Navigation system calibration method and apparatus, electronic device, and storage medium

Abstract

The application relates to a navigation system calibration method and device, electronic equipment and a storage medium. The method comprises the following steps: starting a multi-source navigation positioning device to collect multi-source original data; screening an effective measurement point set meeting a slant range difference constraint condition; if a current track in the effective measurement point set is in a slant range constraint invalidation scene, a measurement vector and an observation equation are constructed by using the effective measurement point set; if not, attention weights of each measurement dimension are calculated by using an attention weight function, and the observation equation is constructed by weighting; a K-means clustering initialization algorithm based on deep reinforcement learning optimization is run to update a state in a filtering iteration; if the state converges, installation error parameters are output, and system installation errors are compensated; if the state does not converge, a time synchronization and interpolation completion link of the multi-source original data is returned to continue iteration until a calibration process is completed. The application can improve underwater navigation anti-interference capability, realize high-precision calibration, widen an adaptation range, and achieve a high-quality navigation calibration target.

Description

Technical Field

[0001] This application relates to the field of navigation calibration technology, and in particular to a navigation system calibration method, apparatus, electronic device and storage medium. Background Technology

[0002] Navigation systems composed of multiple navigation and positioning devices have been widely used in navigation and positioning tasks for marine equipment such as underwater robots and submarines due to their outstanding advantages of high positioning accuracy and flexible deployment. In this system, the installation error between the acoustic array of the Ultra-Short Baseline (USBL) positioning system and the carrier platform is a key factor restricting the improvement of navigation accuracy. Even a small error in the heading angle can lead to a significant positioning deviation. Therefore, this installation error must be accurately calibrated before the system is deployed.

[0003] Currently, existing calibration methods mainly include least squares method, numerical iteration method and filtering method, but these methods generally suffer from problems such as susceptibility to equipment motion interference, insufficient calibration stability and limited application scenarios, making it difficult to meet diverse navigation needs.

[0004] Therefore, existing technologies struggle to simultaneously achieve the application goals of strong anti-interference capability, high calibration accuracy, and wide adaptability. Consequently, developing a navigation system calibration method that is highly adaptable, has good anti-interference performance, and provides accurate calibration has become a pressing technical problem in the field of underwater navigation technology. Summary of the Invention

[0005] In view of the above, this application provides a navigation system calibration method, apparatus, electronic device and storage medium, the purpose of which is to solve the above-mentioned technical problems.

[0006] In a first aspect, this application provides a navigation system calibration method, the method comprising: Start the multi-source navigation and positioning equipment to collect raw data from multiple sources; The multi-source raw data is time-synchronized and interpolated to complete the data, and the effective set of measurement points that meet the slope distance difference constraint is selected. Determine whether the current trajectory in the set of valid measurement points is in a slant range constraint failure scenario; If so, the effective set of measurement points is used directly to construct the measurement vector and observation equation; otherwise, the attention weights of each measurement dimension are calculated using the attention weight function and the observation equation is constructed by weighting them. The initial value obtained by the K-means clustering initialization algorithm optimized by deep reinforcement learning is used as the starting point, and the state is updated by running filtering iterations. Determine if the state has converged. If it has converged, output the installation error parameters and compensate for the system installation error. If it has not converged, return to the time synchronization and interpolation completion stage of the multi-source raw data to continue iterating until the calibration process is completed.

[0007] Secondly, this application provides a navigation system calibration apparatus, the apparatus comprising modules of the navigation system calibration method described in any embodiment of the first aspect.

[0008] Thirdly, this application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When the processor executes a program stored in the memory, it implements the steps of the navigation system calibration method according to any embodiment of the first aspect.

[0009] Fourthly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the navigation system calibration method as described in any embodiment of the first aspect.

[0010] Compared with the prior art, the technical solutions provided in this application have the following advantages: The technical solutions of this application can significantly improve the anti-interference capability of the underwater navigation system, ensure high-precision calibration in complex environments such as time-varying noise and equipment movement, effectively broaden the system's adaptability range, successfully solve the failure problem of traditional calibration methods in special trajectory scenarios, and ultimately achieve the underwater navigation calibration goal of not requiring strict clock synchronization, strong anti-interference capability, high calibration accuracy and wide adaptability. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating a preferred embodiment of the navigation system calibration method of this application; Figure 2 This is a flowchart illustrating another preferred embodiment of the navigation system calibration method of this application; Figure 3This is a schematic diagram of a preferred embodiment of the navigation system calibration device of this application; Figure 4 This is a schematic diagram of a preferred embodiment of the electronic device of this application; The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0015] It should be noted that the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of those features. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed in this application.

