A quantum-enhanced navigation data processing method and system based on physical information neural network
By constructing a quantum-enhanced navigation system based on physical information neural networks and utilizing data processing methods from cold atom interferometric quantum magnetometers and inertial measurement units, the problem of navigation accuracy degradation in dynamic environments was solved, achieving high-precision, long-endurance autonomous navigation and magnetic interference suppression.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-09
- Publication Date
- 2026-07-03
AI Technical Summary
In long-endurance missions, the positioning error of inertial navigation systems accumulates over time. Quantum sensors on dynamic carriers are susceptible to interference from metal structures and electronic devices, leading to a decrease in navigation accuracy. Traditional filtering methods are difficult to separate weak geomagnetic signals.
A quantum-enhanced navigation system based on physical information neural networks is constructed. Data is collected using a cold atom interferometric quantum magnetometer and an inertial measurement unit. By combining a deep neural network model with physical constraints, geomagnetic signals are separated and multi-source sensor fusion filtering is performed to achieve high-precision navigation.
Achieve high-precision, long-endurance autonomous navigation in satellite navigation denied environments, effectively suppress platform interference, and reduce navigation drift rate to within 100 meters per hour.
Smart Images
Figure CN122329285A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of navigation and positioning technology, and more specifically to a quantum-enhanced navigation data processing method and system based on a physical information neural network. Background Technology
[0002] In modern navigation applications, inertial navigation systems (INS) are widely used due to their autonomy; however, their positioning errors accumulate over time, failing to meet the requirements of long-endurance missions. Quantum sensing technology, particularly cold atom interferometric quantum magnetometers, can provide extremely sensitive magnetic field measurements and, theoretically, can eliminate drift in INS through geomagnetic matching. However, practical applications face significant challenges: the metallic structure and electronic equipment of the carrier (such as underwater vehicles and underground drilling equipment) generate complex dynamic magnetic interference, and quantum sensors are sensitive to vibration noise. Traditional filtering methods struggle to effectively separate weak geomagnetic signals under strong interference, leading to a sharp decline in navigation accuracy in dynamic environments.
[0003] Therefore, proposing a quantum-enhanced navigation data processing method and system based on physical information neural networks to solve the difficulties existing in the prior art is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0004] In view of this, the present invention provides a quantum-enhanced navigation data processing method and system based on physical information neural networks. It solves the technical problem of quantum sensors being susceptible to interference and drift divergence on dynamic carriers through a mathematical model of physical constraints, and realizes high-precision, long-endurance autonomous navigation and magnetic interference suppression and multi-source navigation fusion in satellite navigation denial environments.
[0005] To achieve the above objectives, the present invention provides the following technical solution: A quantum-enhanced navigation data processing method based on physical information neural networks includes the following steps: S1. Construct a hybrid sensor assembly, use a cold atom interferometric quantum magnetometer to collect raw observations of the ambient magnetic field strength, and use an inertial measurement unit to collect triaxial angular velocity and triaxial specific force data of the carrier; S2. Construct a physics-assisted artificial intelligence engine, which includes a deep neural network model. The training loss function of the deep neural network model contains physical law constraints. S3. Input the original observations, triaxial angular velocity and triaxial force data into the physics-assisted artificial intelligence engine, and output the reconstructed magnetic field data. S4. Match the reconstructed magnetic field data with the pre-stored geomagnetic reference map to obtain the absolute position observation value of the carrier. S5. Establish a multi-source sensor fusion filter to fuse and estimate the absolute position observations and navigation state variables calculated by inertial navigation, and output the corrected carrier position, velocity and attitude information.
[0006] Optionally, the specific details of constructing a hybrid sensor assembly in S1 to acquire raw observations of the environmental magnetic field strength using a cold atom interferometric quantum magnetometer are as follows: Suppose that the cold atom quantum magnetometer is at time 1000. Output magnetic field measurement value for: (1) in, The actual geomagnetic field vector to be extracted depends on the carrier location. ; For platform-induced magnetic interference; This refers to the sensor's noise floor.
