An aviation intelligent positioning system based on inertial navigation

Abstract

The application discloses an aviation intelligent positioning system based on inertial navigation, comprising: a data acquisition module for acquiring inertial navigation information and constructing a flight state sequence; a state evolution module for performing state evolution processing on the flight state sequence, and constructing a flight behavior sequence; an inertial propagation module for performing drift propagation analysis on the flight behavior sequence, and obtaining an inertial drift sequence; a coupling deduction module for performing state-drift coupling deduction processing, and constructing a position deduction sequence; a convergence correction module for performing trajectory correction processing on the position deduction sequence, and forming a positioning trajectory sequence; and a positioning mapping module for forming an aircraft real-time positioning result. The application utilizes a multi-scale gated Mamba model and the like, and has the advantages of high positioning continuity, strong drift suppression capability, high trajectory stability and strong adaptability to complex flight environments.

Description

Technical Field

[0001] This invention relates to the field of intelligent positioning technology, and in particular to an aviation intelligent positioning system based on inertial navigation. Background Technology

[0002] With the rapid development of drones, autonomous aircraft, and intelligent avionics, inertial navigation technology has been widely applied in aviation positioning, flight control, and trajectory prediction scenarios. Existing inertial navigation methods typically acquire acceleration and angular velocity data through inertial measurement units (IMUs) and combine them with attitude and heading angle data to continuously predict the aircraft's current position.

[0003] In existing technologies, most inertial navigation methods primarily employ Kalman filtering, particle filtering, and trajectory smoothing to compensate for inertial drift errors, thereby improving positioning accuracy and navigation stability. Some solutions also combine satellite navigation information with visual measurement information to perform joint optimization processing on the positioning results.

[0004] However, existing technologies lack the ability to dynamically model the continuous evolution of flight states, making it difficult to effectively analyze the state propagation relationships at different time scales. This leads to the continuous accumulation of inertial drift errors during long-term flight. Furthermore, existing technologies lack the ability to jointly extrapolate between flight states and inertial drift, easily causing positioning trajectory deviations and decreased trajectory continuity. In addition, existing technologies lack the ability to continuously converge and dynamically correct trajectory topology relationships, making it difficult to meet the application requirements for highly continuous intelligent positioning in complex aviation scenarios.

[0005] Therefore, how to provide an intelligent aviation positioning system based on inertial navigation is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent aviation positioning system based on inertial navigation. This invention fully utilizes the multi-scale gated Mamba model, the state feedback RBPF algorithm, factor graph global optimization, and the liquid time constant network to perform continuous state evolution analysis, inertial drift propagation deduction, state-drift coupling calculation, and trajectory topology convergence correction on the aircraft's flight state, thereby achieving intelligent positioning of the aircraft's real-time spatial position. It has the advantages of high positioning continuity, strong drift suppression capability, high trajectory stability, and strong adaptability to complex flight environments.

[0007] An aviation intelligent positioning system based on inertial navigation according to an embodiment of the present invention includes: The data acquisition module is used to acquire and preprocess the inertial navigation information of the aircraft during flight, and to construct a flight state sequence. The state evolution module is used to construct a state evolution matrix based on the flight state sequence. It performs bidirectional state association encoding and continuous state evolution fitting operations on the state evolution matrix through a multi-scale gated Mamba model to generate state evolution features and construct flight behavior sequences. The inertial propagation module is used to perform inertial drift propagation analysis on the flight behavior sequence, construct the drift evolution trajectory, perform state estimation operation on the drift evolution trajectory using the state feedback RBPF algorithm, and perform global optimization of the factor graph to obtain the inertial drift sequence. The coupled extrapolation module is used to perform state-drift coupled extrapolation operations on the flight state sequence based on the inertial drift sequence and through the multi-scale gated Mamba model to construct the position extrapolation sequence; The convergence correction module is used to perform trajectory topology convergence operation on the continuous position extrapolation sequence and perform continuous trajectory correction to form a positioning trajectory sequence; The positioning mapping module is used to perform real-time spatial position mapping processing on the positioning trajectory sequence to generate real-time positioning results for the aircraft.

[0008] Optionally, the inertial navigation information represents acceleration data, angular velocity data, attitude angle data, and heading angle data collected by intelligent sensors installed inside the aircraft. The preprocessing includes time synchronization, outlier removal, noise filtering, data normalization, and unified encoding.