[0016] Reference Figure 1 The diagram shown is a flowchart illustrating an embodiment of the navigation system calibration method of this application. The method is executed by an electronic device, which can be implemented by a software system and / or a hardware system. The navigation system calibration method includes: S101: Start the multi-source navigation and positioning device to collect multi-source raw data.

[0017] Here, multi-source navigation and positioning equipment is a combination of various navigation and positioning devices, including but not limited to Strapdown Inertial Navigation System (SINS), Ultra-Short Baseline Positioning System (USBL), and Real-Time Kinematic Global Positioning System (RTK-GPS). The core function of SINS is to provide continuous output of the carrier's attitude (roll, pitch, heading), velocity, and position; it is the "core of autonomous navigation" and does not rely on external signals. The core function of USBL is to measure the azimuth and slant range of underwater targets through acoustic signals; it is the "external positioning auxiliary module." The core function of RTK-GPS is to provide a high-precision absolute position reference for calibrating the accumulated errors of SINS; it is the "position reference source."

[0018] In addition, the multi-source raw data includes, but is not limited to, SINS attitude, velocity, and position data, USBL acoustic azimuth and signal reception time data, RTK-GPS high-precision position data, etc.

[0019] In practice, multi-source navigation and positioning devices such as SINS and USBL are activated to simultaneously collect raw measurement data (i.e., multi-source raw data) required for underwater navigation, such as azimuth angle and slant range.

[0020] S102: Perform time synchronization and interpolation completion on the multi-source raw data, and filter out the effective set of measurement points that meet the slant distance difference constraint.

[0021] In practice, timestamp alignment is used to synchronize the time of multi-source raw data. Missing values ​​encountered during data acquisition are filled in using linear interpolation to ensure the continuity and integrity of the data sequence, providing a complete dataset with a unified time reference for subsequent calculations. Furthermore, after preprocessing such as synchronization and interpolation completion, valid measurement points that meet the slant range difference constraint are selected from the preprocessed multi-source raw data. This process filters out invalid data, allowing only valid data that meets the constraints to enter the filtering process. This avoids interference from invalid data in subsequent filtering, reduces the risk of filter divergence, and provides high-quality input data for the attention-enhanced filtering algorithm, ensuring the accuracy and stability of the SINS / USBL navigation system installation error calibration.

[0022] For example, valid measurement points can be screened using a dual constraint of "trajectory center distance + continuous slope distance difference", or a constraint of "slope distance difference + transponder relative distance". This application does not specifically limit the methods used, as long as the screening of valid measurement points can be achieved.

[0023] S103: Determine whether the current trajectory in the set of valid measurement points is in a slant distance constraint failure scenario.

[0024] Here, the slope distance constraint failure scenario (also known as the scenario where the slope distances before and after are too close) specifically refers to a scenario where the change in slope distance between consecutive measurement points is extremely small, making it difficult for the traditional slope distance difference model to effectively solve the parameters.

[0025] Here, the "current trajectory" in this embodiment refers to a multi-source navigation and positioning data set characterizing the motion trajectory features of a carrier (such as an underwater robot or submarine). Its core function is to determine whether the current motion scenario meets the applicable conditions of the traditional slant range difference model, providing a scenario basis for the subsequent construction of observation equations. The trajectory data is not single-position data, but a combination of "position reference data + acoustic measurement correlation data," with each part having a clear source and function. Essentially, it is a combination of "carrier motion position + acoustic measurement correlation parameters." This is because in calibration scenarios, the applicability of the model needs to be judged by the "change in the relative position (slant range) between the carrier and the transponder," rather than solely relying on the shape of the carrier's own motion trajectory.

[0026] In other words, determining whether the current trajectory is in a slant range constraint failure scenario requires analysis of trajectory data from a set of valid measurement points. The core criterion is whether the slant range change at consecutive measurement points is minimal. This determination can accurately identify special failure scenarios such as the center of a circular trajectory, providing a basis for constructing observation equations for subsequent scenarios, avoiding calibration deviations caused by the failure of traditional slant range difference models, and expanding the trajectory applicability of SINS / USBL navigation system calibration.

[0027] S104: If so, the measurement vector and observation equation are directly constructed using the effective set of measurement points; otherwise, the attention weights of each measurement dimension are calculated using the attention weight function and the observation equation is constructed by weighting them.

[0028] Here, this embodiment of the application analyzes the trajectory data of the effective measurement point set to determine whether the current trajectory is a slope range constraint failure scenario (special scenario) with extremely small slope range change. Observation equations are then constructed accordingly. Specifically, non-failure scenarios (normal scenarios) are weighted and optimized, while failure scenarios are directly constructed using the original effective measurement point set. This approach overcomes the scenario limitations of traditional slope range difference models, broadens the scope of application, adapts to diverse trajectories, and avoids calibration failures in special scenarios.

