An inertial navigation device error prediction method based on data analysis

Abstract

The application discloses an inertial navigation device error prediction method based on data analysis, which comprises the following steps: collecting self operation data, preprocessing standard inertial navigation operation data; dividing continuous segments to generate an inertial navigation calculation segment set, and extracting device error to generate a device error state sequence; performing time forward inertial navigation recursion calculation and time reverse inertial navigation recursion calculation to generate forward calculation trajectory and backward calculation trajectory; mapping to the same time coordinate and extracting to generate calculation suture crack characteristics; performing attitude closed loop comparison, speed closed loop comparison and position closed loop comparison, and performing time sequence conduction analysis to generate navigation closed loop conservation breakage migration chain; generating error hypothesis set, and performing forward error explanation processing and reverse device checking to generate a credible error hypothesis chain; and generating inertial navigation device error prediction result. Through calculation suture and breakage migration, the application realizes inertial navigation error prediction, and has the advantages of early identification, high credibility and explainability.

Description

Technical Field

[0001] This invention relates to the field of inertial navigation equipment condition monitoring technology, and in particular to an error prediction method for inertial navigation equipment based on data analysis. Background Technology

[0002] Inertial navigation devices typically collect operational data such as angular velocity and acceleration using gyroscopes and accelerometers, and obtain navigation results through recursive calculations of attitude, velocity, and position. Existing error prediction methods are mostly based on sensor bias estimation, temperature compensation, filtering correction, or historical error trend analysis to model and compensate for the drift error of inertial navigation devices.

[0003] However, existing methods typically focus on estimating a single error quantity, making it difficult to identify implicit deviations between forward and backward calculations when external positioning references are lacking. They also struggle to determine whether errors are continuously propagating from the attitude layer to the velocity and position layers, resulting in insufficient explanation of error sources, low reliability of prediction results, and difficulty in timely reflecting the future error expansion trend of inertial navigation equipment. Summary of the Invention

[0004] One objective of this invention is to propose an error prediction method for inertial navigation devices based on data analysis. This invention achieves inertial navigation error prediction by solving stitching and defect migration, and has the advantages of early identification, high reliability, and interpretability.

[0005] An error prediction method for inertial navigation devices based on data analysis according to an embodiment of the present invention includes the following steps: Collect the inertial navigation device's own operational data during operation, and preprocess the data to generate standard inertial navigation operational data; The standard inertial navigation system (INS) operating data is divided into continuous segments to generate an INS solution segment set. The solution state at the beginning and end of each segment is extracted from the INS solution segment. Device errors are extracted from the standard INS operating data to generate a device error state sequence. Perform forward time inertial navigation recursive calculation and backward time inertial navigation recursive calculation on the same inertial navigation solution segment to generate forward and backward solution trajectories; The forward and backward calculation trajectories corresponding to the same inertial navigation calculation segment are mapped to the same time coordinate, and attitude deviation, velocity deviation and position deviation are extracted to generate the calculation stitch crack features; Based on standard inertial navigation operation data, attitude closed-loop comparison, velocity closed-loop comparison, and position closed-loop comparison are performed, and time-series propagation analysis is conducted to generate navigation closed-loop conservation failure transfer chain. Based on the characteristics of the stitching crack, the navigation closed-loop conservation failure migration chain, and the device error state sequence, an error hypothesis set is generated, and forward error interpretation processing and reverse device verification are performed on each error hypothesis to generate a credible error hypothesis chain. Error prediction results for inertial navigation devices are generated based on the characteristics of the suture crack, the navigation closed-loop conservation failure migration chain, and the credible error hypothesis chain.

[0006] Optionally, the preprocessing includes time synchronization, outlier removal, coordinate system unification, and scale normalization.

[0007] Optionally, the generation of the device error state sequence includes: Standard inertial navigation operation data is read in the order of sampling time. Based on the continuity of sampling time and the integrity of the solution state, the standard inertial navigation operation data is divided into continuous segments. Each continuous segment is used as an inertial navigation solution segment to generate an inertial navigation solution segment set. Read the attitude, velocity, and position data corresponding to the initial sampling time of each inertial navigation solution segment from the inertial navigation solution segment set to generate the segment start solution state, and read the attitude, velocity, and position data corresponding to the end sampling time to generate the segment end solution state; Zero-bias state extraction is performed on the triaxial angular velocity and triaxial acceleration data in the standard inertial navigation system (INS) operating data. Response state extraction is performed on the temperature data, vibration data, and attitude calculation data in the standard INS operating data to generate a device error state sequence.

[0008] Optionally, the generation of the forward and backward solution trajectories includes: Read triaxial angular velocity data, triaxial acceleration data, and sampling time data from the data range corresponding to the same inertial navigation solution segment, and generate segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence in ascending order of sampling time; Align the initial solution state of the segment with the initial sampling time of the segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence to generate a forward recursion starting point; align the final solution state of the segment with the final sampling time of the segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence to generate a reverse recursion starting point. Starting from the forward recursion starting point, perform forward time inertial navigation recursion calculation on the same inertial navigation solution segment, and generate forward attitude state, forward velocity state and forward position state at each sampling time. The forward attitude state, forward velocity state, and forward position state are connected in ascending order of sampling time to generate the forward calculated trajectory; Starting from the reverse recursion starting point, time-reverse inertial navigation recursion is performed on the same inertial navigation solution segment, generating reverse attitude state, reverse velocity state and reverse position state at each sampling time. The reverse attitude state, reverse velocity state, and reverse position state are connected in ascending order of sampling time to generate the backward calculated trajectory.

[0009] Optionally, the generation of the solved stitch crack features includes: Using the sampling time of the same inertial navigation solution segment as the time reference, the sampling times of the forward solution trajectory and the backward solution trajectory are aligned to generate the same time coordinates. The forward solution trajectory and the backward solution trajectory are then mapped to the same time coordinates to generate forward and backward trajectories with the same coordinates. Pair the forward and backward trajectories with the same time coordinate at each sampling time to generate a trajectory stitching region; Within the trajectory stitching region, the difference between the positive attitude state and the reverse attitude state is extracted to generate an attitude deviation sequence, the difference between the positive velocity state and the reverse velocity state is extracted to generate a velocity deviation sequence, and the difference between the positive position state and the reverse position state is extracted to generate a position deviation sequence. The attitude deviation sequence, velocity deviation sequence and position deviation sequence are synchronously arranged according to the same time coordinate to generate a three-layer stitched deviation sequence. Then, the continuously increasing deviation segment, the deviation direction-maintaining segment and the deviation amplitude abrupt segment are identified from the three-layer stitched deviation sequence. The features of continuously increasing deviation segments, deviation direction-maintaining segments, and deviation amplitude abruptly changed segments are summarized to generate the features of the suture crack solution.