[0009] Optionally, the state evolution module includes: A state evolution matrix is ​​constructed based on the flight state sequence, and continuous state association mapping is performed on the state evolution matrix to generate state propagation paths; Introducing a multi-scale gated Mamba model; Compared with the traditional Mamba model, the multi-scale gated Mamba model adds bidirectional propagation units, continuous evolution units, and drift coupling units; In the bidirectional propagation unit, bidirectional state association encoding is performed on the state propagation path to extract state propagation features at different time scales and construct a bidirectional state association sequence. The bidirectional state association sequence is input into the continuous evolution unit, and continuous state evolution fitting processing is performed to generate state evolution features. Correlation analysis is performed on the evolution characteristics of each state to extract the corresponding attitude evolution state, trajectory change state and motion change state, and to construct a flight behavior sequence.

[0010] Optionally, the calculation process in the bidirectional propagation unit specifically includes: A bidirectional diffusion algorithm based on time inversion mechanism is adopted to perform reverse time backtracking diffusion and forward time recursion diffusion processing on the state propagation path, and calculate the historical state influence value and future state influence value of each flight state. Based on the historical state influence value and the future state influence value, bidirectional state association coding is performed on the flight state sequence to calculate the state association strength, state propagation distance and state transition continuity between each flight state, and combine them to obtain the state association features. Dynamic gating contention handling is performed on the associated features of each state: Based on the correlation characteristics of each state, the gating competition weight value corresponding to each flight state is calculated, and the competitive ranking process is performed on each flight state based on the gating competition weight value. Based on the competition ranking results, propagation suppression processing is performed on flight states with gating competition weights lower than the preset competition threshold, and priority propagation processing is performed on flight states with gating competition weights higher than the preset competition threshold, in order to identify abrupt and stable states and construct a state label sequence. Based on the state label sequence, the state propagation path is divided into multiple time scales to construct propagation intervals at multiple time scales. In each propagation interval, the state propagation direction and state association continuity are extracted to generate state propagation features at each time scale. Fusion encoding is performed on the state propagation features at different time scales to construct a bidirectional state association sequence.

[0011] Optionally, the computation process in the continuous evolution unit specifically includes: The bidirectional state association sequence is input into the continuous evolution unit, and the continuous state evolution field is constructed based on the neural ordinary differential equation. In a continuous state evolution field, a continuous state flow expansion process is performed on the bidirectional state association sequence to construct a state evolution manifold; Based on the state evolution manifold, continuous state trajectory tracking is performed on each flight state to calculate the state evolution velocity, state evolution direction and state evolution curvature corresponding to each flight state, forming a convergence parameter set; Based on the convergence parameter set, state convergence analysis is performed on the state evolution manifold to identify abrupt state change regions and stable state regions, and to construct the state evolution trajectory. Continuous state prediction processing is performed on the state evolution trajectory to generate state evolution features and construct flight behavior sequences.

[0012] Optionally, the inertial propagation module includes: Based on the flight behavior sequence, the drift propagation relationship between each flight state is extracted, and a drift propagation graph is constructed; Perform continuous drift-diffusion analysis on the drift propagation graph to construct the drift evolution trajectory; The state feedback RBPF algorithm is used to perform particle state prediction and feedback update processing on the drift evolution trajectory to construct the drift state sequence; A drift factor graph is constructed based on the drift state sequence, and global association constraint processing is performed on each node in the drift factor graph. The state association error and state propagation deviation between each node are calculated to obtain the constraint factor graph. Based on the constraint factor diagram, a global convergence optimization process is performed on the drift state sequence to generate an inertial drift sequence.

[0013] Optionally, the particle state prediction and feedback update process specifically includes: Multiple drift state particles are constructed based on the drift evolution trajectory, and the propagation weights of each drift state particle are initialized. Perform inertial propagation prediction processing on each drifting particle to generate predicted particle states; Based on the flight state sequence, state matching processing is performed on each predicted particle state, and the state deviation vector of each predicted particle state is calculated. Based on the state deviation vector, feedback weight update processing is performed on the state of each predicted particle to adjust the corresponding propagation weights; Based on the adjusted propagation weights, the states of each predicted particle are corrected by feedback to construct a drift state sequence; Remove drifting state particles whose propagation weight is less than a preset weight threshold; Based on the retained drift state particles, continuous convergence update processing is performed on the drift evolution trajectory to generate an inertial drift sequence.

[0014] Optionally, the coupled deduction module includes: A state-drift coupling matrix is ​​constructed based on the inertial drift sequence and the flight state sequence, and a continuous correlation mapping process is performed on the state-drift coupling matrix to obtain the coupling propagation path; In the drift coupling unit of the multi-scale gated Mamba model, bidirectional coupling propagation processing is performed on the coupling propagation path to construct a coupling state sequence; Based on the coupled state sequence, state-drift correlation analysis is performed on each flight state to calculate the coupling parameter vector. Then, based on the coupling parameter vector, continuous state deduction is performed on the coupled state sequence to construct the coupled deduction trajectory. Multi-timescale state propagation and continuous position fitting are performed on the coupled inference trajectory to generate a position inference sequence.