[0029] The measurement vector is a quantized data set that integrates measured data from the USBL acoustic array in underwater navigation (such as the azimuth angle between the carrier and the transponder, and the acoustic signal propagation time) and derived parameters (such as the slant range difference). It is the core observation input reflecting the status of the navigation system. Its construction needs to be adapted to different scenarios: in failure scenarios, it is directly constructed using the original data of the effective measurement point set; in normal scenarios, it is constructed by combining an attention weight function and adaptively allocating weights according to the reliability of the data in each measurement dimension (such as the degree of deviation between the current measurement value and the historical effective measurement value). This not only links USBL and SINS multi-source data and provides a reliable basis for calculating installation errors, but also eliminates the strict dependence of traditional ranging on clock synchronization.

[0030] The observation equation is the core expression that establishes the mathematical mapping relationship between the measurement vector and the state to be estimated (installation angle error, lever error, etc.), quantifying the correlation between the navigation system state and the measured data. Its construction integrates trajectory features and an attention mechanism. In failure scenarios, it is directly constructed based on the original valid data. In conventional scenarios, it incorporates attention weights to optimize the measurement residual and noise covariance matrix. This not only solves the failure problem of traditional models in special scenarios but also provides optimized observation input for attention-enhanced filtering, supporting stable iteration in time-varying noise environments and ensuring calibration accuracy.

[0031] In addition, for the attention weighting mechanism in non-failure scenarios, the weight of high-confidence data is close to 1, and its deviation can fully dominate the state update; the weight of low-confidence data (including abnormal interference and noise data) is close to 0, and its impact on the state update is greatly weakened. This effectively filters out invalid interference factors, strengthens the role of reliable data, avoids filter divergence, and significantly improves the anti-interference capability of underwater navigation.

[0032] S105: Starting from the initial value obtained by the K-means clustering initialization algorithm optimized based on deep reinforcement learning, the state is updated by running filtering iteration.

[0033] In practical implementation, this step provides accurate initial values ​​for filtering iterations, ensuring calibration accuracy. Technically, a reinforcement learning framework can be constructed first, using initial measurement data features as the state and cluster center selection as the action. A reward function is designed based on the silhouette coefficient to train the agent. Then, the initial data is clustered using optimized K-means clustering, selecting the optimal cluster center as the initial value for filtering and adapting it to the covariance matrix to initiate filtering iterations. This approach avoids the local optima problem of traditional clustering, ensuring the initial value closely approximates the true state, thus providing accurate initial states, reducing filtering convergence bias, supporting stable convergence of underwater navigation filters, and guaranteeing high-precision calibration of installation errors.

[0034] K-means clustering is an unsupervised learning algorithm that iteratively divides data into K clusters, ensuring compactness within clusters and separation between clusters. Its core principle is optimizing cluster centers to minimize the distance error between data points and the centers. Deep reinforcement learning combines deep learning and reinforcement learning, using deep networks to fit value functions or policies. This allows agents to learn optimal policies through reward and punishment signals in environmental interactions, enabling decision-making for complex tasks. The silhouette coefficient is calculated from the average distance from each valid measurement point to its respective cluster center (intra-cluster compactness) and the minimum average distance to other cluster centers (inter-cluster separation).

[0035] S106: Determine whether the state has converged. If it has converged, output the installation error parameters and compensate for the system installation error. If it has not converged, return to the time synchronization and interpolation completion of the multi-source raw data to continue iterating until the calibration process is completed.

[0036] In practical implementation, the core purpose of this step is to verify the reliability of the state estimation during the filtering iteration, ensure accurate output of installation error parameters and complete system compensation, and avoid single-calibration deviations through the iterative mechanism. Technically, a convergence threshold can be preset (e.g., consecutive rounds of angle or lever arm error fluctuations below a set value). After filtering iteration, if the convergence threshold is met, convergence is determined, error parameters are output and written to the system compensation module; if not, the process returns to the time synchronization and interpolation completion stage of the multi-source raw data, re-filters valid data, and repeats the subsequent process until convergence.

[0037] In this way, in underwater navigation and positioning scenarios, stable installation error estimates are selected through convergence judgment, avoiding non-converged deviation parameters that are affected by time-varying noise and water flow interference. If convergence fails, the process returns to the preprocessing stage for iteration, correcting the accumulated errors from the initial data selection and model building, ensuring that the final parameters closely resemble the actual installation state, and achieving high-precision calibration.