[0010] Optionally, the generation of the navigation closed-loop conservation violation migration chain includes: Extract triaxial angular velocity data, triaxial acceleration data, attitude calculation data, velocity calculation data, position calculation data, and timestamps corresponding to each sampled data from standard inertial navigation operation data, and generate a sampling time series in ascending order of timestamps; The three-axis angular velocity data, three-axis acceleration data and velocity solution data are integrated and recursively processed according to the sampling time series, and the corresponding integrated and recursive processing results are compared in closed loop to generate attitude closed loop broken sequence, velocity closed loop broken sequence and position closed loop broken sequence. The attitude closed-loop failure sequence, velocity closed-loop failure sequence, and position closed-loop failure sequence are synchronously arranged according to the sampling time series, and the failure start time, failure duration interval, failure change direction, and failure amplitude change are marked to generate a closed-loop failure state sequence. Based on the closed-loop broken state sequence, temporal adjacency matching is performed on the attitude closed-loop broken sequence and the velocity closed-loop broken sequence to filter and generate candidate attitude-to-velocity transmission segments. Based on the closed-loop broken state sequence, temporal adjacency matching is performed on the velocity closed-loop broken sequence and the position closed-loop broken sequence to filter and generate candidate velocity-to-position transmission segments. The attitude closed-loop failure sequence, attitude-to-velocity candidate transmission segments, and velocity-to-position candidate transmission segments are time-series processed, and closed-loop failure segments with conflicting failure time order, conflicting failure change direction, and broken failure duration intervals are removed to generate the target failure transmission segment. Determine the failure start time and failure end time from the target failure transmission segment, and determine the failure transmission time when there is cross-layer transmission in the target failure transmission segment. Generate a navigation closed-loop conservation failure migration chain according to the closed-loop failure sequence contained in the target failure transmission segment.

[0011] Optionally, the generation of the credible error hypothesis chain includes: An error hypothesis set is generated based on the characteristics of the stitched crack, the navigation closed-loop conservation failure migration chain, and the device error state sequence. Each error hypothesis in the set of error hypotheses is mapped to the hypothetical attitude deviation, hypothetical velocity deviation, hypothetical position deviation, hypothetical deviation change direction, hypothetical deviation change magnitude, hypothesis failure start time, hypothesis failure propagation time, hypothesis failure termination time, and hypothesis failure duration interval, thereby generating the hypothesis interpretation state; A consistency comparison is performed between the hypothesized explanation state and the solved stitch crack features and the navigation closed-loop conservation failure migration chain to generate an explanation consistency error hypothesis set; For each error hypothesis in the consistent error hypothesis set, reverse device verification is performed. Each error hypothesis is backtracked to the assumed device error state, and the assumed device error state is compared with the corresponding device error state in the device error state sequence to generate a consistent error hypothesis set. A credible error hypothesis chain is generated by chaining the error hypotheses in the consistent error hypothesis set with the corresponding failure start time, failure end time, and failure propagation time during cross-layer propagation in the navigation closed-loop conservation failure migration chain.

[0012] Optionally, the generation of the inertial navigation device error prediction result includes: From the solved suture crack features, attitude deviation, velocity deviation, position deviation, deviation change direction and deviation change magnitude are extracted, and future error types and error growth stages are determined; Extract the failure start time, failure end time, and failure propagation time during cross-layer propagation from the navigation closed-loop conservation failure migration chain, and determine the range of error influence; Read the error hypotheses, their chain order, and the error source types corresponding to the error hypotheses from the credible error hypothesis chain, and determine the error source types and prediction confidence. The error prediction results for inertial navigation devices are generated by associating and encapsulating the future error type, error growth stage, error impact range, error source type, and prediction reliability.

[0013] The beneficial effects of this invention are: This invention collects the operating data of an inertial navigation device during its operation, and preprocesses, divides, and extracts device errors from this data to form a sequence of standard inertial navigation operating data, inertial navigation solution segments, solution states at the beginning and end of segments, and device error states. This allows subsequent error prediction to no longer rely solely on external positioning references or single historical error curves, but instead to construct an analyzable and traceable error observation basis from the operating state of the inertial navigation device itself, thus improving the applicability of error prediction under conditions without external references.

[0014] This invention uses the initial and final solution states of the same inertial navigation (INS) segment as the starting points for solution calculation, performing forward and backward inertial navigation recursive calculations. The forward and backward solution trajectories are mapped to the same time coordinate to generate a trajectory stitching region, thereby extracting attitude deviation, velocity deviation, and position deviation to generate solution stitching crack features. This processing method can transform implicit inconsistencies that are difficult to observe directly between forward and backward calculations into explicit error precursor features, avoiding the need to judge INS errors solely based on final position drift or single-point error thresholds, thus improving the ability to identify early error expansion trends.

[0015] This invention further performs attitude closed-loop comparison, velocity closed-loop comparison, and position closed-loop comparison based on standard inertial navigation operational data to generate attitude closed-loop violation sequences, velocity closed-loop violation sequences, and position closed-loop violation sequences. Furthermore, it generates a navigation closed-loop conservation violation transfer chain through time-series propagation analysis. This transfer chain allows for the determination of whether errors propagate continuously from the attitude layer to the velocity and position layers, clarifying the scope of error influence and the stages of error growth. This transforms error prediction from simple numerical estimation to hierarchical propagation relationship analysis, improving the structure and accuracy of error prediction results.

[0016] This invention also generates a set of error hypotheses based on the characteristics of the stitched cracks, the navigation closed-loop conservation failure migration chain, and the device error state sequence. Each error hypothesis undergoes forward error interpretation processing and reverse device verification to eliminate inconsistent error hypotheses and generate a credible error hypothesis chain. By combining forward interpretation and reverse verification, the future error type, error source type, and prediction credibility can be correlated and output, enhancing the interpretability and reliability of error prediction results. This provides a more stable basis for error warning, maintenance decisions, and navigation result correction for inertial navigation equipment. Attached Figure Description

[0017] 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 1 This is an overall flowchart of an inertial navigation device error prediction method based on data analysis proposed in this invention; Figure 2 This is a schematic diagram illustrating the process of generating forward and backward calculated trajectories in an inertial navigation device error prediction method based on data analysis proposed in this invention. Figure 3 This is a schematic diagram illustrating the process of generating the navigation closed-loop conservation failure migration chain in the data analysis-based inertial navigation device error prediction method proposed in this invention. Detailed Implementation

[0018] 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.