[0015] Optionally, the continuous state deduction process specifically includes: Based on the coupling parameter vector, a continuous state mapping process is performed on the coupled state sequence to generate a state evolution sequence; By using a liquid time constant network, dynamic time constant adjustment is performed on each flight state according to the state evolution sequence to construct a state recursion sequence; Based on the state recursive sequence, continuous drift feedback processing is performed on each flight state to calculate the drift propagation deviation and state residual corresponding to each flight state, and a drift feedback sequence is constructed. Based on the drift feedback sequence, dynamic state convergence processing is performed on the state recursion sequence to generate a coupled inference trajectory.

[0016] Optionally, the convergence correction module includes: Extract the trajectory connectivity in the location extrapolation sequence, construct the trajectory topology map, identify trajectory anomaly regions, and generate trajectory anomaly sequences; Based on the trajectory anomaly sequence, perform topological convergence propagation operation along the trajectory topology graph, calculate the topological convergence direction and topological convergence strength corresponding to each inferred node, and construct the topological convergence sequence. Perform continuous neighborhood association analysis on the topological convergence sequence, and perform dynamic trajectory convergence correction on the position extrapolation sequence based on the results of the continuous neighborhood association analysis to generate a trajectory convergence sequence; Multi-timescale continuous trajectory fitting and trajectory consistency correction operations are performed on the trajectory convergence sequence to form a positioning trajectory sequence.

[0017] The beneficial effects of this invention are: First, this invention performs bidirectional state association coding and continuous state evolution analysis on the flight state through a multi-scale gated Mamba model, which can improve the continuous modeling capability of aircraft flight state and enhance the positioning continuity and state recognition stability in complex flight environments.

[0018] Secondly, this invention uses the state feedback RBPF algorithm and factor graph global optimization technology to perform dynamic feedback and global convergence processing on the inertial drift propagation process, which can reduce the cumulative error of inertial drift during long-term flight and improve the aircraft's positioning accuracy and drift suppression capability.

[0019] Finally, by performing coupled deduction of flight state and inertial drift, and combining trajectory topology convergence and continuous trajectory correction processing, this invention can improve the continuity of positioning trajectory and spatial mapping stability, reduce trajectory jump problems in complex flight scenarios, and thus improve the reliability of aviation intelligent positioning results. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1This is a block diagram of an intelligent aviation positioning system based on inertial navigation proposed in this invention. Figure 2 This is a flowchart illustrating the flight state evolution and behavior recognition of an aviation intelligent positioning system based on inertial navigation proposed in this invention. Figure 3 This is a flowchart illustrating the inertial drift propagation and state coupling derivation of an aviation intelligent positioning system based on inertial navigation proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-3 An intelligent positioning system for aviation based on inertial navigation, comprising: The data acquisition module is used to acquire and preprocess the inertial navigation information of the aircraft during flight, and to construct a flight state sequence. The state evolution module is used to construct a state evolution matrix based on the flight state sequence. It performs bidirectional state association encoding and continuous state evolution fitting operations on the state evolution matrix through a multi-scale gated Mamba model to generate state evolution features and construct flight behavior sequences. The inertial propagation module is used to perform inertial drift propagation analysis on the flight behavior sequence, construct the drift evolution trajectory, perform state estimation operation on the drift evolution trajectory using the state feedback RBPF algorithm, and perform global optimization of the factor graph to obtain the inertial drift sequence. The coupled extrapolation module is used to perform state-drift coupled extrapolation operations on the flight state sequence based on the inertial drift sequence and through the multi-scale gated Mamba model to construct the position extrapolation sequence; The convergence correction module is used to perform trajectory topology convergence operation on the continuous position extrapolation sequence and perform continuous trajectory correction to form a positioning trajectory sequence; The positioning mapping module is used to perform real-time spatial position mapping processing on the positioning trajectory sequence to generate real-time positioning results for the aircraft.

[0023] In this embodiment, the inertial navigation information represents acceleration data, angular velocity data, attitude angle data, and heading angle data collected by intelligent sensors installed inside the aircraft. Preprocessing includes time synchronization, outlier data removal, noise filtering, data normalization, and unified encoding.