[0038] The installation error parameter is a quantitative indicator describing the installation deviation between the USBL acoustic array and the carrier platform in the SINS / USBL navigation system. Its core components are angular error (deviation in heading, pitch, and roll) and lever error (positional deviation between the acoustic array and the origin of the carrier coordinate system). These are key parameters affecting positioning accuracy in underwater navigation. System installation error refers to the deviation in the SINS / USBL integrated navigation system caused by inaccurate installation of the USBL acoustic array and the carrier platform. It consists of angular error (leading to slant range positioning error) and lever error (causing coordinate conversion deviation), directly affecting the navigation and positioning accuracy of underwater robots, submarines, and other equipment, and requires calibration and compensation.

[0039] Here, one way to determine whether the state has converged is to calculate the change in the posterior state estimate over N consecutive iterations; if the change in all state parameters satisfies the condition that the change in the posterior state estimate is less than the convergence threshold, then the state is considered converged; otherwise, it is considered non-converged.

[0040] The convergence threshold can be determined based on the calibration accuracy requirements of the navigation system, typically using an angle error threshold of 0.1 degrees or a lever error threshold of 0.01 meters. N can be set between 10 and 20 to ensure stability. The change in the posterior state estimate is also considered.

[0041] In this embodiment, a multi-source navigation and positioning device is activated to collect multi-source raw data; a valid set of measurement points that meets the slant range difference constraint is selected; if the current trajectory in the valid set of measurement points is in a slant range constraint failure scenario, the valid set of measurement points is used to construct a measurement vector and observation equation; otherwise, the attention weights of each measurement dimension are calculated using an attention weight function and weighted to construct the observation equation; a K-means clustering initialization algorithm optimized by deep reinforcement learning is used to run filtering iterations to update the state; if the state converges, the installation error parameters are output and the system installation error is compensated; if the state does not converge, the process returns to the time synchronization and interpolation completion stage of the multi-source raw data to continue iterating until the calibration process is completed. Thus, this application can significantly improve the underwater navigation anti-interference capability, achieve high-precision calibration, broaden the adaptation range, and achieve the goal of high-quality navigation calibration.

[0042] Here, in this embodiment of the application, trajectory adaptive pre-processing data filtering can be performed through a two-step progressive constraint of trajectory center critical distance and continuous data difference. This ensures that the data participating in subsequent filtering satisfies both the "spatial conditions applicable to the slant range difference model" and the "temporal slant range constraint conditions," providing a reliable data foundation for subsequent accurate calibration. That is, in one possible implementation, Figure 2 This is a flowchart illustrating another preferred embodiment of the navigation system calibration method of this application; as shown below. Figure 2 As shown, the steps in S102 for selecting valid measurement points that satisfy the slope difference constraint include: S1021: Based on the trajectory design parameters of the current navigation task, determine the failure threshold of the slant range difference model, and calculate the target spatial distance between each current sample measurement point and the trajectory center.

[0043] In practice, the relevant parameters of the trajectory center critical distance are first calculated. Specifically, based on the trajectory design parameters of the current navigation task, the failure threshold of the slant range difference model (i.e., the critical distance of the trajectory center) is determined, and the spatial distance between each measurement point and the trajectory center is calculated. A failure zone determination radius is set. When the target spatial distance is less than or equal to the failure zone determination radius, the measurement point is determined to be invalid data (it is near the center of the circular trajectory, and the slant range difference model cannot be effectively constructed).

[0044] Here, trajectory design parameters are pre-defined quantitative indicators of trajectory characteristics for adapting to underwater navigation tasks, including trajectory type (such as circular or straight line), trajectory radius, and vehicle speed. Specifically, the product of the scene adaptation coefficient and the trajectory radius in the trajectory design parameters can be used as the failure threshold of the slant range difference model.

[0045] S1022: When the target spatial distance is greater than the failure area determination radius, the sample measurement point is used as a candidate measurement point.

[0046] In practice, when the target spatial distance is greater than the failure area determination radius, the next step of determination is initiated.

[0047] S1023: Calculate the slope distance difference between the candidate measurement point and its adjacent candidate measurement points. If the slope distance difference is greater than or equal to the determination threshold, then determine that the candidate measurement point is a valid measurement point that satisfies the slope distance difference constraint condition.

[0048] Furthermore, if the slope distance difference is less than the determination threshold, the candidate measurement point is determined to be an invalid measurement point that does not meet the slope distance difference constraint condition.

[0049] In practice, the slope distance difference of continuous data is calculated. Specifically, for the candidate measurement points that have passed the above screening, the slope distance difference between them and their adjacent measurement points is calculated. A judgment threshold is set. Invalid data (invalid measurement points) that do not meet the slope distance difference constraint are directly blocked; valid data (valid measurement points) are retained in the valid measurement point set for subsequent observation equation construction.