[0019] refer to Figures 1-3 A data analysis-based error prediction method for inertial navigation devices includes the following steps: Collect the inertial navigation device's own operational data during operation, and preprocess the data to generate standard inertial navigation operational data; The standard inertial navigation (INS) operation data is divided into continuous segments to generate an INS solution segment set. The start-of-segment and end-of-segment solution states of each INS solution segment in the INS solution segment set are extracted. Device errors are extracted from the standard INS operation data to generate a device error state sequence. Using the initial and final solution states of the same inertial navigation solution segment as the solution starting points, the same inertial navigation solution segment is subjected to forward time inertial navigation recursive solution and backward time inertial navigation recursive solution to generate forward solution trajectory and backward solution trajectory. The forward and backward calculation trajectories corresponding to the same inertial navigation calculation segment are mapped to the same time coordinate to generate a trajectory stitching region. Attitude deviation, velocity deviation and position deviation are extracted from the trajectory stitching region to generate calculation stitching crack features. Based on standard inertial navigation operation data, attitude closed-loop comparison, velocity closed-loop comparison and position closed-loop comparison are performed to generate attitude closed-loop broken sequence, velocity closed-loop broken sequence and position closed-loop broken sequence, and time series propagation analysis is performed to generate navigation closed-loop conservation broken transfer chain. An error hypothesis set is generated based on the characteristics of the solved stitch crack, the navigation closed-loop conservation failure migration chain, and the device error state sequence. Forward error interpretation processing and reverse device verification are performed on each error hypothesis in the error hypothesis set. Error hypotheses that are inconsistent with the characteristics of the solved stitch crack, the navigation closed-loop conservation failure migration chain, and the device error state sequence are eliminated to generate a credible error hypothesis chain. Based on the characteristics of the suture crack, the navigation closed-loop conservation failure migration chain, and the credible error hypothesis chain, the error prediction results of the inertial navigation device are generated. The error prediction results of the inertial navigation device include the future error type, the error growth stage, the error impact range, the error source type, and the prediction credibility.

[0020] In this embodiment, the preprocessing includes time synchronization, outlier sampling removal, coordinate system unification, and scale normalization.

[0021] In this embodiment, the generation of the device error state sequence includes: Standard inertial navigation operation data is read in the order of sampling time. Based on the continuity of sampling time and the integrity of the solution state, the standard inertial navigation operation data is divided into continuous segments. Each continuous segment is used as an inertial navigation solution segment to generate an inertial navigation solution segment set. When dividing continuous segments, a sampling index chain is first generated in ascending order of timestamps. The time interval between adjacent sampling points is calculated one by one, and the effective state of attitude solution data, velocity solution data and position solution data at each sampling point is recorded synchronously. The continuous segments of sampling indexes with continuous time intervals and complete correspondence of the three types of solution data are organized into candidate segments. Then, the starting sampling index, the ending sampling index and the segment length are recorded for each candidate segment. The candidate segments are arranged in the order of sampling time to generate a set of inertial navigation solution segments. Read the attitude, velocity, and position data corresponding to the initial sampling time of each inertial navigation solution segment from the inertial navigation solution segment set to generate the segment start solution state, and read the attitude, velocity, and position data corresponding to the end sampling time to generate the segment end solution state; Zero-bias state extraction is performed on the triaxial angular velocity data and triaxial acceleration data in the standard inertial navigation operation data. Response state extraction is performed on the temperature data, vibration data, and attitude calculation data in the standard inertial navigation operation data to generate a device error state sequence. The device error state sequence includes gyroscope zero-bias state, accelerometer zero-bias state, temperature response state, vibration response state, and installation angle offset state. During zero-bias state extraction, the attitude change and velocity change between adjacent sampling points are first calculated, and the attitude change and velocity change are continuously averaged to generate stable attitude change sequences and stable velocity change sequences. Then, the data segments that continuously overlap at the same sampling time are organized into zero-bias extraction data segments. The drift mean is calculated for the three-axis angular velocity data to generate the gyroscope zero-bias state. The residual mean is calculated for the three-axis acceleration data after deducting the gravity component to generate the accelerometer zero-bias state. Then, the temperature data, vibration data, attitude calculation data, gyroscope zero-bias state and accelerometer zero-bias state are associated according to the same sampling time to generate temperature response state, vibration response state and installation angle offset state. Finally, the above states are arranged in the order of sampling time to generate the device error state sequence. The specific steps of extracting the response state include: associating temperature data with the gyroscope zero-bias state and the accelerometer zero-bias state at the same sampling time to generate a temperature response state; associating vibration data with short-term fluctuations of the gyroscope zero-bias state and the accelerometer zero-bias state at the same sampling time to generate a vibration response state; and after excluding the continuous unidirectional attitude offsets in the attitude calculation data from the gyroscope zero-bias state and the accelerometer zero-bias state, generating an installation angle offset state, which is then arranged in the order of sampling time to form a device error state sequence.

[0022] In this embodiment, the generation of the forward and backward solution trajectories includes: Read triaxial angular velocity data, triaxial acceleration data, and sampling time data from the data range corresponding to the same inertial navigation solution segment, and generate segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence in ascending order of sampling time; Align the initial solution state of the segment with the initial sampling time of the segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence to generate a forward recursion starting point; align the final solution state of the segment with the final sampling time of the segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence to generate a reverse recursion starting point. Starting from the forward recursion starting point, the segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence are called in ascending order of sampling time to perform forward time inertial navigation recursion calculation on the same inertial navigation solution segment, generating forward attitude state, forward velocity state, and forward position state at each sampling time. The specific steps of performing the forward inertial navigation recursive solution include: during the forward inertial navigation recursive solution, the attitude, velocity, and position at the starting point of the forward recursion are used as the initial solution state. The segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence are read in ascending order of sampling time. First, the forward attitude state is updated using the segment angular velocity sequence and segment time interval sequence. Then, the segment acceleration sequence is converted to the navigation coordinate system and the forward velocity state is updated. Subsequently, the forward position state is updated according to the forward velocity state and segment time interval sequence. The forward attitude state, forward velocity state, and forward position state generated at each sampling time are arranged in chronological order to form the forward recursive result. The forward attitude state, forward velocity state, and forward position state are connected in ascending order of sampling time to generate the forward calculated trajectory; Starting from the reverse recursion starting point, the segment angular velocity sequence, segment acceleration sequence and segment time interval sequence are called in descending order of sampling time. The same inertial navigation solution segment is subjected to time-reverse inertial navigation recursion solution, and the reverse attitude state, reverse velocity state and reverse position state are generated at each sampling time. The specific steps of performing time-reverse inertial navigation recursive calculation include: during time-reverse inertial navigation recursive calculation, the attitude, velocity, and position at the reverse recursion starting point are used as the initial calculation state. The segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence are read in descending order of sampling time. The attitude change, velocity change, and position change between adjacent sampling points are deducted in reverse. Reverse attitude state, reverse velocity state, and reverse position state are generated at each sampling time. Then, the reverse attitude state, reverse velocity state, and reverse position state are rearranged in ascending order of sampling time to form a backward calculation trajectory that corresponds to the forward calculation trajectory. The reverse attitude state, reverse velocity state, and reverse position state are connected in ascending order of sampling time to generate the backward calculated trajectory.