[0024] In this embodiment, the state evolution module includes: A state evolution matrix is ​​constructed based on the flight state sequence, and continuous state association mapping is performed on the state evolution matrix to generate state propagation paths; Introducing a multi-scale gated Mamba model; Compared with the traditional Mamba model, the multi-scale gated Mamba model adds bidirectional propagation units, continuous evolution units, and drift coupling units; In the bidirectional propagation unit, bidirectional state association encoding is performed on the state propagation path to extract state propagation features at different time scales and construct a bidirectional state association sequence. The bidirectional state association sequence is input into the continuous evolution unit, and continuous state evolution fitting processing is performed to generate state evolution features. Correlation analysis is performed on the evolution characteristics of each state to extract the corresponding attitude evolution state, trajectory change state and motion change state, and to construct a flight behavior sequence.

[0025] In this embodiment, the calculation process in the bidirectional propagation unit specifically includes: A bidirectional diffusion algorithm based on time inversion mechanism is adopted to perform reverse time backtracking diffusion and forward time recursion diffusion processing on the state propagation path. During the continuous turning process of the aircraft, 12 consecutive flight state nodes are used as the propagation window. Reverse time backtracking diffusion processing is performed on the first 6 time steps, and forward time recursion diffusion processing is performed on the last 6 time steps to obtain the corresponding historical state influence value and future state influence value. Based on the historical and future state influence values, bidirectional state association coding is performed on the flight state sequence. During the continuous climb of the aircraft, the state association strength, state propagation distance, and state transition continuity between adjacent flight states are calculated. The state association strength is maintained between 0.83 and 0.91, the state propagation distance is maintained between 3.6 and 5.2, and the state transition continuity is maintained above 0.88. State association features are constructed based on the state association strength, state propagation distance, and state transition continuity. Dynamic gating contention handling is performed on the associated features of each state: Based on the correlation characteristics of each state, the gating competition weights corresponding to each flight state are calculated. During the continuous cruise phase, the gating competition weights corresponding to the stable flight state are maintained between 0.74 and 0.93, and the gating competition weights corresponding to the attitude change state are maintained between 0.28 and 0.46. The competition ranking process is then performed on each flight state based on the gating competition weights. Based on the competition ranking results, propagation suppression processing is performed on flight states with gating competition weights below 0.45, and priority propagation processing is performed on flight states with gating competition weights above 0.72, in order to identify abrupt and stable states and construct a state label sequence. Based on the state label sequence, the state propagation path is divided into a multi-timescale partitioning process, which divides the state propagation path into a propagation interval containing 8 time steps, a propagation interval containing 16 time steps, and a propagation interval containing 32 time steps. In each propagation interval, the state propagation direction and state association continuity are extracted to generate state propagation features at each time scale. Fusion encoding is performed on the state propagation features at different time scales to construct a bidirectional state association sequence.

[0026] In this embodiment, the calculation process in the continuous evolution unit specifically includes: The bidirectional state association sequence is input into the continuous evolution unit, and a continuous state evolution field is constructed based on the neural ordinary differential equation. During continuous flight, 256 consecutive state nodes are used as a state evolution batch to perform continuous state evolution calculation on the bidirectional state association sequence. In a continuous state evolution field, continuous state flow expansion processing is performed on the bidirectional state association sequence to construct a state evolution manifold. During the continuous turning and climbing of the aircraft, the state flow distribution density in the state evolution manifold is maintained between 0.68 and 0.94. Based on the state evolution manifold, continuous state trajectory tracking is performed on each flight state to calculate the state evolution velocity, state evolution direction, and state evolution curvature corresponding to each flight state. The state evolution velocity is maintained between 2.4 and 6.8, the state evolution direction change angle is maintained between 11.3° and 27.6°, and the state evolution curvature is maintained between 0.32 and 0.87, forming a convergence parameter set. Based on the convergence parameter set, state convergence analysis is performed on the state evolution manifold. When the state evolution velocity is higher than 6.1 and the state evolution curvature is higher than 0.75, the corresponding region is identified as a state change region. When the state evolution velocity is lower than 3.2 and the state evolution curvature is lower than 0.45, the corresponding region is identified as a state stable region, thereby constructing the state evolution trajectory. Continuous state prediction processing is performed on the state evolution trajectory. Within 32 consecutive time steps, continuous prediction is performed on the subsequent flight state to generate state evolution features and construct flight behavior sequences.

