A ship indoor UWB multi-mode ad hoc network trusted positioning system

Abstract

The application relates to the technical field of ship indoor positioning, and particularly relates to a ship indoor UWB multi-mode ad hoc network trusted positioning system, which comprises the following modules: a trustworthiness evaluation module for evaluating the trustworthiness of deep feature representation of UWB observation; a hierarchical shunting module for shunting and disposing the observation based on the trustworthiness score; an error compensation module for compensating the medium-trustworthy observation based on the deep representation feature and the trustworthiness score of the medium-trustworthy observation; a candidate positioning solution construction module for constructing a candidate positioning solution set according to the high-trustworthy observation and the compensated medium-trustworthy observation, calculating the positioning trustworthiness according to the trustworthiness score of the observation in the candidate positioning solution set, and selecting the candidate positioning solution with the highest positioning trustworthiness as the optimal positioning solution; and a positioning solution module for constructing a weight according to the trustworthiness score of the observation corresponding to the optimal positioning solution and solving the UWB positioning coordinates. The application improves the precision, stability and robustness of UWB positioning.

Description

Technical Field

[0001] This invention relates to the field of ship indoor positioning technology, specifically to a ship indoor UWB multi-mode self-organizing network trusted positioning system. Background Technology

[0002] With the development of demands for intelligent ships, ship operation safety supervision, cabin personnel location monitoring, and emergency rescue, high-precision indoor positioning technology for ships has received increasing attention. The interior space of ships is typically characterized by narrow, enclosed compartments, dense metal structures, significant obstruction and multipath effects, limited base station deployment, and a complex and variable propagation environment, making indoor positioning for ships face complex application conditions. Existing indoor positioning technologies for ships mainly include Wi-Fi, Bluetooth, RFID, visual positioning, and Ultra Wideband (UWB) positioning. While Wi-Fi, Bluetooth, and RFID technologies are relatively easy to deploy, their positioning accuracy and stability are easily affected by signal fluctuations, environmental interference, and coverage conditions; visual positioning is easily affected by changes in lighting, obstruction, and limited field of view. In contrast, UWB has advantages such as high temporal resolution, high measurement accuracy, and strong resistance to multipath effects, thus showing promising application prospects in high-precision indoor positioning for ships.

[0003] However, existing UWB positioning methods still have shortcomings in the complex propagation environment of ship interiors. The quality of distance, horizontal angle, and pitch angle observations corresponding to different base stations varies significantly under conditions of non-line-of-sight propagation, enhanced local reflections, and dynamic changes in visibility. Existing technologies typically lack a unified quantitative characterization of the reliability of multiple types of observations, making it difficult to perform refined processing for different observations. Simultaneously, the observation quality assessment, anomaly screening, error compensation, and positioning calculation processes in existing methods are often separated. This makes it difficult to dynamically construct an adapted positioning calculation mode based on the number of available base stations, base station combinations, observation types, and their quality states. It also makes it difficult to select the most reliable positioning result from multiple feasible solutions, thus affecting the accuracy, stability, and reliability of the final positioning output. Therefore, it is necessary to propose a reliable UWB positioning system for the complex indoor environment of ships to address these problems. Summary of the Invention

[0004] To address the technical problems of unifying the characterization of various observation types, disconnect between observation processing and positioning calculation, insufficient adaptability of positioning modes, and low reliability of positioning results in complex ship indoor UWB positioning methods under complex propagation environments, this invention provides a ship indoor UWB multi-mode self-organizing network reliable positioning system. This invention primarily achieves reliability assessment of distance, horizontal angle, and pitch angle observations by deeply characterizing the channel impulse response (CIR) sequence. Based on the reliability, it performs hierarchical distribution and differentiated error compensation for various observations. Furthermore, it constructs a candidate positioning solution set by combining the number of available base stations, base station combinations, and observation types at the current time. The optimal candidate positioning solution is selected based on the positioning reliability of each candidate solution. Weights are constructed based on the reliability scores of the observations actually participating in the positioning calculation in the selected optimal candidate solution, and the positioning solution is then obtained. This improves the accuracy, stability, and reliability of UWB positioning in complex ship indoor environments.

[0005] The technical means employed in this invention are as follows:

[0006] A shipboard indoor UWB multi-mode self-organizing network trusted positioning system includes: The raw data acquisition module is used to acquire observations and corresponding CIR sequences, wherein the observations include distance, horizontal angle, and pitch angle; A deep characterization extraction module is used to extract the deep characterization features of the CIR sequence; The credibility assessment module is used to assess the credibility of the deep representation features and obtain a credibility score. The hierarchical distribution module is used to perform hierarchical distribution processing on the observations based on the confidence score, so as to obtain high confidence observations, medium confidence observations and low confidence observations; An error compensation module is used to compensate for medium-confidence observations based on the depth characterization features and the confidence score of the medium-confidence observations. The candidate location solution construction module is used to construct a set of candidate location solutions based on the number of base stations, the combination of base stations, high-confidence observations and compensated medium-confidence observations. On the other hand, it is used to calculate the location confidence based on the confidence scores corresponding to the observations in the candidate location solution set, and select the candidate location solution with the highest location confidence as the optimal location solution. The localization solution module is used to construct weights based on the confidence scores corresponding to the observations of the optimal localization solution, and to obtain UWB localization coordinates based on the weights and the least squares localization solution.