[0007] Optionally, the physical law constraints included in the training loss function of the deep neural network model in S2 include the magnetic field divergence-free constraint and the platform magnetic disturbance model constraint. Deep neural network models are used to separate geomagnetic signal components that conform to Maxwell's equations from noisy data.
[0008] Optionally, the core of the physics-assisted AI engine built in S2 is a deep neural network model. The optimization objective is to minimize the total loss function. : (2) in, This is the data fitting error term. For physical constraints, For kinematic constraints, and These are the weighting coefficients; Physical constraints Based on Gauss's law for magnetic fields in Maxwell's equations, the reconstructed geomagnetic field is required. Satisfy the divergence-free condition: (3) After obtaining clean magnetic field data, geomagnetic map matching is performed to determine the observation location. And combined with the state prediction of the inertial navigation system The system state is updated using the optimal estimation criterion: (4) in, For Kalman gain, For the observation matrix, and These represent the states before and after the update, respectively.
[0009] Optionally, in S3, the carrier attitude information provided by the inertial measurement unit is used to fit the Tolles-Lawson magnetic interference model through a neural network to compensate for the induced magnetic field and eddy current magnetic field generated by the carrier's maneuver in real time.
[0010] Optionally, the multi-source sensor fusion filter in S5 adopts an error state extended Kalman filter, and the state vector includes position error, velocity error, attitude error, gyroscope zero bias and accelerometer zero bias.
[0011] A quantum-enhanced navigation data processing system based on a physical information neural network, applying any one of the above-mentioned quantum-enhanced navigation data processing methods based on a physical information neural network, includes: a data acquisition module, a physical auxiliary processing module, and a navigation solution module; The data acquisition module, connected to the input of the physical auxiliary processing module, is used to construct a hybrid sensor assembly. It uses a cold atom interferometric quantum magnetometer to acquire raw observations of the environmental magnetic field strength and an inertial measurement unit to acquire triaxial angular velocity and triaxial specific force data of the carrier. The physical auxiliary processing module is connected to the input end of the navigation calculation module. It is used to build a physical auxiliary artificial intelligence engine. The original observation values, three-axis angular velocity and three-axis specific force data are input into the physical auxiliary artificial intelligence engine, and the reconstructed magnetic field data is output. The reconstructed magnetic field data is matched with the pre-stored geomagnetic reference map to obtain the absolute position observation value of the carrier. The navigation calculation module, connected to the output of the physical auxiliary processing module, is used to establish a multi-source sensor fusion filter, fuse the absolute position observations with the navigation state variables calculated by inertial navigation, and output the corrected carrier position, velocity and attitude information.
[0012] As can be seen from the above technical solution, compared with the prior art, the present invention provides a quantum-enhanced navigation data processing method and system based on physical information neural networks, which has the following beneficial effects: (1) This invention solves the technical problem of quantum sensors being susceptible to interference and drifting on dynamic carriers through a mathematical model of physical constraints, and realizes high-precision, long-endurance autonomous navigation in satellite navigation denied environments; (2) The present invention can effectively suppress platform interference and reduce the navigation drift rate in the absence of GNSS environment to less than 100 meters / hour. Attached Figure Description
[0013] 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0014] Figure 1 A flowchart of a quantum-enhanced navigation data processing method based on a physical information neural network provided by the present invention; Figure 2 A schematic diagram of a quantum-enhanced navigation data processing system based on a physical information neural network provided by the present invention; Figure 3 The data processing flowchart of the physics-assisted artificial intelligence engine provided by this invention; Figure 4 The present invention provides a logic block diagram of a multi-source sensor fusion algorithm. Detailed Implementation
[0015] 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.