[0027] In this embodiment, the inertial propagation module includes: Based on the flight behavior sequence, the drift propagation relationship between each flight state is extracted and a drift propagation map is constructed. During the continuous turning process of the aircraft, the drift propagation connection relationship is established for the flight state nodes with attitude change amplitude greater than 12°, so that a continuous drift propagation path is formed between adjacent flight state nodes. A continuous drift-diffusion analysis was performed on the drift propagation map to construct the drift evolution trajectory. During the continuous climbing phase, continuous diffusion calculations were performed on the drift propagation nodes according to the drift propagation direction, so that the drift propagation range gradually expanded from the initial 3.2m to 9.8m. The state feedback RBPF algorithm is used to perform particle state prediction and feedback update processing on the drift evolution trajectory to construct the drift state sequence; A drift factor graph is constructed based on the drift state sequence, and global association constraint processing is performed on each node in the drift factor graph. In the drift factor graph, 16 consecutive drift state nodes are constructed as a constraint association region, and the state association error and state propagation deviation between each node are calculated, so that the state association error is reduced from 0.63 to 0.21 and the state propagation deviation is reduced from 0.74 to 0.28, thus obtaining the constraint factor graph. Based on the constraint factor diagram, global convergence optimization is performed on the drift state sequence. During 32 consecutive rounds of drift convergence, global drift correction is performed on the drift anomaly region, reducing the cumulative drift error during the aircraft's continuous flight from 11.6m to 2.4m, and finally generating the inertial drift sequence.

[0028] In this embodiment, the particle state prediction and feedback update process specifically includes: During continuous flight, 512 drift state particles are generated based on the drift evolution trajectory, and the initial propagation weight of each drift state particle is set to 0.5. During the continuous turning phase of the aircraft, inertial propagation prediction is performed based on the current propagation direction and drift change of each drifting particle to maintain the predicted drift propagation distance between 4.3m and 8.7m, thereby generating the predicted particle state. Based on the flight state sequence, state matching processing is performed on each predicted particle state, and the state deviation vector of each predicted particle state is calculated. Based on the state deviation vector, feedback weight update processing is performed on the state of each predicted particle, and the corresponding propagation weight is adjusted. When the state deviation vector is lower than 0.2, the propagation weight of the corresponding drifting state particle is increased, and when the state deviation vector is higher than 0.45, the propagation weight of the corresponding drifting state particle is decreased, so that the updated propagation weight is maintained between 0.31 and 0.93. Based on the adjusted propagation weights, the states of each predicted particle are corrected by feedback to construct a drift state sequence; During continuous propagation, drifting state particles with a propagation weight lower than 0.3 are eliminated; Based on the retained drift state particles, continuous convergence update processing is performed on the drift evolution trajectory to generate an inertial drift sequence.

[0029] In this embodiment, the coupled deduction module includes: A state-drift coupling matrix is ​​constructed based on the inertial drift sequence and the flight state sequence. A continuous association mapping process is performed on the state-drift coupling matrix. During continuous flight, a state-drift association relationship is established between 32 consecutive flight state nodes and the corresponding inertial drift nodes, so that the state association strength is maintained between 0.61 and 0.92, thereby obtaining the coupling propagation path. In the drift coupling unit of the multi-scale gated Mamba model, bidirectional coupling propagation processing is performed on the coupling propagation path. During the continuous turning phase of the aircraft, forward state propagation and reverse drift propagation are performed on the coupling propagation path simultaneously, so that the coupling propagation distance is maintained between 5.4m and 12.8m, thereby forming a coupling state sequence. Based on the coupled state sequence, state-drift correlation analysis is performed on each flight state to calculate the coupling parameter vector. Then, based on the coupling parameter vector, continuous state extrapolation is performed on the coupled state sequence. In 24 consecutive time steps, continuous position extrapolation is performed on the subsequent flight states based on the coupling parameter vector, reducing the position extrapolation deviation from 8.6m to 2.7m, thereby forming the coupled extrapolation trajectory. Multi-timescale state propagation and continuous position fitting are performed on the coupled inference trajectory to keep the continuous position fitting error within 1.9m and generate a position inference sequence.

[0030] In this embodiment, the continuous state deduction process specifically includes: Based on the coupling parameter vector, continuous state mapping is performed on the coupled state sequence to generate a state evolution sequence. The state change amplitude in the state evolution sequence is maintained between 0.18 and 0.76. Through the liquid time constant network, dynamic time constant adjustment is performed on each flight state according to the state evolution sequence. During the continuous turning phase of the aircraft, the time constant is dynamically adjusted according to the rate of change of the flight state, so that the time constant is adjusted from 0.32 to 0.87. Based on the adjusted time constant, continuous recursive processing is performed on each flight state to form a state recursive sequence. Based on the state recursive sequence, continuous drift feedback processing is performed on each flight state, and the drift propagation deviation and state residual corresponding to each flight state are calculated. During the continuous climb, the drift propagation deviation is maintained between 1.6m and 4.3m, and the state residual is maintained between 0.11 and 0.38, thus forming a drift feedback sequence. Based on the drift feedback sequence, dynamic state convergence processing is performed on the state recursion sequence to generate a coupled inference trajectory.