[0007] Furthermore, the raw data acquisition module includes UWB base stations that measure UWB tags, forming a UWB base station list. The UWB base station list includes the observations of each UWB base station and the corresponding CIR sequences.

[0008] Furthermore, the deep representation extraction module is a Transformer encoder, which is obtained through self-supervised pre-training based on unlabeled CIR sequences. The pre-trained Transformer encoder encodes the CIR sequences to obtain deep representations. The self-supervised pre-training employs several pre-training tasks to learn the representations of unlabeled CIR sequences, including CIR sequence mask reconstruction tasks and contrastive learning tasks.

[0009] Furthermore, the credibility assessment module inputs the deep representation features into the credibility assessment network that has been fine-tuned and trained, and outputs a credibility score for the observations.

[0010] Furthermore, the hierarchical distribution module classifies observations with a confidence level below a first threshold as low-confidence observations, observations with a confidence level greater than or equal to a second threshold as high-confidence observations, and observations with a confidence level greater than or equal to the first threshold and lower than the second threshold as medium-confidence observations, wherein the first threshold is less than the second threshold.

[0011] Furthermore, the error compensation module inputs the depth characterization features, the confidence score of the medium-confidence observations, and the medium-confidence observations into the error compensation network, outputs the observation error estimate, and compensates the medium-confidence observations based on the observation error estimate.

[0012] Furthermore, the candidate positioning solution construction module constructs candidate positioning solution patterns based on the number of base stations, the combination of base stations, and the type of observation. It then determines whether the candidate positioning solution pattern meets the requirements of three-dimensional positioning solution based on the rank of the Jacobian matrix of the observation equation corresponding to the candidate positioning solution pattern. If the requirements of three-dimensional positioning solution are met, the positioning solution pattern is included in the candidate positioning solution set; otherwise, it is discarded.

[0013] Furthermore, the candidate localization solution construction module constructs observation equations based on the observations in the candidate localization solution mode, and constructs observation functions based on all observation equations; it constructs a Jacobian matrix based on the observation functions and the three-dimensional coordinates of the target to be located; it performs singular value decomposition on the Jacobian matrix to obtain singular values, and counts the number of singular values ​​greater than a third threshold. When the number is greater than or equal to 3, the current candidate localization solution mode meets the three-dimensional localization solution requirements; when the number is less than 3, it is discarded.

[0014] Furthermore, the formula for calculating the UWB positioning coordinates is as follows:

[0015] in, Locate coordinates for UWB. Let the label coordinates be the variables to be estimated. For the residual vector, The weight is denoted as .

[0016] Compared with the prior art, the present invention has the following advantages: This invention assesses the reliability of range, horizontal angle, and pitch angle observations using depth characterization information based on channel impulse response (CIR) sequences, and uniformly represents them as reliability scores. This provides a consistent basis for the unified processing and positioning calculation of multiple types of observations. Based on the reliability scores, various observations are categorized and distributed, and error estimation and compensation are implemented using depth characterization information. This enables differentiated processing of observations with different quality levels, improving observation utilization efficiency and error compensation effectiveness. Furthermore, a candidate positioning solution set is dynamically constructed based on the number of available base stations, base station combinations, and observation types. This allows the positioning calculation mode to adaptively adjust to changes in base station visibility, observation integrity, and propagation conditions, thereby improving the method's handling of complex shipboard indoor obstructions, multipath propagation, and partial obstructions. The invention enhances the adaptability to constrained conditions. It calculates the positioning reliability based on the reliability scores of each observation in the candidate positioning solution and selects the candidate positioning solution with the highest reliability as the optimal candidate positioning solution, which improves the reliability of the positioning results. A weight matrix is ​​constructed based on the reliability scores of the observations actually involved in the positioning calculation in the optimal candidate positioning solution, and the reliability-weighted least squares method is used for positioning. This effectively suppresses the adverse effects of low-reliability observations and improves the accuracy, stability, and robustness of UWB positioning in complex shipboard indoor environments. Furthermore, the candidate positioning solution constructed in this invention can adapt to different base station combinations and different observation combinations, thus being compatible with scenarios such as single observation type availability and multi-type observation joint calculation, exhibiting good flexibility and engineering applicability.

[0017] Based on the above reasons, this invention can be widely applied in fields such as ship indoor positioning. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the structure of a shipboard indoor UWB multi-mode self-organizing network trusted positioning system according to the present invention.

[0020] Figure 2 This is a schematic diagram of the workflow of the present invention.

[0021] Figure 3This is a schematic diagram illustrating the principle of the present invention.