[0023] In this embodiment, the generation of the solution for the suture crack features includes: Using the sampling time of the same inertial navigation solution segment as the time reference, the sampling times of the forward solution trajectory and the backward solution trajectory are aligned to generate the same time coordinates. The forward solution trajectory and the backward solution trajectory are then mapped to the same time coordinates to generate forward and backward trajectories with the same coordinates. The forward and backward trajectories with the same time coordinate are paired up at each sampling moment according to the same time coordinate. The continuous paired sampling moments that simultaneously have positive attitude state, reverse attitude state, positive velocity state, reverse velocity state, positive position state, and reverse position state are retained to generate the trajectory stitching region. The generation of the trajectory stitching region specifically includes: after generating the same time coordinate, reading the attitude state, velocity state and position state corresponding to the forward and backward trajectories at the same sampling time, organizing the sampling times where there is a corresponding relationship between the forward attitude state, backward attitude state, forward velocity state, backward velocity state, forward position state and backward position state into effective stitching sampling points, and then combining the continuous effective stitching sampling points and their corresponding forward and backward states to generate the trajectory stitching region; Within the trajectory stitching region, the difference between the positive attitude state and the reverse attitude state is extracted to generate an attitude deviation sequence, the difference between the positive velocity state and the reverse velocity state is extracted to generate a velocity deviation sequence, and the difference between the positive position state and the reverse position state is extracted to generate a position deviation sequence. Specifically, within the trajectory stitching region, the attitude state in the forward same coordinate trajectory and the attitude state in the backward same coordinate trajectory are calculated to generate an attitude deviation sequence; the velocity state in the forward same coordinate trajectory and the velocity state in the backward same coordinate trajectory are calculated to generate a velocity deviation sequence; the position state in the forward same coordinate trajectory and the position state in the backward same coordinate trajectory are calculated to generate a position deviation sequence, and the attitude deviation sequence, velocity deviation sequence, and position deviation sequence are kept to correspond to the sampling time of the same time coordinate. The attitude deviation sequence, velocity deviation sequence and position deviation sequence are synchronously arranged according to the same time coordinate to generate a three-layer stitched deviation sequence. Then, the continuously increasing deviation segment, the deviation direction-maintaining segment and the deviation amplitude abrupt segment are identified from the three-layer stitched deviation sequence. The identification of continuously increasing deviation segments, deviation direction-maintaining segments, and deviation amplitude abrupt change segments specifically includes: calculating the deviation increment, deviation direction change, and deviation amplitude change between adjacent sampling points for attitude deviation sequences, velocity deviation sequences, and position deviation sequences respectively; calculating the mean of the stable segment of deviation amplitude change within the segment to generate a stable change benchmark; then subtracting the deviation amplitude change of each sampling segment from the stable change benchmark to generate abrupt deviation amount; and generating deviation amplitude abrupt change segments according to the continuous clustering positions of the abrupt deviation amount on the time coordinate. Features of continuously increasing deviation segments, deviation direction-maintaining segments, and deviation amplitude abruptly changing segments are summarized to generate solution stitch crack features. The solution stitch crack features include attitude deviation, velocity deviation, position deviation, deviation change direction, deviation change amplitude, and deviation continuity. The generation of the solution suture crack features specifically includes: time-overlapping and organizing the continuously growing deviation segments, deviation direction-maintaining segments, and deviation amplitude abrupt change segments; organizing the sampling segments that simultaneously have continuous growth and direction-maintaining characteristics into the main crack interval; organizing the sampling positions of the corresponding deviation amplitude abrupt change segments within the main crack interval into crack expansion positions; and then summarizing the main crack interval, crack expansion position, attitude deviation, velocity deviation, position deviation, deviation change direction, and deviation change amplitude to generate the solution suture crack features.

[0024] In this embodiment, the generation of the navigation closed-loop conservation violation migration chain includes: Extract triaxial angular velocity data, triaxial acceleration data, attitude calculation data, velocity calculation data, position calculation data, and timestamps corresponding to each sampled data from standard inertial navigation operation data, and generate a sampling time series in ascending order of timestamps; The three-axis angular velocity data, three-axis acceleration data and velocity solution data are integrated and recursively processed according to the sampling time series. The corresponding integrated recursive processing results are compared with the attitude solution data, velocity solution data and position solution data in closed loop to generate attitude closed loop broken sequence, velocity closed loop broken sequence and position closed loop broken sequence. The attitude closed-loop failure sequence, velocity closed-loop failure sequence, and position closed-loop failure sequence are synchronously arranged according to the sampling time series, and the failure start time, failure duration interval, failure change direction, and failure amplitude change are marked to generate a closed-loop failure state sequence. The generation of the closed-loop fault state sequence specifically includes: synchronizing the attitude closed-loop fault sequence, velocity closed-loop fault sequence, and position closed-loop fault sequence; averaging the stationary segments of each closed-loop comparison difference sequence to generate a closed-loop fluctuation benchmark; subtracting the closed-loop comparison difference from the closed-loop fluctuation benchmark to generate the closed-loop fault deviation; generating the fault start time and fault duration interval according to the continuous clustering position of the closed-loop fault deviation in the sampling time sequence; and calculating the direction and magnitude of the difference change within the fault duration interval to generate the closed-loop fault state sequence. Based on the closed-loop broken state sequence, temporal adjacency matching is performed on the attitude closed-loop broken sequence and the velocity closed-loop broken sequence to select continuous intervals where the attitude closed-loop broken occurs first, the velocity closed-loop broken continues afterward, and the direction of the broken change is consistent, thus generating candidate transmission segments from attitude to velocity. The generation of the attitude-to-velocity candidate transmission segment specifically includes: reading the failure start time, failure duration interval, and failure change direction of the attitude closed loop failure from the closed loop failure state sequence, and reading the failure start time, failure duration interval, and failure change direction of the velocity closed loop failure; pairing the attitude closed loop failure and the velocity closed loop failure adjacently according to the sampling time order; and organizing the continuous sampling segments in which the attitude closed loop failure occurs first, the velocity closed loop failure continues after it, and the failure change directions of the two are consistent into attitude-to-velocity candidate transmission segments. Based on the closed-loop broken state sequence, temporal adjacency matching is performed on the velocity closed-loop broken sequence and the position closed-loop broken sequence to screen continuous intervals in which velocity closed-loop broken occurs first, position closed-loop broken continues afterward, and the amplitude of the brokenness changes continuously, thereby generating candidate velocity-to-position transmission segments. The generation of the velocity-to-position candidate transmission segment specifically includes: reading the failure start time, failure duration interval, and failure amplitude change of velocity closed loop failure from the closed loop failure state sequence, and reading the failure start time, failure duration interval, and failure amplitude change of position closed loop failure; pairing velocity closed loop failure and position closed loop failure adjacently according to the sampling time order; and organizing the sampling segments in which velocity closed loop failure occurs first, position closed loop failure continues after it, and the failure amplitudes of the two segments show a continuous increasing relationship as velocity-to-position candidate transmission segments. The attitude closed-loop failure sequence, attitude-to-velocity candidate transmission segments, and velocity-to-position candidate transmission segments are time-series processed, and closed-loop failure segments with conflicting failure time order, conflicting failure change direction, and broken failure duration intervals are removed to generate the target failure transmission segment. The generation of the target fault transmission segment specifically includes: performing time-connection and organization on the attitude-to-velocity candidate transmission segment and the velocity-to-position candidate transmission segment, connecting the candidate transmission segments that can be continuously connected on the sampling time series, and after connecting, organizing the fault nodes according to the time sequence of attitude closed-loop fault, velocity closed-loop fault and position closed-loop fault, and simultaneously retaining the fault change direction and fault duration interval corresponding to each fault node to form the target fault transmission segment; Determine the failure start time and failure end time from the target failure transmission segment, and determine the failure transmission time when there is cross-layer transmission in the target failure transmission segment. Generate a navigation closed-loop conservation failure migration chain according to the closed-loop failure sequence contained in the target failure transmission segment. The generation of the navigation closed-loop conservation fault migration chain specifically includes: reading the earliest fault start time, the fault conduction time corresponding to cross-layer conduction, and the fault termination time corresponding to the terminal fault from the target fault conduction segment; taking the closed-loop fault sequence corresponding to the fault start time as the chain start point; taking the closed-loop fault sequence corresponding to the fault conduction time as the chain intermediate node; and taking the closed-loop fault sequence corresponding to the fault termination time as the chain end point; and connecting the chain start point, chain intermediate node, and chain end point in the order of sampling time to generate the navigation closed-loop conservation fault migration chain.