[0031] In this embodiment, the convergence correction module includes: The trajectory connection relationships in the position projection sequence are extracted, and a trajectory topology map is constructed. During continuous flight, the trajectory connection relationships of 64 consecutive position projection nodes are established. When the position offset distance between adjacent trajectory nodes is greater than 6.5m, the corresponding area is identified as a trajectory anomaly area, thereby generating a trajectory anomaly sequence. Based on the trajectory anomaly sequence, a topological convergence propagation operation is performed along the trajectory topology map. During the continuous turning phase, a topological convergence calculation is performed along the trajectory propagation direction based on the trajectory anomaly region, so that the change angle of the topological convergence direction is maintained between 8.3° and 21.6°, and the topological convergence strength is maintained between 0.52 and 0.91, thereby forming a topological convergence sequence. A continuous neighborhood association analysis operation is performed on the topological convergence sequence. During the continuous neighborhood association analysis, the 16 adjacent position inference nodes are constructed as a continuous neighborhood region, and dynamic trajectory correction is performed based on the trajectory offset change in the neighborhood region, so that the trajectory offset error is reduced from 7.4m to 2.1m, thereby forming a trajectory convergence sequence. Continuous trajectory fitting is performed on the trajectory convergence sequence to keep the continuous trajectory fitting error within 1.8m, and trajectory consistency correction is performed on the trajectory connection area to reduce the number of trajectory jumps from 34 to 5, thus forming a positioning trajectory sequence.

[0032] Example 1: To verify the feasibility of this invention in practice, it was applied to an intelligent navigation scenario for aircraft in a complex low-altitude flight environment. In this scenario, the aircraft continuously performed low-altitude cruise, rapid turns, continuous climbs, and complex trajectory switching operations, maintaining a flight altitude between 1200m and 3800m, an average flight speed between 210km / h and 460km / h, and a continuous flight time of 4.5 hours. During the flight, the aircraft's internal inertial measurement sensors continuously collected acceleration data, angular velocity data, attitude angle data, and heading angle data at a sampling frequency of 200Hz, accumulating approximately 3.24 million sets of flight status data.

[0033] In this embodiment, the acquired flight status data is first processed through time synchronization, outlier removal, noise filtering, and unified encoding to form a flight status sequence. Subsequently, a multi-scale gated Mamba model is used to perform bidirectional state association encoding on the flight status sequence, analyzing the flight status propagation relationship at different time scales and dynamically fitting the continuous state evolution process to form a flight behavior sequence. During continuous flight, the system cumulatively identified 1247 flight status change nodes, including 286 attitude change nodes, 713 stable flight nodes, and 248 continuous turning nodes.

[0034] Because aircraft are prone to inertial drift propagation during rapid turns and continuous climbs, this embodiment further constructs a drift propagation graph based on the flight behavior sequence and performs continuous drift propagation analysis on the graph to form a drift evolution trajectory. Subsequently, the drift state particles are updated using the state feedback RBPF algorithm, and the drift propagation process is continuously converged and optimized using a factor graph global optimization method. Throughout the flight, a total of 48,000 drift state particles are generated, approximately 260,000 particle feedback updates are performed, and approximately 83,000 factor graph constraint optimization nodes are completed.

[0035] After obtaining the inertial drift sequence, this embodiment further constructs a state-drift coupling matrix based on the inertial drift sequence and the flight state sequence, and performs joint inference processing on the correlation between the flight state and inertial drift through the drift coupling unit to form a position inference sequence. Subsequently, trajectory topology convergence and continuous trajectory correction processing are performed on the position inference sequence to finally form the real-time positioning result of the aircraft.

[0036] To verify the practical application effect of the present invention, continuous positioning tests were conducted on the same flight process using both traditional inertial navigation methods and the method of the present invention. Statistical analysis was performed on positioning deviation, cumulative drift error, trajectory continuity, and the number of abnormal trajectories, yielding the following results: Table 1 Comparative Analysis of Aircraft Smart Positioning Performance

[0037] As can be seen from the test data in Table 1, the method of the present invention exhibits high positioning stability and drift suppression capability under complex flight scenarios. Specifically, the average positioning deviation of the method of the present invention is reduced from 8.74m to 1.93m, and the maximum positioning deviation is reduced from 15.28m to 3.67m, indicating that the present invention can effectively improve the real-time spatial positioning accuracy of aircraft. Simultaneously, the cumulative drift error of the method of the present invention is reduced to 2.16m, a reduction of 10.25m compared to traditional inertial navigation methods, demonstrating that the present invention can effectively suppress the inertial drift propagation problem during long-term flight.