[0022] Figure 4 This is a schematic diagram of an indoor UWB positioning application scenario for ships provided by the present invention.

[0023] Figure 5 A comparison chart of positioning error CDF curves provided by the present invention.

[0024] In the diagram: 1. Label; 2. Electrical control board; 3. Equipment frame. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] This invention provides a reliable UWB multi-mode self-organizing network positioning system for ship indoor positioning systems. The positioning system includes UWB tags, a positioning processing unit, and N UWB base stations located within the ship's indoor positioning area, where N is a positive integer. The UWB tags are installed on the target to be positioned and are used for wireless communication with each UWB base station to generate raw measurement data related to positioning. The N UWB base stations are distributed within the ship's indoor positioning area to acquire distance observations, horizontal angle observations, pitch angle observations, and corresponding channel impulse response (CIR) sequences between the tags and each base station. The positioning processing unit is communicatively connected to the UWB tags and receives the raw measurement data. It then performs depth characterization extraction, observation reliability assessment, hierarchical processing, error compensation, candidate positioning solution construction, positioning reliability calculation, optimal candidate positioning solution selection, and final positioning solution calculation, thereby outputting the final UWB positioning coordinates. The ship's indoor positioning area can be a ship's cabin, corridor, engine room, cargo hold, control area, or other enclosed or semi-enclosed internal space. Due to the presence of bulkhead obstructions, metal reflections, multipath propagation, spatial constraints, and varying base station visibility in such environments, the type, quantity, and quality of observables acquired at different positioning times often differ. To address these characteristics, this embodiment introduces an observable reliability assessment mechanism based on CIR depth characterization, combined with a dynamic construction and optimization strategy for candidate positioning solutions, to achieve reliable UWB multi-mode ad hoc network positioning in complex shipboard indoor environments. Figure 1 As shown, the present invention includes: The raw data acquisition module is used to acquire the observations and their corresponding CIR sequences. The observations include distance, horizontal angle, and pitch angle.

[0028] The deep characterization extraction module is used to extract deep characterization features from CIR sequences.

[0029] The credibility assessment module is used to assess the credibility of deep representation features and obtain a credibility score.

[0030] The hierarchical distribution module is used to classify and distribute observations based on their confidence scores, resulting in high-confidence, medium-confidence, and low-confidence observations.

[0031] The error compensation module is used to compensate for medium-confidence observations based on the confidence scores of deep characterization features and medium-confidence observations.

[0032] The candidate location solution construction module is used to construct a set of candidate location solutions based on the number of base stations, the combination of base stations, high-confidence observations, and compensated medium-confidence observations. It is also used to calculate the location confidence based on the confidence scores corresponding to the observations in the candidate location solution set, and select the candidate location solution with the highest location confidence as the optimal location solution.

[0033] The localization solution module is used to construct weights based on the confidence scores corresponding to the observations of the optimal localization solution, and obtain the UWB localization coordinates based on the weights and the least squares localization solution.

[0034] In the raw data acquisition module, the positioning area is pre-set The first UWB base station, the... The spatial coordinates of the UWB base stations are ( The process involves measuring the UWB tags of UWB base stations to create a UWB base station list. This list includes observations for each UWB base station and their corresponding CIR sequences. The observation data for each base station in the UWB base station list includes distance observations. Horizontal angle observation Pitch angle observation and the corresponding CIR sequence .

[0035] Acquire raw UWB measurement data. At the current positioning time. To obtain raw UWB measurement data. For the first... The measurement quantity corresponding to each available UWB base station can be expressed as:

[0036] in, For distance observation, For horizontal angle observation, For pitch angle observation, This is the corresponding channel impulse response (CIR) sequence. Further, the CIR sequence can be expressed as:

[0037] in, The length of the CIR sequence is given. Due to obstruction, multipath propagation, reflection, and local visibility variations in the ship's indoor environment, the type, quantity, and quality of observations output by different base stations at different positioning times may differ, thus providing a basis for subsequent observation reliability assessment, hierarchical triage, and candidate positioning solution construction.

[0038] The deep representation extraction module is a Transformer encoder, which is obtained through self-supervised pre-training based on unlabeled CIR sequences. The pre-trained Transformer encoder processes CIR sequences... Encode to obtain a deep representation To characterize the channel propagation state and observation error characteristics. Self-supervised training employs several pre-training tasks to learn representations of unlabeled CIR sequences, including CIR sequence mask reconstruction and contrastive learning tasks.

[0039] The CIR sequence is input into a pre-built Transformer encoder to extract deep representation features:

[0040] in, This represents a Transformer-based deep representation extraction function. Indicates the first Each base station at time The corresponding deep representation features. Deep representations are used to characterize information related to propagation path states, first-path features, multipath distribution, and energy decay in the CIR sequence. In one embodiment, the Transformer encoder is obtained through self-supervised pre-training based on unlabeled CIR sequences. Self-supervised pre-training employs one or more pre-training tasks to learn representations of the unlabeled CIR sequences, including one or more of CIR sequence mask reconstruction tasks and contrastive learning tasks.