[0050] It should be noted that by quantifying the failure boundary through trajectory design parameters and performing layer-by-layer verification through a dual screening mechanism, invalid data in special scenarios such as the center of a circular trajectory can be accurately shielded, providing highly reliable input for subsequent filtering algorithms and ensuring the anti-interference capability and applicability of the SINS / USBL navigation system calibration.

[0051] In one possible implementation, the step of determining whether the current trajectory in the set of valid measurement points is in a slant range constraint failure scenario in S103 includes: Step 1031: Extract the slope distance data of all valid measurement points from the set of valid measurement points, and calculate the unbiased variance based on the slope distance data of all valid measurement points.

[0052] Step 1032: If the unbiased variance is less than the scene determination threshold, then the current trajectory is determined to be in a slant distance constraint failure scene.

[0053] Step 1033: If the unbiased variance is greater than or equal to the scene determination threshold, then the current trajectory is determined to be in a normal scene.

[0054] In practice, the slant range data of all valid measurement points are extracted from the set of valid measurement points and arranged into a slant range sequence in chronological order. The slant range of the valid measurement points is derived from the acoustic array measurement of the USBL system. Then, the unbiased variance is calculated based on this slant range sequence. If the unbiased variance is less than the scene determination threshold, it indicates that the overall change in slant range corresponding to the current trajectory is weak and the difference is not significant, and the current trajectory is determined to be in a slant range constraint failure scenario; otherwise, it is determined to be a normal scenario.

[0055] The scene determination threshold can be set according to the calibration accuracy requirements to adapt to the actual slant distance scale.

[0056] In one possible implementation, for a special scenario (slant range constraint failure scenario), the step in S104 of constructing the measurement vector and observation equation using the effective measurement point set includes: Step 1041a: Based on the multi-source raw data in the effective measurement point set, determine the measurement vector; the measurement vector includes the azimuth angle measured by the USBL acoustic array, the slant range data of the effective measurement point set, and the compensated propagation time parameter; Step 1042a: Based on the acoustic propagation characteristics and coordinate transformation relationship of the USBL system, determine the nonlinear observation equation of the measurement vector and the state vector.

[0057] In practical implementation, for special scenarios (slant range constraint failure scenarios), the measurement vector is directly constructed using multi-source raw data from the effective measurement point set. Specifically, the measurement vector includes the azimuth angle measured by the USBL acoustic array and the compensated propagation time parameters. Based on the coordinate transformation relationship and acoustic propagation characteristics, a nonlinear observation equation between the measurement vector and the state vector is established.

[0058] For example, the expression for the observation equation is: ;in, For measurement vectors, It is a state vector (including installation angle error, lever error, and the product of sound speed and signal transmission time). For nonlinear observation functions, To measure noise, Let k be the time of the signal reception. For signal transmission period, For the speed of sound, ( Let be the azimuth angle measured by the USBL acoustic array for the kth time. Furthermore, the state vector... Including installation angle error Lever arm error and speed of sound With signal transmission time The product of , is expressed as: .

[0059] It should be noted that, for the slant range constraint failure scenario, this step directly constructs the measurement vector and observation equation using an effective set of measurement points, avoiding the failure problem of the traditional slant range difference model in such scenarios, expanding the trajectory applicability of SINS / USBL navigation system calibration, and ensuring the validity of calibration data and the stability of subsequent filtering estimation in special scenarios.

[0060] In one possible implementation, for a normal scenario (non-failure scenario), the step in S104 of calculating the attention weights of each measurement dimension using the attention weight function and constructing the observation equation by weighting them includes: Step 1041b: Construct an attention weight function based on the credibility of the measurement data of the effective measurement point set, calculate the deviation between the measurement value of the current effective measurement point and the measurement value of the historical effective measurement point under each measurement dimension, and calculate the attention weight of each measurement dimension by combining the noise standard deviation estimate corresponding to each measurement dimension.

[0061] Here, measurement dimensions refer to the multiple core constituent dimensions of the measurement vector in a typical scenario, including but not limited to the azimuth x-dimensional, azimuth y-dimensional, and compensated propagation time dimension measured by the USBL acoustic array. The degree of deviation characterizes the reliability (the smaller the deviation, the higher the reliability, and the closer the weight is to 1; the larger the deviation, the lower the reliability, and the closer the weight is to 0). The noise standard deviation estimate is obtained based on previous measurement data and is used to quantify the noise level of each measurement dimension.

[0062] For example, by substituting the degree of bias and the estimated standard deviation of noise into the weighting function, we obtain the attention weights for each dimension, expressed as: ) / ;in, Let j be the attention weight in dimension j, and i correspond to one of the following dimensions: azimuth x-dimensional, azimuth y-dimensional, or the compensated propagation time t. The degree of bias in dimension j. As the noise standard deviation estimate, the attention weights for each measurement dimension are calculated as follows: x , y , t .