[0038] Furthermore, this invention performs continuous state evolution analysis on flight status using a multi-scale gated Mamba model, achieving a state identification accuracy of 96.5%, which is 13.1% higher than traditional methods, and increasing the continuous position extrapolation accuracy to 95.2%. Simultaneously, through trajectory topology convergence and continuous trajectory correction processing, the number of trajectory jumps is reduced from 37 to 6, and the spatial position mapping stability reaches 98.4%, demonstrating that this invention can effectively improve trajectory continuity and spatial mapping reliability in complex flight environments.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent positioning system for aviation based on inertial navigation, characterized in that, include: The data acquisition module is used to acquire and preprocess the inertial navigation information of the aircraft during flight, and to construct a flight state sequence. The state evolution module is used to construct a state evolution matrix based on the flight state sequence. It performs bidirectional state association encoding and continuous state evolution fitting operations on the state evolution matrix through a multi-scale gated Mamba model to generate state evolution features and construct flight behavior sequences. The inertial propagation module is used to perform inertial drift propagation analysis on the flight behavior sequence, construct the drift evolution trajectory, perform state estimation operation on the drift evolution trajectory using the state feedback RBPF algorithm, and perform global optimization of the factor graph to obtain the inertial drift sequence. The coupled extrapolation module is used to perform state-drift coupled extrapolation operations on the flight state sequence based on the inertial drift sequence and through the multi-scale gated Mamba model to construct the position extrapolation sequence; The convergence correction module is used to perform trajectory topology convergence operation on the continuous position extrapolation sequence and perform continuous trajectory correction to form a positioning trajectory sequence; The positioning mapping module is used to perform real-time spatial position mapping processing on the positioning trajectory sequence to generate real-time positioning results for the aircraft.

2. The aviation intelligent positioning system based on inertial navigation according to claim 1, characterized in that, The inertial navigation information refers to acceleration data, angular velocity data, attitude angle data, and heading angle data collected by intelligent sensors installed inside the aircraft. The preprocessing includes time synchronization, outlier data removal, noise filtering, data normalization, and unified encoding processing.

3. The aviation intelligent positioning system based on inertial navigation according to claim 1, characterized in that, The state evolution module includes: A state evolution matrix is ​​constructed based on the flight state sequence, and continuous state association mapping is performed on the state evolution matrix to generate state propagation paths; Introducing a multi-scale gated Mamba model; Compared with the traditional Mamba model, the multi-scale gated Mamba model adds bidirectional propagation units, continuous evolution units, and drift coupling units; In the bidirectional propagation unit, bidirectional state association encoding is performed on the state propagation path to extract state propagation features at different time scales and construct a bidirectional state association sequence. The bidirectional state association sequence is input into the continuous evolution unit, and continuous state evolution fitting processing is performed to generate state evolution features. Correlation analysis is performed on the evolution characteristics of each state to extract the corresponding attitude evolution state, trajectory change state and motion change state, and to construct a flight behavior sequence.

4. An aviation intelligent positioning system based on inertial navigation according to claim 3, characterized in that, The calculation process in the bidirectional propagation unit specifically includes: A bidirectional diffusion algorithm based on time inversion mechanism is adopted to perform reverse time backtracking diffusion and forward time recursion diffusion processing on the state propagation path, and calculate the historical state influence value and future state influence value of each flight state. Based on the historical state influence value and the future state influence value, bidirectional state association coding is performed on the flight state sequence to calculate the state association strength, state propagation distance and state transition continuity between each flight state, and combine them to obtain the state association features. Dynamic gating contention handling is performed on the associated features of each state: Based on the correlation characteristics of each state, the gating competition weight value corresponding to each flight state is calculated, and the competitive ranking process is performed on each flight state based on the gating competition weight value. Based on the competition ranking results, propagation suppression processing is performed on flight states with gating competition weights lower than the preset competition threshold, and priority propagation processing is performed on flight states with gating competition weights higher than the preset competition threshold, in order to identify abrupt and stable states and construct a state label sequence. Based on the state label sequence, the state propagation path is divided into multiple time scales to construct propagation intervals at multiple time scales. In each propagation interval, the state propagation direction and state association continuity are extracted to generate state propagation features at each time scale. Fusion encoding is performed on the state propagation features at different time scales to construct a bidirectional state association sequence.

