edges connecting the magnetometer and interference source nodes are selected. Distance attenuation and correlation fusion are used to set edge weights, constructing an adjacency matrix. This introduces graph neural networks and their topology modeling mechanisms into the field of magnetic compensation, enabling explicit representation and learning of the spatial dependencies and temporal dynamics between sensors.

[0090] Figure 8 This is a schematic diagram of the hardware structure of an electronic device to implement various embodiments of the present invention. The electronic device 800 includes, but is not limited to: a radio frequency unit 801, a network module 802, an audio output unit 803, an input unit 804, a sensor 805, a display unit 806, a user input unit 807, an interface unit 808, a memory 809, a processor 810, and a power supply 811, etc. Those skilled in the art will understand that... Figure 8 The electronic device structure shown does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements. In embodiments of the present invention, the electronic device includes, but is not limited to, mobile phones, tablet computers, laptops, PDAs, vehicle terminals, wearable devices, and pedometers. The processor 810 is used to execute the steps of the multi-source sensor aeromagnetic compensation method provided in the above embodiments.

[0091] It should be understood that, in this embodiment of the invention, the radio frequency unit 801 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink data from the base station and processes it with the processor 810; additionally, it transmits uplink data to the base station. Typically, the radio frequency unit 801 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier, a duplexer, etc. Furthermore, the radio frequency unit 801 can also communicate with networks and other electronic devices through a wireless communication system.

[0092] Electronic devices provide users with wireless broadband internet access through network module 802, such as helping users send and receive emails, browse web pages, and access streaming media.

[0093] The audio output unit 803 can convert audio data received by the radio frequency unit 801 or the network module 802 or stored in the memory 809 into audio signals and output them as sound. Furthermore, the audio output unit 803 can also provide audio output related to specific functions performed by the electronic device 800 (e.g., call signal reception sound, message reception sound, etc.). The audio output unit 803 includes a speaker, a buzzer, and a receiver, etc.

[0094] Input unit 804 is used to receive audio or video signals. Input unit 804 may include a graphics processing unit (GPU) 8041 and a microphone 8042. The GPU 8041 processes image data of still images or videos acquired by an image capture device (such as a camera) in video capture mode or image capture mode. The processed image frames can be displayed on display unit 806. The image frames processed by GPU 8041 can be stored in memory 809 (or other storage media) or transmitted via radio frequency unit 801 or network module 802. Microphone 8042 can receive sound and process such sound into audio data. The processed audio data can be converted into a format that can be transmitted to a mobile communication base station via radio frequency unit 801 in telephone call mode.

[0095] The electronic device 800 also includes at least one sensor 805, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor includes an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display panel 8061 according to the ambient light level, and the proximity sensor can turn off the display panel 8061 and / or backlight when the electronic device 800 is moved to the ear. As a type of motion sensor, an accelerometer sensor can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity and can be used to identify the posture of the electronic device (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc. The sensor 805 may also include a fingerprint sensor, pressure sensor, iris sensor, molecular sensor, gyroscope, barometer, hygrometer, thermometer, infrared sensor, etc., which will not be described in detail here.

[0096] The display unit 806 is used to display information input by the user or information provided to the user. The display unit 806 may include a display panel 8061, which may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), or the like.

[0097] User input unit 807 can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of electronic devices. Specifically, user input unit 807 includes a touch panel 8071 and other input devices 8072. Touch panel 8071, also known as a touch screen, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near touch panel 8071). Touch panel 8071 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to processor 810, which receives and executes commands from processor 810. In addition, touch panel 8071 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. Besides touch panel 8071, user input unit 807 may also include other input devices 8072. Specifically, other input devices 8072 may include, but are not limited to, physical keyboards, function keys (such as volume control buttons, power buttons, etc.), trackballs, mice, joysticks, etc., which will not be described in detail here.

[0098] Furthermore, the touch panel 8071 can cover the display panel 8061. When the touch panel 8071 detects a touch operation on or near it, it transmits the information to the processor 810 to determine the type of touch event. Subsequently, the processor 810 provides corresponding visual output on the display panel 8061 based on the type of touch event. Although in Figure 8 In this embodiment, the touch panel 8071 and the display panel 8061 are two independent components to realize the input and output functions of the electronic device. However, in some embodiments, the touch panel 8071 and the display panel 8061 can be integrated to realize the input and output functions of the electronic device. The specific implementation is not limited here.

[0099] Interface unit 808 serves as an interface for connecting external devices to electronic device 800. For example, external devices may include a wired or wireless headphone port, an external power supply (or battery charger) port, a wired or wireless data port, a memory card port, a port for connecting a device with an identification module, an audio input / output (I / O) port, a video I / O port, a headphone port, and so on. Interface unit 808 can be used to receive input from external devices (e.g., data, power, etc.) and transmit the received input to one or more components within electronic device 800, or it can be used to transmit data between electronic device 800 and external devices.

[0100] The memory 809 can be used to store software programs and various data. The memory 809 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback, image playback, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory 809 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0101] The processor 810 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 809, and by calling data stored in the memory 809, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. The processor 810 may include one or more processing units; preferably, the processor 810 may integrate an application processor and a modem processor. The application processor mainly handles the operating system, user interface, and applications, while the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 810.

[0102] The electronic device 800 may also include a power supply 811 (such as a battery) that supplies power to various components. Preferably, the power supply 811 can be logically connected to the processor 810 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system.

[0103] Preferably, the present invention also provides an electronic device, including a processor 810, a memory 809, and a computer program stored in the memory 809 and executable on the processor 810. When the computer program is executed by the processor 810, it implements the various processes of the above method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0104] This invention also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here.