[0063] Step 1042b: Incorporate attention weights into the measurement update process of filtering, obtain the predicted measurement values ​​of each measurement dimension through a nonlinear observation function, and calculate the weighted corrected measurement residuals by combining the attention weights.

[0064] For example, weighted correction measurement residuals , To predict the measured value, For the measurement vector, diag() means to construct a diagonal matrix.

[0065] Step 1043b: Based on the weighted correction of the measurement residuals, the original measurement noise covariance matrix is ​​weighted and corrected by the attention weights of each measurement dimension. Combined with the acoustic propagation characteristics and coordinate transformation relationship of the USBL system, the observation equation is constructed in a weighted manner.

[0066] Among them, the corrected measurement noise covariance matrix , The original noise covariance matrix is ​​given. A weighted observation equation is constructed based on the corrected measurement residuals and the noise covariance matrix. The noise covariance of dimensions with high confidence is "compressed," indicating that the data in that dimension has less noise and is more reliable; the noise covariance of dimensions with low confidence is "amplified," and filtering automatically reduces the confidence level in the data of that dimension.

[0067] It should be noted that, in conventional scenarios, this step quantifies the credibility of each measurement dimension through an attention weight function and constructs an observation equation with weights. This can adaptively strengthen high-credibility data, weaken low-credibility data and noise interference, adapt to time-varying noise environments, improve the accuracy and stability of filtered state estimation, and ensure the calibration accuracy of the SINS / USBL navigation system.

[0068] In one possible implementation, the initial value obtained from the K-means clustering initialization algorithm optimized by deep reinforcement learning in step S105, and the step of running filtering iterations to update the state, includes: Step 1051: Define the state space as the feature vector of the initial effective measurement point set, the action space as the selection range of cluster centers, and the reward function as determined based on the contour coefficient after clustering, the average distance between each cluster center and the global state mean, and the penalty coefficient.

[0069] The feature vector includes features such as azimuth, slant range, and propagation time. The distance, for example, is Euclidean distance.

[0070] Step 1052: Train the agent through a deep learning network to learn the optimal initial cluster center selection strategy to maximize the reward function value.

[0071] Here, maximizing the reward function value is equivalent to achieving the optimal clustering effect, with the silhouette coefficient close to 1 and the cluster centers far from outlier samples.

[0072] Step 1053: Using the trained clustering algorithm, cluster the preliminary estimation results corresponding to the effective measurement data of the initial preset group, select the cluster center with the best silhouette coefficient as the initial state estimate of the filter, and adaptively set the initial covariance matrix according to the degree of dispersion of the clustering results.

[0073] Step 1054: Based on variational Bayesian theory, assuming that the measurement noise covariance matrix follows an inverse Wissaud distribution, and combining the power relationship between ranging noise and propagation distance, a distance correction factor is introduced to optimize the prior update of the measurement noise covariance matrix.

[0074] The distance correction factor is a correction term that integrates the forgetting factor, propagation loss factor, and the current measured propagation distance, used to optimize the estimation of time-varying noise covariance. For example, the distance correction factor for the k-th measurement can be determined based on the forgetting factor and the propagation distance of the k-th measurement, as the propagation loss varies at different propagation distances.

[0075] Step 1055: Generate the sampling points required for unscented Kalman filtering. After state prediction and measurement prediction, calculate the weighted Kalman gain by combining the weighted corrected measurement residual and the weighted corrected measurement noise covariance matrix. Use the weighted Kalman gain to update the posterior state estimate and posterior covariance matrix to complete one filtering iteration.

[0076] By quantifying the reliability of each measurement dimension through an attention weighting function, weighting and correcting the measurement residual and noise covariance matrix, and constructing an observation equation by combining acoustic propagation characteristics and coordinate transformation relationships, high-reliability data enhancement and low-reliability data and noise reduction are achieved, thereby improving the accuracy and robustness of filtering estimation in SINS / USBL navigation systems under time-varying noise environments.

[0077] Reference Figure 3 The diagram shown is a functional module schematic of the navigation system calibration device 300 of this application. Figure 3 As shown, the navigation system calibration device 300 includes functions such as... Figure 1 or Figure 2 The navigation system calibration method described in this application comprises modules. The navigation system calibration device 300 is installed in an electronic device. Depending on its function, the navigation system calibration device 300 includes a data acquisition module 310, a filtering module 320, a judgment module 330, a construction module 340, an operation module 350, and a calibration module 360. These modules can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, and are stored in the memory of the electronic device.