5. An aviation intelligent positioning system based on inertial navigation according to claim 3, characterized in that, The calculation process in the continuous evolution unit specifically includes: The bidirectional state association sequence is input into the continuous evolution unit, and the continuous state evolution field is constructed based on the neural ordinary differential equation. In a continuous state evolution field, a continuous state flow expansion process is performed on the bidirectional state association sequence to construct a state evolution manifold; Based on the state evolution manifold, continuous state trajectory tracking is performed on each flight state to calculate the state evolution velocity, state evolution direction and state evolution curvature corresponding to each flight state, forming a convergence parameter set; Based on the convergence parameter set, state convergence analysis is performed on the state evolution manifold to identify abrupt state change regions and stable state regions, and to construct the state evolution trajectory. Continuous state prediction processing is performed on the state evolution trajectory to generate state evolution features and construct flight behavior sequences.

6. An aviation intelligent positioning system based on inertial navigation according to claim 1, characterized in that, The inertial propagation module includes: Based on the flight behavior sequence, the drift propagation relationship between each flight state is extracted, and a drift propagation graph is constructed; Perform continuous drift-diffusion analysis on the drift propagation graph to construct the drift evolution trajectory; The state feedback RBPF algorithm is used to perform particle state prediction and feedback update processing on the drift evolution trajectory to construct the drift state sequence; A drift factor graph is constructed based on the drift state sequence, and global association constraint processing is performed on each node in the drift factor graph. The state association error and state propagation deviation between each node are calculated to obtain the constraint factor graph. Based on the constraint factor diagram, a global convergence optimization process is performed on the drift state sequence to generate an inertial drift sequence.

7. An aviation intelligent positioning system based on inertial navigation according to claim 6, characterized in that, The particle state prediction and feedback update process specifically includes: Multiple drift state particles are constructed based on the drift evolution trajectory, and the propagation weights of each drift state particle are initialized. Perform inertial propagation prediction processing on each drifting particle to generate predicted particle states; Based on the flight state sequence, state matching processing is performed on each predicted particle state, and the state deviation vector of each predicted particle state is calculated. Based on the state deviation vector, feedback weight update processing is performed on the state of each predicted particle to adjust the corresponding propagation weights; Based on the adjusted propagation weights, the states of each predicted particle are corrected by feedback to construct a drift state sequence; Remove drifting state particles whose propagation weight is less than a preset weight threshold; Based on the retained drift state particles, continuous convergence update processing is performed on the drift evolution trajectory to generate an inertial drift sequence.

8. An aviation intelligent positioning system based on inertial navigation according to claim 1, characterized in that, The coupled deduction module includes: A state-drift coupling matrix is ​​constructed based on the inertial drift sequence and the flight state sequence, and a continuous correlation mapping process is performed on the state-drift coupling matrix to obtain the coupling propagation path; In the drift coupling unit of the multi-scale gated Mamba model, bidirectional coupling propagation processing is performed on the coupling propagation path to construct a coupling state sequence; Based on the coupled state sequence, state-drift correlation analysis is performed on each flight state to calculate the coupling parameter vector. Then, based on the coupling parameter vector, continuous state deduction is performed on the coupled state sequence to construct the coupled deduction trajectory. Multi-timescale state propagation and continuous position fitting are performed on the coupled inference trajectory to generate a position inference sequence.

9. An aviation intelligent positioning system based on inertial navigation according to claim 8, characterized in that, The continuous state deduction process specifically includes: Based on the coupling parameter vector, a continuous state mapping process is performed on the coupled state sequence to generate a state evolution sequence; By using a liquid time constant network, dynamic time constant adjustment is performed on each flight state according to the state evolution sequence to construct a state recursion sequence; Based on the state recursive sequence, continuous drift feedback processing is performed on each flight state to calculate the drift propagation deviation and state residual corresponding to each flight state, and a drift feedback sequence is constructed. Based on the drift feedback sequence, dynamic state convergence processing is performed on the state recursion sequence to generate a coupled inference trajectory.

10. An aviation intelligent positioning system based on inertial navigation according to claim 1, characterized in that, The convergence correction module includes: Extract the trajectory connectivity in the location extrapolation sequence, construct the trajectory topology map, identify trajectory anomaly regions, and generate trajectory anomaly sequences; Based on the trajectory anomaly sequence, perform topological convergence propagation operation along the trajectory topology graph, calculate the topological convergence direction and topological convergence strength corresponding to each inferred node, and construct the topological convergence sequence. Perform continuous neighborhood association analysis on the topological convergence sequence, and perform dynamic trajectory convergence correction on the position extrapolation sequence based on the results of the continuous neighborhood association analysis to generate a trajectory convergence sequence; Multi-timescale continuous trajectory fitting and trajectory consistency correction operations are performed on the trajectory convergence sequence to form a positioning trajectory sequence.

Images