[0041] The credibility assessment module will deeply characterize features The data is fed into a finely tuned and trained reliability evaluation network, which outputs reliability scores for the observations. When the observations include distance, horizontal angle, and pitch angle observations, reliability scores are obtained to characterize the reliability of the distance, horizontal angle, and pitch angle observations, respectively. , and ,in: , , ,and , and All belong to the interval A higher score indicates greater reliability of the corresponding observation. Using this method, the quality status of different types of observations can be uniformly represented by a reliability score, providing a basis for subsequent tiered and differentiated handling, error compensation, and location calculation. Score the reliability of distance observations. Score the reliability of the horizontal angle observations. To score the reliability of the pitch angle observations, Let be the confidence mapping function for distance observations. Let be the confidence mapping function for the horizontal angle observation. This is the confidence mapping function for pitch angle observations.

[0042] The hierarchical distribution module classifies observations with a confidence level below a first threshold as low-confidence observations, observations with a confidence level greater than or equal to a second threshold as high-confidence observations, and observations with a confidence level greater than or equal to the first threshold but lower than the second threshold as medium-confidence observations. Less than the second threshold .

[0043] Specifically, observations are categorized and processed according to their credibility scores; for any given observation, a credibility score is assigned... ,in Indicates the observation type, using a preset threshold. and Perform credibility grading and meet the following requirements. Its hierarchical rule is expressed as: when When, it is determined to be a high-confidence observation; when When, it is determined to be a moderately reliable observation; when When the observation is determined to be low confidence, high confidence observations are directly used in the localization solution, medium confidence observations are used in the localization solution after weight reduction or error compensation, and low confidence observations are eliminated or used only for error compensation reference.

[0044] The error compensation module inputs the depth characterization features, the confidence score of the medium-confidence observations, and the medium-confidence observations into the error compensation network, outputs the observation error estimate, and compensates the medium-confidence observations based on the observation error estimate.

[0045] Specifically, the first Depth representation of each base station The observed measurements after hierarchical and triaged processing, along with their corresponding confidence scores, are input into the error compensation network. It outputs the corresponding observation error estimate and obtains the compensated observations based on the observation error estimate; when the observations include distance observations... Horizontal angle observation and pitch angle observation At that time, error compensation network Output distance error estimate Horizontal angle error estimate and pitch angle error estimate ,in: Based on the error estimate, the compensated distance observation is obtained. Horizontal angle observation and pitch angle observation ,in: , , .

[0046] The error compensation network is a multilayer perceptron network. The deep representation features, the observations after hierarchical and diversion processing, and the confidence scores corresponding to the observations are input into the multilayer perceptron network, which outputs the corresponding observation error estimate and compensates the observations according to the observation error estimate to obtain the compensated observations.

[0047] The above methods can be used to provide differentiated compensation for observations with different quality levels and error states, thereby improving the availability and reliability of observations in subsequent positioning calculations.

[0048] The candidate localization solution construction module determines the candidate localization solution based on the number of base stations, the combination of base stations, and the type of observations, according to the rank of the Jacobian matrix of the corresponding observation equation. When the requirements for three-dimensional localization solution are met, the corresponding localization solution mode is included in the candidate localization solution set; otherwise, it is discarded. Each candidate localization solution in the candidate localization solution set corresponds to a localization solution mode, which is determined by the combination of base stations participating in the current localization solution and the corresponding combination of observations.

[0049] Based on the number of available base stations, the combination of base stations, and the type of observations at the current moment, a set of candidate positioning solutions that satisfy the conditions for 3D positioning calculation is constructed. ; where each candidate localization solution A corresponding positioning solution mode is defined by the combination of base stations participating in the current positioning solution. and its corresponding combination of observations Determine the combination of observations. The set of observations participating in the current positioning solution includes observations directly retained after hierarchical distribution and / or observations after error compensation; base station combinations and observation combinations that do not meet the conditions for 3D positioning solution are removed to obtain the candidate positioning solution set. .

[0050] Specifically, based on the number of available base stations, the combination of base stations, and the type of observations at the current time, a set of candidate positioning solutions that satisfy the conditions for 3D positioning calculation is constructed. Let the tag to be positioned be at time... The location is:

[0051] in, Let k be the position of the tag to be located at time k.

[0052] No. The coordinates of the base stations are:

[0053] in, Let be the coordinates of the i-th base station.

[0054] In one specific embodiment, a set of candidate localization solutions is constructed:

[0055] in, For the set of candidate localization solutions, each candidate localization solution A corresponding positioning solution mode is defined by the combination of base stations participating in the current positioning solution. and its corresponding combination of observations Determine the combination of observations. Including distance observations ( , ), Horizontal angle observation ( , ) and pitch angle observation ( , One or more of the following. For each combination of base stations and observations, determine whether it meets the conditions for solving the three-dimensional positioning problem.