[0025] In this embodiment, the generation of the credible error hypothesis chain includes: An error hypothesis set is generated based on the characteristics of the stitched crack, the navigation closed-loop conservation failure migration chain, and the device error state sequence. The error hypothesis set includes the gyroscope zero bias dominant hypothesis, the accelerometer zero bias dominant hypothesis, the temperature response dominant hypothesis, the vibration response dominant hypothesis, and the mounting angle offset dominant hypothesis. The generation of the error hypothesis set specifically includes: first, reading the deviation level, deviation change direction, and deviation change amplitude from the solved stitch crack features; then, reading the failure propagation sequence from the navigation closed-loop conservation failure migration chain; and finally, reading the gyroscope zero-bias state, accelerometer zero-bias state, temperature response state, vibration response state, and mounting angle offset state from the device error state sequence. The above information is combined into the gyroscope zero-bias dominant hypothesis, accelerometer zero-bias dominant hypothesis, temperature response dominant hypothesis, vibration response dominant hypothesis, and mounting angle offset dominant hypothesis, respectively. Each error hypothesis includes the hypothesis source, corresponding deviation level, corresponding failure node, and corresponding device error state. Each error hypothesis in the set of error hypotheses is mapped to the hypothetical attitude deviation, hypothetical velocity deviation, hypothetical position deviation, hypothetical deviation change direction, hypothetical deviation change magnitude, hypothesis failure start time, hypothesis failure propagation time, hypothesis failure termination time, and hypothesis failure duration interval, thereby generating the hypothesis interpretation state; The interpretation state is compared with the characteristics of the solved suture crack and the conservation failure of the navigation closed loop. Error assumptions that cannot match the characteristics of the solved suture crack and the conservation failure of the navigation closed loop are eliminated, and a set of interpretation consistent error assumptions is generated. The generation of the interpretation consistency error hypothesis set specifically includes: during consistency comparison, the hypothetical attitude deviation, hypothetical velocity deviation, and hypothetical position deviation in the hypothesis interpretation state are respectively matched with the attitude deviation, velocity deviation, and position deviation in the solved suture crack features. Then, the direction of change of the hypothesis deviation and the magnitude of change of the hypothesis deviation are matched with the direction of change of the deviation and the magnitude of change of the deviation in the solved suture crack features. The hypothesis failure start time, hypothesis failure propagation time, and hypothesis failure termination time are matched with the corresponding failure time in the navigation closed-loop conservation failure migration chain. The error hypotheses that have completed the corresponding hierarchy, direction, magnitude, and time sequence are organized into the interpretation consistency error hypothesis set. For each error hypothesis in the consistent error hypothesis set, reverse device verification is performed. Each error hypothesis is backtracked to the assumed device error state, and the assumed device error state is compared with the corresponding device error state in the device error state sequence. Error hypotheses that cannot match the corresponding device error state are eliminated, and a consistent error hypothesis set is generated. The reverse device verification specifically includes: during reverse device verification, mapping each error hypothesis in the interpretation consistency error hypothesis set to the corresponding device error state, such as the gyroscope zero bias dominant hypothesis corresponding to the gyroscope zero bias state, the accelerometer zero bias dominant hypothesis corresponding to the accelerometer zero bias state, the temperature response dominant hypothesis corresponding to the temperature response state, the vibration response dominant hypothesis corresponding to the vibration response state, and the mounting angle offset dominant hypothesis corresponding to the mounting angle offset state. Then, reading the change process of the device error state before the failure start time along the sampling time in reverse, forming a device leader change segment, and corresponding and organizing the device leader change segment with the hypothesis deviation change direction and hypothesis failure start time in the error hypothesis to generate a verification consistency error hypothesis set. A credible error hypothesis chain is generated by chaining each error hypothesis in the consistent error hypothesis set with the corresponding failure start time, failure end time, and failure propagation time during cross-layer propagation in the navigation closed-loop conservation failure migration chain. The generation of the credible error hypothesis chain specifically includes: sorting each error hypothesis in the verification consistency error hypothesis set according to its corresponding failure start time, failure propagation time, and failure termination time; when multiple error hypotheses correspond to the same failure interval, arranging them according to the degree of correspondence between them and the solution stitch crack feature, the navigation closed-loop conservation failure migration chain, and the device error state sequence; and then connecting the sorted error hypotheses in sequence to generate a credible error hypothesis chain. The degree of correspondence is jointly generated by the number of corresponding deviation levels, the number of corresponding failure timings, and the number of corresponding device leader changes. First, the number of corresponding errors between each error hypothesis and the solution stitch crack feature, the navigation closed-loop conservation failure migration chain, and the device error state sequence is counted separately; then, the error hypotheses within the same failure interval are arranged in descending order of the number of corresponding errors.