[0016] See Figure 1 As shown, this invention discloses a quantum-enhanced navigation data processing method based on a physical information neural network, comprising the following steps: S1. Construct a hybrid sensor assembly, use a cold atom interferometric quantum magnetometer to collect raw observations of the ambient magnetic field strength, and use an inertial measurement unit to collect triaxial angular velocity and triaxial specific force data of the carrier; S2. Construct a physics-assisted artificial intelligence engine, which includes a deep neural network model. The training loss function of the deep neural network model contains physical law constraints. S3. Input the original observations, triaxial angular velocity and triaxial force data into the physics-assisted artificial intelligence engine, and output the reconstructed magnetic field data. S4. Match the reconstructed magnetic field data with the pre-stored geomagnetic reference map to obtain the absolute position observation value of the carrier. S5. Establish a multi-source sensor fusion filter to fuse and estimate the absolute position observations and navigation state variables calculated by inertial navigation, and output the corrected carrier position, velocity and attitude information.
[0017] Furthermore, the specific details of constructing a hybrid sensor assembly in S1 and using a cold atom interferometric quantum magnetometer to collect raw observations of the environmental magnetic field strength are as follows: Suppose that the cold atom quantum magnetometer is at time 1000. Output magnetic field measurement value for: (1) in, The actual geomagnetic field vector to be extracted depends on the carrier location. ; For platform-induced magnetic interference; This refers to the sensor's noise floor.
[0018] Furthermore, the physical law constraints included in the training loss function of the deep neural network model in S2 include the magnetic field divergence-free constraint and the platform magnetic disturbance model constraint. Deep neural network models are used to separate geomagnetic signal components that conform to Maxwell's equations from noisy data.
[0019] Furthermore, the core of the physics-assisted artificial intelligence engine built in S2 is a deep neural network model. The optimization objective is to minimize the total loss function. : (2) in, This is the data fitting error term. For physical constraints, For kinematic constraints, and These are the weighting coefficients; Physical constraints Based on Gauss's law for magnetic fields in Maxwell's equations, the reconstructed geomagnetic field is required. Satisfy the divergence-free condition: (3) After obtaining clean magnetic field data, geomagnetic map matching is performed to determine the observation location. And combined with the state prediction of the inertial navigation system The system state is updated using the optimal estimation criterion: (4) in, For Kalman gain, For the observation matrix, and These represent the states before and after the update, respectively.
[0020] Furthermore, in S3, the carrier attitude information provided by the inertial measurement unit is used to fit the Tolles-Lawson magnetic interference model through a neural network, thereby compensating in real time for the induced magnetic field and eddy current magnetic field generated by the carrier's maneuver.
[0021] Furthermore, the multi-source sensor fusion filter in S5 adopts an error state extended Kalman filter, and the state vector includes position error, velocity error, attitude error, gyroscope zero bias and accelerometer zero bias.
[0022] Example 1: Sensor Error Model and State Space Definition
[0023] First, the data acquisition model of the system is defined. The system uses a cold atom interferometer as the main magnetometer and a MEMS IMU as the motion reference.
[0024] For MEMS gyroscopes, measuring angular velocity The model is as follows: (5) in, For true angular velocity, To achieve zero bias in the gyroscope, This represents angular random walk noise.
[0025] For MEMS accelerometers, measuring specific force The model is as follows: (6) in, Let n be the attitude rotation matrix from the vehicle coordinate system (b-frame) to the navigation coordinate system (n-frame). For the acceleration of motion in the navigation system, It is the gravitational acceleration vector. To achieve zero bias in the accelerometer, For velocity random walk noise; Therefore, a 15-dimensional error state vector for the inertial navigation system is defined. : (7) in, For positional error, For speed error, The angle represents the attitude error.
[0026] Example 2: Deep Processing of a Physics-Assisted AI Engine
[0027] like Figure 3 As shown in the figure, this embodiment details how the AI engine uses physical constraints to extract the real magnetic field from mixed data.