[0056] Preferably, the determination can be made based on the rank of the Jacobian matrix of the corresponding observation equation; when the three-dimensional positioning solution requirements are met, the corresponding positioning solution mode is included in the candidate positioning solution set. If the solution is selected, it is discarded. Through this method, a set of candidate positioning solutions can be dynamically generated based on changes in base station visibility, observation integrity, and propagation conditions at the current moment. These solutions can be used for subsequent positioning reliability calculations, selection of optimal candidate positioning solutions, and final positioning solutions.

[0057] The candidate positioning solution construction module constructs candidate positioning solution patterns based on the number of base stations, base station combinations, and observation types. It then determines whether the pattern meets the 3D positioning solution requirements based on the rank of the Jacobian matrix corresponding to the observation equation. Specifically, for any candidate base station combination, the candidate positioning solution construction module selects the corresponding distance observation equation, horizontal angle observation equation, and / or pitch angle observation equation based on the available observation types in that combination, thus constructing an observation function.

[0058] Where S represents the current candidate base station combination and its corresponding observation type, p represents the three-dimensional coordinates of the target to be located, and m represents the number of available observation equations under the current candidate localization solution mode. The observation function is defined as follows: the distance observation equation is used to characterize the spatial distance relationship between the base station and the target to be located; the horizontal angle observation equation is used to characterize the horizontal orientation relationship between the base station and the target to be located; and the pitch angle observation equation is used to characterize the vertical angle relationship between the base station and the target to be located.

[0059] The candidate localization solution construction module performs observation functions. Taking the partial derivative with respect to the three-dimensional coordinates of the target to be located, we obtain the corresponding observation equation Jacobian matrix:

[0060] in, Let be the Jacobian matrix. When the Jacobian matrix of the observation equation satisfies: When determining that the current candidate base station combination and its corresponding observation type can meet the requirements of three-dimensional positioning calculation, its corresponding positioning calculation mode is included in the candidate positioning solution set; when the observation equation Jacobian matrix satisfies If the current candidate base station combination and its corresponding observation type do not meet the requirements of three-dimensional positioning solution, they are removed from the candidate positioning solution set.

[0061] The candidate localization solution construction module is specifically used for: obtaining the set of base stations available at the current time and the available observation types corresponding to each base station; generating candidate base station combinations based on the set of available base stations, and generating observation combinations based on the available observation types corresponding to each base station in the candidate base station combinations; determining each candidate base station combination and its corresponding observation combination as a localization solution mode, and setting a mode identifier for the localization solution mode; performing localization solutions on the localization solution modes that meet the requirements of three-dimensional localization solution, and obtaining the corresponding candidate localization solutions; and associating and storing the mode identifier, base station combination, observation combination, confidence score, and candidate localization solution so that the candidate localization solution and the localization solution mode form a one-to-one correspondence.

[0062] Calculate the location confidence of candidate localization solutions and select the optimal candidate localization solution. Based on the confidence scores of the observations included in each candidate localization solution in the candidate localization solution set, calculate the location confidence of each candidate localization solution, and select the candidate localization solution with the highest location confidence as the optimal candidate localization solution. Let the th... The candidate localization solutions are at time... The location credibility is , can be represented as:

[0063] in, , and These are the weighting coefficients for distance observations, horizontal angle observations, and pitch angle observations, respectively. , and They represent the first The number of distance, horizontal angle, and pitch angle observations in each candidate positioning solution. Positioning reliability. Used for comprehensive characterization of the first The overall reliability of various observation information in each candidate localization solution is calculated. The numerator is used to weight and summarize the reliability scores of various observations, while the denominator is used to normalize the differences in the number of observations among different candidate localization solutions, to avoid inflated localization reliability solely due to a large number of observations. Based on the localization reliability of each candidate localization solution, the candidate localization solution with the highest localization reliability is selected as the optimal candidate localization solution, i.e.:

[0064] in, This indicates the number corresponding to the optimal candidate localization solution.

[0065] In the localization solution module, a weight matrix is ​​constructed based on the confidence scores corresponding to the observations actually involved in the localization solution in the optimal candidate localization solution, and the localization solution is obtained using the confidence-weighted least squares method. Let the optimal candidate localization solution be at time... The corresponding observation vector is The theoretical observation function is ,in To estimate the label location, the final location estimation result can be obtained through the following objective function:

[0066] in, This is a weight matrix constructed based on credibility scores. This is the location estimation result. Preferably, It can be a diagonal matrix, whose diagonal elements can be obtained by mapping the confidence scores of the corresponding observations, i.e.:

[0067] in, Indicates the first The credibility score corresponding to each observation. This represents the mapping function from credibility to weight. is an element in the matrix corresponding to the j-th credibility score. This represents the total number of observations actually involved in the localization solution in the current optimal candidate localization solution. Preferably, a confidence-weighted least squares solution can be performed iteratively, where the... The increment of the next iteration can be expressed as:

[0068] in, For the first iteration For theoretical observation function For position variables The Jacobian matrix. Update the position estimate based on the incremental values:

[0069] in, This is the position estimation result at time t. This outputs the position estimation result at time t+1. When the iteration increment satisfies the preset convergence condition, the current position estimation result is output. As the final UWB positioning coordinates.