[0026] In this embodiment, the generation of the inertial navigation device error prediction result includes: From the solved suture crack features, attitude deviation, velocity deviation, position deviation, deviation change direction and deviation change magnitude are extracted, and future error types and error growth stages are determined; The determination of future error types and error growth stages specifically includes: reading the order of occurrence and continuous change of attitude deviation, velocity deviation and position deviation from the solved stitch crack features; taking the deviation level that first continues to expand as the leading level of future error types; generating attitude error expansion type when attitude deviation first continues to expand; generating velocity error expansion type when velocity deviation subsequently continues to expand; generating position error expansion type when position deviation subsequently continues to expand; and generating error growth stage by the degree of deviation between the deviation change amplitude and the stable deviation change benchmark within the inertial navigation solution segment. Extract the failure start time, failure end time, and failure propagation time during cross-layer propagation from the navigation closed-loop conservation failure migration chain, and determine the range of error influence; The determination of the error impact range specifically includes: reading the names of the connected closed-loop failure sequences and the order of failure nodes from the navigation closed-loop conservation failure migration chain; when the chain only connects the attitude closed-loop failure sequence, the error impact range is generated as the attitude layer impact range; when the chain connects the attitude closed-loop failure sequence and the velocity closed-loop failure sequence, the error impact range is generated as the attitude-velocity joint impact range; when the chain connects the attitude closed-loop failure sequence, the velocity closed-loop failure sequence, and the position closed-loop failure sequence, the error impact range is generated as the attitude-velocity-position joint impact range. Read the error hypotheses, their chain order, and the error source types corresponding to the error hypotheses from the credible error hypothesis chain, and determine the error source types and prediction confidence. The determination of prediction confidence specifically includes: reading the top-ranked error hypotheses and their corresponding device error states from the confidence error hypothesis chain, taking the source of the error hypothesis as the error source type, and then counting the number of object correspondences between the confidence error hypothesis chain and the solution stitch crack feature, the navigation closed-loop conservation failure migration chain, and the device error state sequence. A high confidence result is generated when the confidence error hypothesis chain corresponds to all three types of objects, a medium confidence result is generated when it corresponds to two types of objects, and error hypotheses that correspond to only one type of object are transferred to the unadopted record and do not participate in the association encapsulation of the inertial navigation device error prediction results. The error prediction results for inertial navigation devices are generated by associating and encapsulating the future error type, error growth stage, error impact range, error source type, and prediction reliability.

[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to a continuous operation error prediction scenario for a vehicle-mounted inertial navigation device. The inertial navigation device is mounted on a mobile carrier, which continuously travels within a closed test environment. The operation includes straight-line driving, turning, acceleration and deceleration, passage through short-term vibration sections, and periods of unstable external positioning signals. In this scenario, the inertial navigation device needs to continuously output attitude, velocity, and position calculation results under conditions of insufficient external positioning reference. Existing methods typically require waiting until significant position drift occurs before error judgment can be made, making it difficult to promptly identify the process of attitude error expanding into velocity and position errors. It also makes it difficult to determine whether the error source is gyroscope bias, accelerometer bias, temperature response, vibration response, or installation angle offset, resulting in lag and insufficient interpretation of the error prediction results.

[0028] In this embodiment, the inertial navigation device collects its own operational data according to a set sampling period. The collected data includes three-axis angular velocity data, three-axis acceleration data, attitude calculation data, velocity calculation data, position calculation data, temperature data, vibration data, and timestamps corresponding to each sampled data. The system performs time synchronization, abnormal sampling removal, coordinate system unification, and scale normalization on its own operational data to generate standard inertial navigation operational data. When adjacent sampling times exceed the normal sampling interval, or when there are continuous missing values ​​in the attitude, velocity, and position calculation data, the data is segmented at the corresponding sampling position. The segmented data is used as an inertial navigation calculation fragment only when the sampling time is continuous, the attitude, velocity, and position calculation data are complete. In the sample data, the system forms a total of 36 valid inertial navigation calculation fragments, of which 12 fragments exhibit the phenomenon that the forward and backward calculation trajectories cannot be stably stitched together.

[0029] Within each inertial navigation (INS) calculation segment, the system reads the attitude, velocity, and position calculation data corresponding to the initial sampling time to generate the segment's initial calculation state, and reads the attitude, velocity, and position calculation data corresponding to the final sampling time to generate the segment's final calculation state. Subsequently, it reads the three-axis angular velocity data, three-axis acceleration data, and sampling time data from the data range corresponding to the same INS calculation segment, generating segment angular velocity sequences, segment acceleration sequences, and segment time interval sequences in ascending order of sampling time. The system aligns the segment's initial calculation state with the sequence's initial sampling time to generate a forward recursion starting point, and aligns the segment's final calculation state with the sequence's final sampling time to generate a reverse recursion starting point.

[0030] During forward inertial navigation recursive calculation, the system uses the attitude, velocity, and position at the starting point of the forward recursion as the initial solution state. It reads the segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence in ascending order of sampling time, sequentially updating the forward attitude state, forward velocity state, and forward position state. The forward attitude state, forward velocity state, and forward position state at each sampling time are then connected to generate the forward solution trajectory. During backward inertial navigation recursive calculation, the system uses the attitude, velocity, and position at the starting point of the backward recursion as the initial solution state. It reads the segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence in descending order of sampling time. The attitude change, velocity change, and position change are subtracted in reverse to generate the reverse attitude state, reverse velocity state, and reverse position state. These are then connected in ascending order of sampling time to generate the backward solution trajectory.

[0031] After the forward and backward calculated trajectories are generated, the system maps them to the same time coordinate to generate a trajectory stitching region. Within the trajectory stitching region, the system extracts the differences in attitude, velocity, and position states from the forward and backward calculated trajectories, generating attitude deviation sequences, velocity deviation sequences, and position deviation sequences, which are then simultaneously arranged into a three-layer stitching deviation sequence. Taking one running segment as an example, in the first part of the segment, the attitude deviation remains within 0.02 degrees, the velocity deviation within 0.015 meters per second, and the position deviation within 0.40 meters. After entering the turning and vibration superposition zone, the attitude deviation increases to 0.09 degrees, the velocity deviation increases to 0.061 meters per second, and the position deviation increases to 1.72 meters. Based on this change, the system identifies segments with continuously increasing deviations, segments with maintained deviation direction, and segments with abrupt changes in deviation amplitude, and generates the characteristics of the calculated stitching crack.

[0032] To determine whether errors have propagated hierarchically, the system performs attitude, velocity, and position closed-loop comparisons based on standard inertial navigation data. The system integrates and recursively processes the three-axis angular velocity, three-axis acceleration, and velocity calculation data, and then performs closed-loop comparisons with the attitude, velocity, and position calculation data respectively, generating attitude, velocity, and position closed-loop failure sequences. After synchronizing the three types of closed-loop failure sequences, the system marks the failure start time, failure duration interval, failure direction, and failure amplitude changes, generating a closed-loop failure state sequence. Furthermore, it generates candidate transmission segments from attitude to velocity, velocity to position, and target failure transmission segments, forming a navigation closed-loop conservation failure transfer chain.

[0033] In error source analysis, the system generates an error hypothesis set based on the characteristics of the solved stitch crack, the navigation closed-loop conservation failure migration chain, and the device error state sequence. Each error hypothesis is mapped to a hypothesis interpretation state. The system compares the hypothesis interpretation state with the characteristics of the solved stitch crack and the navigation closed-loop conservation failure migration chain, retaining the error hypotheses that can match the deviation level, deviation direction, and failure propagation sequence. Subsequently, reverse device verification is performed, pushing each error hypothesis back to the hypothetical device error state and comparing it with the corresponding device error state in the device error state sequence to generate a verified consistent error hypothesis set, thus forming a credible error hypothesis chain.