[0078] In this embodiment, the functions of each module / unit are as follows: The acquisition module 310 is used to start the multi-source navigation and positioning device and acquire multi-source raw data; The filtering module 320 is used to perform time synchronization and interpolation completion on the multi-source raw data, and to filter out the effective set of measurement points that meet the slope distance difference constraint. The determination module 330 is used to determine whether the current trajectory in the set of valid measurement points is in a slant range constraint failure scenario; The construction module 340 is used to directly construct the measurement vector and observation equation using the effective measurement point set if the condition is met; otherwise, it calculates the attention weights of each measurement dimension through the attention weight function and constructs the observation equation by weighting them. The running module 350 is used to run filtering iterations to update the state, starting from the initial value obtained by the K-means clustering initialization algorithm optimized based on deep reinforcement learning. The calibration module 360 ​​determines whether the status has converged. If it has converged, it outputs the installation error parameters and compensates for the system installation error. If it has not converged, it returns to the time synchronization and interpolation completion of the multi-source raw data to continue the iteration until the calibration process is completed.

[0079] The specific implementation of the navigation system calibration device in this application is largely the same as the specific implementation of the navigation system calibration method described above, and will not be repeated here.

[0080] Reference Figure 4 The diagram shown is a schematic representation of a preferred embodiment of the electronic device of this application.

[0081] The electronic device includes a processor 411, a communication interface 412, a memory 413, and a communication bus 414, wherein the processor 411, the communication interface 412, and the memory 413 communicate with each other through the communication bus 414. Memory 413 is used to store computer programs; In some embodiments, the processor 411 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 411 is typically used to control the overall operation of the electronic device, such as performing data interaction or communication-related control and processing. In this embodiment, the processor 411 is used to run program code stored in the memory 413 or process data.

[0082] The communication interface 412 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The communication interface 412 may also be used to establish a communication connection between the electronic device and other electronic devices.

[0083] The memory 413 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 413 may be an internal storage unit of the electronic device, such as the hard disk or memory of the electronic device. In other embodiments, the memory 413 may also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. of the electronic device. Of course, the memory 413 may include both internal storage units and external storage devices of the electronic device. In this embodiment, the memory 413 is typically used to store the operating system and various computer programs installed on the electronic device, such as the program code of a navigation system calibration program. In addition, the memory 413 may also be used to temporarily store various types of data that have been output or will be output.

[0084] Figure 4 Only an electronic device with a processor 411, a communication interface 412, a memory 413, and a communication bus 414 is shown. However, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0085] In one embodiment of this application, when the processor 411 executes the program stored in the memory 413, it implements the navigation system calibration method provided in any of the foregoing method embodiments, including: Start the multi-source navigation and positioning equipment to collect raw data from multiple sources; The multi-source raw data is time-synchronized and interpolated to complete the data, and the effective set of measurement points that meet the slope distance difference constraint is selected. Determine whether the current trajectory in the set of valid measurement points is in a slant range constraint failure scenario; If so, the effective set of measurement points is used directly to construct the measurement vector and observation equation; otherwise, the attention weights of each measurement dimension are calculated using the attention weight function and the observation equation is constructed by weighting them. The initial value obtained by the K-means clustering initialization algorithm optimized by deep reinforcement learning is used as the starting point, and the state is updated by running filtering iterations. Determine if the state has converged. If it has converged, output the installation error parameters and compensate for the system installation error. If it has not converged, return to the time synchronization and interpolation completion stage of the multi-source raw data to continue iterating until the calibration process is completed.

[0086] For a detailed explanation of the above steps, please refer to the above. Figure 1 Description of the flowchart of an embodiment of the navigation system calibration method.

[0087] Furthermore, embodiments of this application also propose a computer-readable storage medium that is both non-volatile and volatile. This computer-readable storage medium is any one or any combination of several of the following: hard disk, multimedia card, SD card, flash memory card, SMC, read-only memory (ROM), erasable programmable read-only memory (EPROM), portable compact disc read-only memory (CD-ROM), USB memory, etc. The computer-readable storage medium includes a data storage area and a program storage area, the program storage area storing a computer program.

[0088] The specific implementation of the computer-readable storage medium in this application is largely the same as the specific implementation of the navigation system calibration method described above, and will not be repeated here.

[0089] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0090] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware simulation platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0091] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A navigation system calibration method, characterized in that, The method includes: Start the multi-source navigation and positioning equipment to collect raw data from multiple sources; The multi-source raw data is time-synchronized and interpolated to complete the data, and the effective set of measurement points that meet the slope distance difference constraint is selected. Determine whether the current trajectory in the set of valid measurement points is in a slant range constraint failure scenario; If so, the effective set of measurement points is used directly to construct the measurement vector and observation equation; otherwise, the attention weights of each measurement dimension are calculated using the attention weight function and the observation equation is constructed by weighting them. The initial value obtained by the K-means clustering initialization algorithm optimized by deep reinforcement learning is used as the starting point, and the state is updated by running filtering iterations. Determine if the state has converged. If it has converged, output the installation error parameters and compensate for the system installation error. If it has not converged, return to the time synchronization and interpolation completion stage of the multi-source raw data to continue iterating until the calibration process is completed.