[0070] In summary, the formula for calculating UWB positioning coordinates is:

[0071] in, Locate coordinates for UWB. Let the label coordinates be the variables to be estimated. To determine the optimal candidate solution Corresponding base station combination and observation combination The residual vector constructed from the observations actually used in the positioning solution. The weights are constructed based on the credibility scores corresponding to the observations that actually participated in the positioning solution.

[0072] like Figure 2 As shown, this embodiment also includes a trusted positioning method for shipboard indoor UWB multi-mode self-organizing network, the steps of which are as follows: Step A: Obtain raw UWB measurement data. The raw UWB measurement data includes one or more observations and the corresponding channel impulse response (CIR) sequence. The observations include one or more of the following: range observations, horizontal angle observations, and pitch angle observations.

[0073] Step B involves inputting the CIR sequence into the feature extraction network to extract the depth representation features corresponding to the observations.

[0074] Step C: Based on deep representation features, perform credibility assessment on one or more observations to obtain the credibility score corresponding to each observation.

[0075] Step D: Based on the confidence score corresponding to each observation, classify and categorize one or more observations for different processing levels.

[0076] Step E involves estimating and compensating errors in at least some of the observations based on deep characterization features, the observations processed through hierarchical triage, and their corresponding confidence scores, to obtain the compensated observations.

[0077] Step F: Based on the number of available base stations, the combination of base stations, and the type of observation at the current time, construct a set of candidate positioning solutions that meet the positioning solution conditions.

[0078] Step G: Calculate the location confidence of each candidate location solution based on the confidence score of each observation in the candidate location solution, and select the candidate location solution with the highest location confidence as the optimal candidate location solution.

[0079] Step H involves constructing weights based on the credibility scores of the observations that actually participated in the localization solution in the optimal candidate localization solution, and then performing localization calculations to obtain the final UWB localization coordinates.

[0080] like Figure 3 As shown, the shipboard indoor UWB multi-mode ad hoc network trusted positioning method of this embodiment takes the original measurement data as input and uses trustedness-driven observation processing and positioning calculation as the main line. First, the channel impulse response (CIR) sequence output by the UWB system, as well as distance, horizontal angle, and pitch angle observations, are acquired. The CIR sequence is then input into the Transformer depth representation network to extract depth features reflecting the propagation state. Based on this, the trustedness of each type of observation is evaluated to obtain the corresponding trustedness score. Then, each observation is processed in a hierarchical manner according to the trustedness score. High-trusted observations can directly participate in the positioning calculation, while medium-trusted or low-trusted observations can be combined with an error compensation model for error estimation and correction. Further, based on the set of base stations that can participate in positioning at the current time and the types of observations that each base station can provide, a candidate positioning solution set is dynamically constructed. The trustedness scores of the observations contained in each candidate positioning solution are aggregated to calculate the corresponding positioning trustedness. Finally, the candidate positioning solution with the highest trustedness is selected as the optimal candidate positioning solution. Finally, a weight matrix is ​​constructed based on the credibility scores of the observations actually involved in the positioning solution in the optimal candidate positioning solution, and the credibility-weighted least squares method is used for the final positioning solution, thus outputting the ship's indoor UWB ad hoc network positioning result. Through the above process, a seamless processing is achieved from CIR propagation feature perception, observation credibility assessment, hierarchical traffic diversion processing, error compensation, dynamic construction and optimization of candidate positioning solutions to the final credible positioning solution. This enables the positioning mode to adaptively adjust to changes in occlusion, multipath propagation, base station visibility, and observation integrity in the ship's indoor environment, thereby improving the accuracy, stability, and robustness of UWB positioning in complex ship indoor scenarios.

[0081] The process of depth characterization extraction and observation reliability assessment based on CIR sequences includes the definition and offline calibration of observation reliability, as well as the construction of an online observation reliability assessment model. In this embodiment, observation reliability is used to characterize the ability of a single observation to support the positioning result. Specifically, the reliability of an observation is defined as the conditional probability that, after a single observation participates in positioning under its current error state, the resulting positioning error does not exceed a preset positioning accuracy threshold. It can be expressed as:

[0082] in, Indicates the reliability of the observation. This indicates the positioning error corresponding to the observation used in the positioning process. This indicates the preset positioning accuracy threshold. This indicates the observation error corresponding to the observation. Preferably, the observation confidence value ranges from [0,1]. The larger the value, the higher the probability that the corresponding observation can be directly used for positioning and meets the preset accuracy requirements; the smaller the value, the greater the risk that the positioning result will not meet the accuracy requirements when the corresponding observation is directly used for positioning.