[0034] Sample results show that traditional position drift threshold judgment methods only provide anomaly warnings after a significant increase in position deviation, while this invention can generate error prediction results during the initial expansion stages of attitude and velocity deviations. In 36 valid inertial navigation solution segments, the traditional method outputs valid anomaly warnings in 8 segments, while this invention outputs error prediction results in 12 segments. Of these, 10 segments continued to show increased velocity or position deviations in subsequent runs. Regarding error source judgment, the system outputs gyroscope zero bias dominance in 7 segments, vibration response amplification in 3 segments, and installation angle offset influence in 2 segments. After reverse verification of the device error state sequence, the gyroscope zero bias state, vibration response state, and installation angle offset state all show a consistent relationship with the error change direction of the corresponding segments.

[0035] As can be seen from the above sample data, the present invention can identify early changes in attitude and velocity deviations by solving the stitching crack features before the position drift significantly expands, and determine whether the error continues to propagate to the position layer by using the navigation closed-loop conservation migration chain. At the same time, the credible error hypothesis chain can correspond the error prediction results with the device error state sequence, so that the error prediction results of the inertial navigation device can not only give the future error type, error growth stage and error impact range, but also output the error source type and prediction credibility, thereby improving the timeliness, interpretability and reliability of error prediction.

[0036] Table 1. Comparison of Overall Performance Percentage of Error Prediction for Inertial Navigation Equipment

[0037] In Table 1, the correct early identification rate and the missed identification rate are based on the inertial navigation solution segment where error expansion actually occurs, while the false alarm rate is based on the inertial navigation solution segment where continuous error expansion does not occur. The attitude error prediction compliance rate, velocity error prediction compliance rate, and position error prediction compliance rate are used to characterize the proportion of each method's prediction results for attitude error, velocity error, and position error falling within the allowable deviation range, respectively. The error source interpretation consistency rate is used to characterize the degree of consistency between the predicted error source type and the device error state sequence.

[0038] As shown in Table 1, the correct early identification rate of the position drift threshold judgment method is 58.33%, and the missed identification rate is 41.67%. This indicates that the method mainly triggers judgment after the position drift has significantly expanded, and is not sensitive to the early expansion of attitude deviation and velocity deviation. Its attitude error prediction compliance rate, velocity error prediction compliance rate, and position error prediction compliance rate are 52.78%, 55.56%, and 61.11%, respectively. Although it can make some judgments on segments that have already shown position drift, it is not capable of explaining early changes in error and error sources. The consistency rate of error source explanation is only 16.67%.

[0039] The correct early identification rate of the zero-bias compensation and residual threshold method was 66.67%, which was higher than that of the position drift threshold method. This indicates that by correcting the zero bias of the gyroscope and the zero bias of the accelerometer, it can identify the error expansion trend caused by some device errors in advance. The accuracy rates of attitude error prediction, velocity error prediction and position error prediction were 66.67%, 69.44% and 72.22%, respectively. The consistency rate of error source explanation was 50.00%, indicating that the method has a certain explanatory ability for zero-bias errors. However, it is insufficient in characterizing vibration response, installation angle offset and the transmission relationship of attitude error to velocity error and position error.

[0040] The recurrent neural network error prediction method achieved a correct early identification rate of 75.00% and a false negative rate of 25.00%, indicating that the prediction method based on historical time-series data can capture some error growth trends. The method achieved accuracy rates of 75.00%, 77.78%, and 80.56% for attitude error prediction, velocity error prediction, and position error prediction, respectively, demonstrating good error trend prediction capabilities. However, its false alarm rate reached 16.67%, mainly because it easily identifies short-term disturbances caused by turning, short-term vibrations, or unstable external positioning references as continuous error expansion, and lacks reverse verification with the device error state sequence. Therefore, the consistency rate of error source explanation was only 23.08%.

[0041] The method of this invention achieves a correct early identification rate of 83.33%, reduces the missed identification rate to 16.67%, and maintains a false alarm rate of 8.33%, outperforming other comparative methods overall. Its attitude error prediction accuracy rate is 86.11%, velocity error prediction accuracy rate is 88.89%, and position error prediction accuracy rate is 91.67%, indicating that this invention can more accurately identify error changes at the attitude, velocity, and position levels. Furthermore, the method of this invention achieves an error source interpretation consistency rate of 83.33%, demonstrating its ability to reliably output the error source type and prediction confidence.

[0042] The performance improvement of this invention lies in its approach: instead of relying solely on position drift thresholds, zero-bias residuals, or historical time-series trends for judgment, it generates solution stitching crack features through the unstable stitching differences between the forward and backward solution trajectories. This allows early manifestation of changes in attitude, velocity, and position deviations. Simultaneously, through temporal propagation analysis of attitude, velocity, and position closed-loop failure sequences, it generates a navigation closed-loop conservation failure migration chain, thereby determining whether the error is continuously expanding from the attitude layer to the velocity and position layers. Furthermore, this invention uses a credible error hypothesis chain to reverse-verify the prediction results with the device error state sequence, thus achieving better overall performance in early error identification, prediction accuracy, and error source explanation.

[0043] The above are merely preferred embodiments 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. A method for predicting errors in inertial navigation devices based on data analysis, characterized in that, Includes the following steps: Collect the inertial navigation device's own operational data during operation, and preprocess the data to generate standard inertial navigation operational data; The standard inertial navigation system (INS) operating data is divided into continuous segments to generate an INS solution segment set. The solution state at the beginning and end of each segment is extracted from the INS solution segment. Device errors are extracted from the standard INS operating data to generate a device error state sequence. Perform forward time inertial navigation recursive calculation and backward time inertial navigation recursive calculation on the same inertial navigation solution segment to generate forward and backward solution trajectories; The forward and backward calculation trajectories corresponding to the same inertial navigation calculation segment are mapped to the same time coordinate, and attitude deviation, velocity deviation and position deviation are extracted to generate the calculation stitch crack features; Based on standard inertial navigation operation data, attitude closed-loop comparison, velocity closed-loop comparison, and position closed-loop comparison are performed, and time-series propagation analysis is conducted to generate navigation closed-loop conservation failure transfer chain. Based on the characteristics of the stitching crack, the navigation closed-loop conservation failure migration chain, and the device error state sequence, an error hypothesis set is generated, and forward error interpretation processing and reverse device verification are performed on each error hypothesis to generate a credible error hypothesis chain. Error prediction results for inertial navigation devices are generated based on the characteristics of the suture crack, the navigation closed-loop conservation failure migration chain, and the credible error hypothesis chain.

2. The method for predicting errors in inertial navigation devices based on data analysis according to claim 1, characterized in that, The preprocessing includes time synchronization, outlier removal, coordinate system unification, and scale normalization.