[0028] 1. Input data preprocessing
[0029] Input tensor of neural network Includes time window Internal sensor sequence: (8) 2. Physical Modeling of Platform Magnetic Interference To enable the AI network to identify interference, a Tolles-Lawson model constraint is introduced into the loss function to address the interference magnetic field generated by the platform. Decomposed into a constant field, an induced field, and an eddy current field: (9) Its mathematical expression is further expanded to include the carrier magnetic field. And the form related to its rate of change: (10) in, The magnetic field vector generated by a constant magnetic moment. The matrix of inductive magnetic coefficients is (3x3). The eddy current coefficient matrix is 3x3. This is the time derivative of the magnetic field; 3. Specification of the Physical Constraint Loss Function Loss function of physics-assisted AI engine Designed as: (11) in, To constrain consistency in the Tolles-Lawson model, the network is forced to predict interference terms. It conforms to formula (10). Physical structure: (12) Here , , These are the trainable parameters that the network needs to learn, corresponding to physical coefficients; To constrain the magnetic field divergence, the gradient of the network output field with respect to spatial coordinates is calculated using automatic differentiation: (13) By minimizing equation (11), the network can output a clean estimate of the geomagnetic field. .
[0030] Example 3: Geomagnetic Matching and Multi-Source Fusion Algorithm
[0031] like Figure 4 As shown, this embodiment describes the navigation solution process.
[0032] 1. Magnetic Map Matching
[0033] Using noise reduction In the pre-stored magnetic map The search for the optimal matching position is performed, and the objective function is constructed using the mean squared error (MSD) criterion. : (14) Solving the problem using optimization algorithms (such as particle swarm optimization or related matching) makes it possible to achieve the desired result. The minimum position correction is used to obtain the position observation value. ; 2. Extended Kalman Filter (EKF) Fusion An indirect Kalman filter is used. First, the error state transition equation is established: (15) in, The state transition matrix is composed of the following submatrices at its core: Position error transfer: (16) Velocity error transfer (considering specific force integral and attitude error): (17) Attitude error transfer (considering Earth's rotation) ): (18) in, Representing vectors antisymmetric matrix, This represents the discretization time step.
[0034] The observation equations are constructed as follows: (19) Observation matrix for: (20) Finally, the standard Kalman filter update steps are performed to calculate the error state estimate. And correct the inertial navigation system: (twenty one) Through the above closed-loop correction, the cumulative error of inertial navigation diverging over time is eliminated, and high-precision positioning with long endurance is achieved.
[0035] A quantum-enhanced navigation data processing system based on physical information neural networks, such as Figure 2 As shown, a quantum-enhanced navigation data processing method based on a physical information neural network, applying any of the above, includes: a data acquisition module, a physical auxiliary processing module, and a navigation solution module; The data acquisition module, connected to the input of the physical auxiliary processing module, is used to construct a hybrid sensor assembly. It uses a cold atom interferometric quantum magnetometer to acquire raw observations of the environmental magnetic field strength and an inertial measurement unit to acquire triaxial angular velocity and triaxial specific force data of the carrier. The physical auxiliary processing module is connected to the input end of the navigation calculation module. It is used to build a physical auxiliary artificial intelligence engine. The original observation values, three-axis angular velocity and three-axis specific force data are input into the physical auxiliary artificial intelligence engine, and the reconstructed magnetic field data is output. The reconstructed magnetic field data is matched with the pre-stored geomagnetic reference map to obtain the absolute position observation value of the carrier. The navigation calculation module, connected to the output of the physical auxiliary processing module, is used to establish a multi-source sensor fusion filter, fuse the absolute position observations with the navigation state variables calculated by inertial navigation, and output the corrected carrier position, velocity and attitude information.
[0036] Specifically, the inertial measurement unit is a MEMS inertial measurement unit.
[0037] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.