[0083] To obtain observation reliability calibration values ​​under different error levels, an observation reliability calibration model is constructed in the offline stage. The following uses distance observations as an example to illustrate the offline calibration process of observation reliability; the reliability calibration process for horizontal angle observations and pitch angle observations is similar. Specifically, a statistical correspondence between ranging error and positioning error can be constructed through simulation. Preferably, the location coordinates of multiple UWB base stations and one UWB tag are randomly generated within a preset positioning area, and the true distance between the tag and each base station is calculated based on the base station coordinates and tag coordinates. While keeping the remaining observation results as true values, different levels of ranging error are sequentially injected into a single ranging result to construct a ranging sample set of a single distance observation under different error states. Let the first... The ranging error corresponding to each error level is: Then the corresponding ranging sample can be represented as:

[0084] in, Indicates the first The actual distance corresponding to each distance observation. This indicates the ranging result after injection error. Indicates the first Distance measurement error at several error levels; in this embodiment... Can be used as an error state variable The specific implementation form is as follows. Preferably, the ranging error can be gradually varied within a preset error range according to a preset step size to cover the range of common ranging errors, for example, it can vary within the range of 0.001m to 5m.

[0085] For each set of ranging samples, the corresponding positioning result and positioning error are calculated using a positioning solution method. Preferably, a weighted least squares method can be used for positioning solution, and the corresponding positioning error can be expressed as:

[0086] in, Indicates the first In the second trial The positioning results corresponding to each error level This indicates the corresponding actual label position. This represents the corresponding positioning error. Furthermore, by repeating independent simulation experiments, the proportion of positioning errors not exceeding a preset positioning accuracy threshold at the same error level is statistically analyzed, and this proportion is used as the confidence level calibration value, which can be expressed as:

[0087] in, Indicates the first The confidence level corresponds to a certain error level. Indicates the number of repeated trials. This indicates the preset positioning accuracy threshold. The function represents the indicator function, taking a value of 1 when the condition within the parentheses is true, and a value of 0 otherwise. To improve the stability of the calibration results, multiple sets of repeated experiments can be performed on the above simulation process to obtain multiple sets of confidence calibration values ​​at the same error level. Furthermore, the discrete confidence calibration values ​​can be fitted to establish a mapping relationship between the observation error and the confidence calibration value, which can be expressed as:

[0088] in, As a confidence level calibration value, This represents the mapping function from error to confidence calibration value. Using the above method, reference calibration results for observation confidence at different error levels can be obtained, serving as a supervisory basis for subsequent online observation confidence assessment models.

[0089] After obtaining the confidence rating, an online observation confidence assessment model is constructed. The online observation confidence assessment model includes a deep representation extraction network and a confidence prediction network. The deep representation extraction network extracts deep features related to the propagation state from the CIR sequence, and the confidence prediction network outputs the confidence score of the corresponding observation based on the deep features. Preferably, the deep representation extraction network is implemented using a Transformer encoder. In one embodiment, the Transformer encoder can first undergo self-supervised pre-training based on unlabeled CIR sequences to learn the temporal structure features and propagation state representations in the CIR sequences; preferably, the self-supervised pre-training task includes one or more of a CIR sequence mask reconstruction task and a contrastive learning task. Based on this, the confidence rating generated by the offline calibration model is used as a supervision label to train the online observation confidence assessment model, enabling the model to learn the mapping relationship between CIR sequence features and observation confidence. Preferably, the model training loss function can be expressed as:

[0090] in, Indicates the number of training samples. Indicates the first The confidence prediction value for each sample Indicates the first The confidence rating value corresponding to each sample This represents the loss function between the predicted value and the calibration value. Set the loss function for model training. The loss function can be the mean squared error loss function, the cross-entropy loss function, or other loss functions suitable for confidence prediction.

[0091] During the online positioning phase, after obtaining the CIR sequence corresponding to the current time observation, it can be processed using the same preprocessing method as the offline phase and input into the trained online observation credibility assessment model to output the credibility score of the corresponding observation. For distance observations, horizontal angle observations, and pitch angle observations, corresponding credibility prediction branches can be set separately, or a unified prediction network can be used to output the credibility results of multiple types of observations.