2. The method as described in claim 1, characterized in that, The steps for selecting valid measurement points that satisfy the slope difference constraint include: Based on the trajectory design parameters of the current navigation task, determine the failure threshold of the slant range difference model, and calculate the target spatial distance between each current sample measurement point and the trajectory center; When the target spatial distance is greater than the failure area determination radius, the sample measurement point is used as a candidate measurement point; Calculate the slope distance difference between the candidate measurement point and its adjacent candidate measurement points. If the slope distance difference is greater than or equal to the determination threshold, then the candidate measurement point is determined to be a valid measurement point that satisfies the slope distance difference constraint condition.

3. The method as described in claim 1, characterized in that, The step of determining whether the current trajectory in the set of valid measurement points is in a slant range constraint failure scenario includes: Extract the slope distance data of all valid measurement points from the set of valid measurement points, and calculate the unbiased variance based on the slope distance data of all valid measurement points; If the unbiased variance is less than the scene determination threshold, then the current trajectory is determined to be in a slant range constraint failure scenario.

4. The method as described in claim 1, characterized in that, The steps for constructing the measurement vector and observation equation using the effective set of measurement points include: Based on the multi-source raw data in the effective measurement point set, a measurement vector is determined; the measurement vector includes the azimuth angle measured by the USBL acoustic array, the slant range data of the effective measurement point set, and the compensated propagation time parameter. Based on the acoustic propagation characteristics and coordinate transformation relationship of the USBL system, the nonlinear observation equations of the measurement vector and the state vector are determined.

5. The method as described in claim 1, characterized in that, The steps of calculating the attention weights for each measurement dimension using the attention weight function and constructing the observation equation by weighting them include: An attention weight function is constructed based on the credibility of the measurement data of the effective measurement point set. The deviation between the measurement value of the current effective measurement point and the measurement value of the historical effective measurement point is calculated under each measurement dimension. Combined with the noise standard deviation estimate corresponding to each measurement dimension, the attention weight of each measurement dimension is calculated. Attention weights are incorporated into the measurement update process of filtering, and the predicted measurement values ​​of each measurement dimension are obtained through a nonlinear observation function. The weighted correction measurement residuals are then calculated by combining the attention weights. Based on the weighted correction of the measurement residuals, the original measurement noise covariance matrix is ​​also weighted and corrected by the attention weights of each measurement dimension. Combined with the acoustic propagation characteristics and coordinate transformation relationship of the USBL system, the observation equation is constructed in a weighted manner.

6. The method as described in claim 1, characterized in that, The initial value obtained from the K-means clustering initialization algorithm optimized by deep reinforcement learning is used as the starting point. The steps of running filtering iterations to update the state include: The state space is defined as the feature vector of the initial effective measurement point set, the action space is the selection range of cluster centers, and the reward function is determined based on the contour coefficient after clustering, the average distance between each cluster center and the global state mean, and the penalty coefficient. The agent is trained by deep learning network to learn the optimal initial cluster center selection strategy to maximize the reward function value; The trained clustering algorithm is used to cluster the preliminary estimation results corresponding to the effective measurement data of the initial preset group. The cluster center with the best profile coefficient is selected as the initial state estimate of the filter. The initial covariance matrix is ​​adaptively set according to the degree of dispersion of the clustering results. Based on variational Bayesian theory, assuming that the measurement noise covariance matrix follows an inverse Wissaud distribution, and combining the power relationship between ranging noise and propagation distance, a distance correction factor is introduced to optimize the prior update of the measurement noise covariance matrix. The sampling points required for unscented Kalman filtering are generated. After state prediction and measurement prediction, the weighted Kalman gain is calculated by combining the weighted corrected measurement residual and the weighted corrected measurement noise covariance matrix. The posterior state estimate and posterior covariance matrix are updated using the weighted Kalman gain, thus completing one filtering iteration.

7. The method as described in claim 6, characterized in that, The distance correction factor is a correction term that integrates the forgetting factor, the propagation loss factor, and the currently measured propagation distance.

8. A navigation system calibration device, characterized in that, The device includes: The apparatus includes a module that performs the navigation system calibration method as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the navigation system calibration method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the navigation system calibration method as described in any one of claims 1 to 7.

Images