3. The method for predicting errors in inertial navigation devices based on data analysis according to claim 1, characterized in that, The generation of the device error state sequence includes: Standard inertial navigation operation data is read in the order of sampling time. Based on the continuity of sampling time and the integrity of the solution state, the standard inertial navigation operation data is divided into continuous segments. Each continuous segment is used as an inertial navigation solution segment to generate an inertial navigation solution segment set. Read the attitude, velocity, and position data corresponding to the initial sampling time of each inertial navigation solution segment from the inertial navigation solution segment set to generate the segment start solution state, and read the attitude, velocity, and position data corresponding to the end sampling time to generate the segment end solution state; Zero-bias state extraction is performed on the triaxial angular velocity and triaxial acceleration data in the standard inertial navigation system (INS) operating data. Response state extraction is performed on the temperature data, vibration data, and attitude calculation data in the standard INS operating data to generate a device error state sequence.

4. The method for predicting errors in inertial navigation devices based on data analysis according to claim 1, characterized in that, The generation of the forward and backward solution trajectories includes: Read triaxial angular velocity data, triaxial acceleration data, and sampling time data from the data range corresponding to the same inertial navigation solution segment, and generate segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence in ascending order of sampling time; Align the initial solution state of the segment with the initial sampling time of the segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence to generate a forward recursion starting point; align the final solution state of the segment with the final sampling time of the segment angular velocity sequence, segment acceleration sequence, and segment time interval sequence to generate a reverse recursion starting point. Starting from the forward recursion starting point, perform forward time inertial navigation recursion calculation on the same inertial navigation solution segment, and generate forward attitude state, forward velocity state and forward position state at each sampling time. The forward attitude state, forward velocity state, and forward position state are connected in ascending order of sampling time to generate the forward calculated trajectory; Starting from the reverse recursion starting point, time-reverse inertial navigation recursion is performed on the same inertial navigation solution segment, generating reverse attitude state, reverse velocity state and reverse position state at each sampling time. The reverse attitude state, reverse velocity state, and reverse position state are connected in ascending order of sampling time to generate the backward calculated trajectory.

5. The method for predicting errors in inertial navigation devices based on data analysis according to claim 1, characterized in that, The generation of the solution for the suture crack features includes: Using the sampling time of the same inertial navigation solution segment as the time reference, the sampling times of the forward solution trajectory and the backward solution trajectory are aligned to generate the same time coordinates. The forward solution trajectory and the backward solution trajectory are then mapped to the same time coordinates to generate forward and backward trajectories with the same coordinates. Pair the forward and backward trajectories with the same time coordinate at each sampling time to generate a trajectory stitching region; Within the trajectory stitching region, the difference between the positive attitude state and the reverse attitude state is extracted to generate an attitude deviation sequence, the difference between the positive velocity state and the reverse velocity state is extracted to generate a velocity deviation sequence, and the difference between the positive position state and the reverse position state is extracted to generate a position deviation sequence. The attitude deviation sequence, velocity deviation sequence and position deviation sequence are synchronously arranged according to the same time coordinate to generate a three-layer stitched deviation sequence. Then, the continuously increasing deviation segment, the deviation direction-maintaining segment and the deviation amplitude abrupt segment are identified from the three-layer stitched deviation sequence. The features of continuously increasing deviation segments, deviation direction-maintaining segments, and deviation amplitude abruptly changed segments are summarized to generate the features of the suture crack solution.

6. The method for predicting errors in inertial navigation devices based on data analysis according to claim 1, characterized in that, The generation of the navigation closed-loop conservation violation migration chain includes: Extract triaxial angular velocity data, triaxial acceleration data, attitude calculation data, velocity calculation data, position calculation data, and timestamps corresponding to each sampled data from standard inertial navigation operation data, and generate a sampling time series in ascending order of timestamps; The three-axis angular velocity data, three-axis acceleration data and velocity solution data are integrated and recursively processed according to the sampling time series, and the corresponding integrated and recursive processing results are compared in closed loop to generate attitude closed loop broken sequence, velocity closed loop broken sequence and position closed loop broken sequence. The attitude closed-loop failure sequence, velocity closed-loop failure sequence, and position closed-loop failure sequence are synchronously arranged according to the sampling time series, and the failure start time, failure duration interval, failure change direction, and failure amplitude change are marked to generate a closed-loop failure state sequence. Based on the closed-loop broken state sequence, temporal adjacency matching is performed on the attitude closed-loop broken sequence and the velocity closed-loop broken sequence to filter and generate candidate attitude-to-velocity transmission segments. Based on the closed-loop broken state sequence, temporal adjacency matching is performed on the velocity closed-loop broken sequence and the position closed-loop broken sequence to filter and generate candidate velocity-to-position transmission segments. The attitude closed-loop failure sequence, attitude-to-velocity candidate transmission segments, and velocity-to-position candidate transmission segments are time-series processed, and closed-loop failure segments with conflicting failure time order, conflicting failure change direction, and broken failure duration intervals are removed to generate the target failure transmission segment. Determine the failure start time and failure end time from the target failure transmission segment, and determine the failure transmission time when there is cross-layer transmission in the target failure transmission segment. Generate a navigation closed-loop conservation failure migration chain according to the closed-loop failure sequence contained in the target failure transmission segment.

7. The method for predicting errors in inertial navigation devices based on data analysis according to claim 1, characterized in that, The generation of the credible error hypothesis chain includes: An error hypothesis set is generated based on the characteristics of the stitched crack, the navigation closed-loop conservation failure migration chain, and the device error state sequence. Each error hypothesis in the set of error hypotheses is mapped to the hypothetical attitude deviation, hypothetical velocity deviation, hypothetical position deviation, hypothetical deviation change direction, hypothetical deviation change magnitude, hypothesis failure start time, hypothesis failure propagation time, hypothesis failure termination time, and hypothesis failure duration interval, thereby generating the hypothesis interpretation state; A consistency comparison is performed between the hypothesized explanation state and the solved stitch crack features and the navigation closed-loop conservation failure migration chain to generate an explanation consistency error hypothesis set; For each error hypothesis in the consistent error hypothesis set, reverse device verification is performed. Each error hypothesis is backtracked to the assumed device error state, and the assumed device error state is compared with the corresponding device error state in the device error state sequence to generate a consistent error hypothesis set. A credible error hypothesis chain is generated by chaining the error hypotheses in the consistent error hypothesis set with the corresponding failure start time, failure end time, and failure propagation time during cross-layer propagation in the navigation closed-loop conservation failure migration chain.

8. The method for predicting errors in inertial navigation devices based on data analysis according to claim 1, characterized in that, The generation of the inertial navigation device error prediction results includes: From the solved suture crack features, attitude deviation, velocity deviation, position deviation, deviation change direction and deviation change magnitude are extracted, and future error types and error growth stages are determined; Extract the failure start time, failure end time, and failure propagation time during cross-layer propagation from the navigation closed-loop conservation failure migration chain, and determine the range of error influence; Read the error hypotheses, their chain order, and the error source types corresponding to the error hypotheses from the credible error hypothesis chain, and determine the error source types and prediction confidence. The error prediction results for inertial navigation devices are generated by associating and encapsulating the future error type, error growth stage, error impact range, error source type, and prediction reliability.

Images