[0038] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A quantum-enhanced navigation data processing method based on a physical information neural network, characterized in that, Includes the following steps: S1. Construct a hybrid sensor assembly, use a cold atom interferometric quantum magnetometer to collect raw observations of the ambient magnetic field strength, and use an inertial measurement unit to collect triaxial angular velocity and triaxial specific force data of the carrier; S2. Construct a physics-assisted artificial intelligence engine, which includes a deep neural network model. The training loss function of the deep neural network model contains physical law constraints. S3. Input the original observations, triaxial angular velocity and triaxial force data into the physics-assisted artificial intelligence engine, and output the reconstructed magnetic field data. S4. Match the reconstructed magnetic field data with the pre-stored geomagnetic reference map to obtain the absolute position observation value of the carrier. S5. Establish a multi-source sensor fusion filter to fuse and estimate the absolute position observations and navigation state variables calculated by inertial navigation, and output the corrected carrier position, velocity and attitude information.
2. The quantum-enhanced navigation data processing method based on a physical information neural network according to claim 1, characterized in that, The specific details of constructing a hybrid sensor assembly in S1, and using a cold atom interferometric quantum magnetometer to collect raw observations of the environmental magnetic field strength, are as follows: Suppose that the cold atom quantum magnetometer is at time 1000. Output magnetic field measurement value for: （1） in, The actual geomagnetic field vector to be extracted depends on the carrier location. ; For platform-induced magnetic interference; This refers to the sensor's noise floor.
3. The quantum-enhanced navigation data processing method based on a physical information neural network according to claim 1, characterized in that, The physical law constraints included in the training loss function of the deep neural network model in S2 include the magnetic field divergence-free constraint and the platform magnetic disturbance model constraint. Deep neural network models are used to separate geomagnetic signal components that conform to Maxwell's equations from noisy data.
4. A quantum-enhanced navigation data processing method based on a physical information neural network according to claim 1 or 3, characterized in that, The core of the physics-assisted AI engine built in S2 is a deep neural network model. The optimization objective is to minimize the total loss function. : （2） in, This is the data fitting error term. For physical constraints, For kinematic constraints, and These are the weighting coefficients; Physical constraints Based on Gauss's law for magnetic fields in Maxwell's equations, the reconstructed geomagnetic field is required. Satisfy the divergence-free condition: （3） After obtaining clean magnetic field data, geomagnetic map matching is performed to determine the observation location. And combined with the state prediction of the inertial navigation system The system state is updated using the optimal estimation criterion: （4） in, For Kalman gain, For the observation matrix, and These represent the states before and after the update, respectively.
5. The quantum-enhanced navigation data processing method based on a physical information neural network according to claim 1, characterized in that, In S3, the carrier attitude information provided by the inertial measurement unit is used to fit the Tolles-Lawson magnetic interference model through a neural network, thereby compensating in real time for the induced magnetic field and eddy current magnetic field generated by the carrier's maneuver.
6. The quantum-enhanced navigation data processing method based on a physical information neural network according to claim 1, characterized in that, The multi-source sensor fusion filter in S5 adopts an error state extended Kalman filter, and the state vector includes position error, velocity error, attitude error, gyroscope zero bias, and accelerometer zero bias.
7. A quantum-enhanced navigation data processing system based on a physical information neural network, characterized in that, The quantum-enhanced navigation data processing method based on a physical information neural network according to any one of claims 1-6 includes: a data acquisition module, a physical auxiliary processing module, and a navigation solution module; The data acquisition module, connected to the input of the physical auxiliary processing module, is used to construct a hybrid sensor assembly. It uses a cold atom interferometric quantum magnetometer to acquire raw observations of the environmental magnetic field strength and an inertial measurement unit to acquire triaxial angular velocity and triaxial specific force data of the carrier. The physical auxiliary processing module is connected to the input end of the navigation calculation module. It is used to build a physical auxiliary artificial intelligence engine. The original observation values, three-axis angular velocity and three-axis specific force data are input into the physical auxiliary artificial intelligence engine, and the reconstructed magnetic field data is output. The reconstructed magnetic field data is matched with the pre-stored geomagnetic reference map to obtain the absolute position observation value of the carrier. The navigation calculation module, connected to the output of the physical auxiliary processing module, is used to establish a multi-source sensor fusion filter, fuse the absolute position observations with the navigation state variables calculated by inertial navigation, and output the corrected carrier position, velocity and attitude information.