[0105] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the above-described method embodiments and achieves the same technical effects. To avoid repetition, further details are omitted here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0106] The computer-readable storage medium provided in this embodiment of the invention acquires magnetic field values ​​from a magnetometer and interference field observations from other sensors at multiple historical moments within an aircraft in flight. Using the magnetic field values ​​and interference field observations as input values, a graph neural network model is used to extract spatiotemporal feature representations of the input values. Based on these spatiotemporal feature representations, the magnetic field compensation value for the current moment is predicted. The graph neural network model is trained based on measurements from multiple sources and a true reference value of the magnetic field after interference elimination. These multiple sources include a magnetometer and other sensors. The magnetic field compensation value predicted through this process is more accurate than the acquired magnetic field value because it eliminates the influence of interfering magnetic fields, thus improving compensation precision. Prediction based on a graph neural network model enhances generalization and robustness, resulting in stronger stability. It eliminates the need for a tail boom magnetometer, is not hardware-dependent, and reduces cost and complexity.

[0107] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0109] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0111] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0112] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0113] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0114] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0115] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A multi-source sensor aeromagnetic compensation method, characterized in that, The method comprises the following steps: In an aircraft in flight, magnetic field values of a magnetometer at multiple historical time points and interference field observation values of other sensors are acquired; The magnetic field values and the interference field observation values are used as input values, a graph neural network model is used to extract a spatio-temporal feature representation of the input values, and a magnetic field compensation value at a current time point is predicted according to the spatio-temporal feature representation. The graph neural network model is trained based on measurement values of multiple source sensors and reference true values of magnetic fields after interference elimination, and the multiple source sensors comprise the magnetometer and the other sensors.

2. The method of claim 1, wherein, The prediction of the magnetic field compensation value at the current time point according to the spatio-temporal feature representation comprises the following steps: The graph neural network model uses a long short-term memory network to calculate the spatio-temporal feature representation, and outputs the magnetic field compensation value at the current time point after layer normalization processing and full connection layer processing.

3. The method of claim 1, wherein, The extraction of the spatio-temporal feature representation of the input values by using the graph neural network model comprises the following steps: The graph neural network model comprises a spatio-temporal attention network, and the spatio-temporal attention network comprises a spatial attention network and a temporal attention network; The spatial attention network is used to extract spatial enhanced features of the input values; The temporal attention network is used to generate the spatio-temporal feature representation based on the spatial enhanced features.

4. The method of claim 3, wherein, The spatio-temporal attention network comprises magnetometer nodes and interference source nodes, and a connection relationship of nodes in the spatio-temporal attention network comprises the following steps: The magnetometer nodes are connected by edges; The magnetometer nodes and the interference source nodes are selectively connected by edges based on statistical correlation or spatial proximity; The interference source nodes are not connected by edges.

5. The method of claim 4, wherein, The selective connection of the magnetometer nodes and the interference source nodes based on the statistical correlation comprises the following steps: A Pearson correlation coefficient between the magnetic field values of the magnetometer nodes and the interference field observation values of the interference source nodes is calculated; In a case where the Pearson correlation coefficient is greater than or equal to a correlation threshold, the magnetometer nodes and the interference source nodes are connected by edges; In a case where the Pearson correlation coefficient is less than the correlation threshold, the magnetometer nodes and the interference source nodes are not connected by edges.

6. The method of claim 4, wherein, The selective connection of the magnetometer nodes and the interference source nodes based on the spatial proximity comprises the following steps: A distance between the magnetometer nodes and the interference source nodes is calculated; In a case where the distance is less than a distance threshold, the magnetometer nodes and the interference source nodes are connected by edges; In a case where the distance is greater than or equal to the distance threshold, the magnetometer nodes and the interference source nodes are not connected by edges.

7. The method of claim 4, wherein, The connection relationship of the nodes in the spatio-temporal attention network is defined by an adjacency matrix, and a calculation formula of the adjacency matrix is as follows: ； ； ； where A is an initial adjacency matrix, is an adjacency matrix between magnetometer nodes and magnetometer nodes, is an adjacency matrix between magnetometer nodes and interference source nodes, is is a transpose matrix of is an adjacency matrix between interference source nodes and interference source nodes, is an intermediate adjacency matrix, is an identity matrix, is an adjacency matrix, is a diagonal matrix.

8. The method of claim 4, wherein, The edges in the spatio-temporal attention network also have weights, and a calculation formula of the weights is as follows: ； ； wherein, node i and node j are any two nodes in the spatio-temporal attention network, is a distance kernel for the node i and node j, is a correlation kernel for the node i and node j, is the Euclidean distance of the node i and node j in the body coordinate system, is a distance scale, is the Pearson correlation coefficient of the node i and node j, p is a power operation coefficient, p = 1 or p = 2, is the weight of the edge between the node i and node j, is a weighting coefficient.

9. The method of claim 3, wherein, The spatio-temporal attention network comprises two layers, wherein a first layer extracts the spatio-temporal feature representation of the input values, and a second layer extracts a spatio-temporal feature representation of the spatio-temporal feature representation again, as an output value of the spatio-temporal attention network.

10. A multi-source sensor aeromagnetic compensation device, characterized in that, The method comprises the following steps: An acquisition module is configured to acquire, in an aircraft in flight, magnetic field values of a magnetometer at multiple historical time points and interference field observation values of other sensors; A prediction module is configured to use a graph neural network model to extract a spatio-temporal feature representation of input values including the magnetic field value and the disturbance field observation value, and predict a magnetic field compensation value at a current time according to the spatio-temporal feature representation. The graph neural network model is trained based on measurement values of multiple source sensors and reference true values of the magnetic field after interference is eliminated, and the multiple source sensors include a magnetometer and other sensors.

Images