[0092] Figure 4 This embodiment illustrates a typical cabin application scenario within a ship. Under complex structural conditions involving piping, bulkheads, electrical control cabinets, and equipment partitions, base stations 1 to 5 (A1 to A5) are deployed within the positioning area. The target to be located, carrying a UWB tag, moves within the cabin. Each base station can acquire distance measurements, horizontal angle measurements, pitch angle measurements, and the corresponding CIR sequence of the tag relative to its own base station. Due to the prevalence of obstructions, multipath propagation, metal reflections, and dynamic changes in base station visibility within the ship's indoor environment, the number of base stations available for positioning and the types of measurements vary at different times. Therefore, this invention can adaptively select different positioning calculation modes based on real-time observation conditions. In one specific implementation, when the tag is located in a locally open area, and a base station can simultaneously provide highly reliable distance, horizontal angle, and pitch angle observations, a candidate positioning solution can be constructed based on the observations corresponding to that base station, and the tag's three-dimensional position can be determined. When the tag is located in complex areas such as the middle of a cabin, a densely populated equipment area, or a doorway transition area, and multiple base stations are visible simultaneously, a candidate positioning solution can be constructed by combining the distance and angle observations from multiple base stations, and the optimal positioning mode can be selected for solution through positioning reliability evaluation. Therefore, this invention can dynamically switch between different positioning solution modes during continuous movement within the ship's interior, depending on changes in tag position, obstruction status, and observation integrity, thereby improving the accuracy, stability, and environmental adaptability of UWB positioning within the ship's interior. (Figure shows...) These are the elevation angles from the tag to base stations 1 through 5, respectively. These represent the horizontal angles from the tag to base station 1 through 5, respectively. These represent the distances from the tag to base stations 1 through 5, respectively.

[0093] In one specific embodiment, a simulated ship compartment is used as the positioning area, with dimensions of 18 m × 10 m × 5 m. Four UWB base stations, A1, A2, A3 and A4, are deployed inside the compartment, with coordinates A1 (0.8, 0.6, 2.9), A2 (17.2, 0.7, 2.8), A3 (1.1, 3.9, 2.7) and A4 (16.8, 4.0, 2.9), respectively. These coordinates are pre-calibrated and stored in the positioning processing unit.

[0094] At any moment The data obtained by the tag from base station A1 is as follows: ° ° The data obtained by the tag from the A2 base station is as follows: Unavailable Unavailable The data obtained by the tag from the A3 base station is as follows: ° Unavailable The data obtained by the tag from the A4 base station is as follows: Unavailable Unavailable . These are the distance observations from base station A1 to base station A5. These are the horizontal angle observations from base station A1 to base station A5. These are the elevation angle observations from base station A1 to base station A5. The CIR sequence is from base station A1 to base station A5.

[0095] The positioning processing unit performs in-depth characterization extraction and reliability assessment on the CIR sequences corresponding to each observation, and obtains the reliability score corresponding to base station A1. The credibility score corresponding to base station A2 is The credibility score corresponding to base station A3 is The credibility score corresponding to base station A4 is When a high confidence threshold is selected... Low confidence threshold It can be determined that all observations of A1 and the distance observation of A2 are high-confidence observations, the distance observation and horizontal angle observation of A3 are medium-confidence observations, and the distance observation of A4 is a low-confidence observation and is discarded. These are the reliability scores for the distance observations from base station A1 to base station A5. The reliability scores for the horizontal angle observations from base station A1 to base station A5 are respectively. The confidence scores are for the elevation angle observations from base station A1 to base station A5.

[0096] Furthermore, error compensation is performed on the medium-confidence observations of A3 to obtain the compensated distance observations as follows: The compensated horizontal angle observation is Based on this, a set of candidate localization solutions is constructed, and the localization reliability is calculated according to the reliability score of the observations in each candidate localization solution, thus obtaining candidate localization solution S1 ( S2 S3 ) and S4 ( The location confidence scores of the samples were 0.85, 0.86, 0.88, and 0.78, respectively. Therefore, S3, with the highest location confidence score, was selected as the optimal candidate location solution. Subsequently, a weight matrix was constructed based on the confidence scores of the observations actually involved in the location solution, and the confidence-weighted least squares method was used for iterative solution. After three iterations, the solution converged, and the estimated label position at that moment was (9.54, 2.18, 1.09), corresponding to the true label position (9.60, 2.10, 1.15), with a 3D location error of 0.12m.

[0097] Figure 5 The experimental results of the positioning error CDF curves shown indicate that, under the condition of progressively increasing interference levels in the three groups, the error of the unprocessed original positioning results increases significantly with increasing interference. However, after correction using the method of this invention, all error indicators are significantly reduced. Specifically, as shown in Table 1, the average error, root mean square error, and 90% positioning error of trajectory 1 are reduced by 70.19%, 60.31%, and 61.16%, respectively, and after correction, they are 0.045 m, 0.050 m, and 0.113 m; those of trajectory 2 are reduced by 68.72%, 57.73%, and 58.94%, respectively, and after correction, they are 0.076 m, 0.071 m, and 0.202 m; and those of trajectory 3 are reduced by 58.78%, 53.22%, and 59.27%, respectively, and after correction, they are 0.115 m, 0.167 m, and 0.380 m. Overall, the average error decreased from 0.224 m to 0.078 m, a reduction of 64.93%; the root mean square error decreased from 0.217 m to 0.096 m, a reduction of 55.76%; and the 90% positioning error decreased from 0.572 m to 0.231 m, a reduction of 59.49%. These results demonstrate that, even under conditions of gradually increasing interference, the present invention can effectively suppress the adverse effects of complex interference on positioning results, significantly improving the accuracy, stability, and robustness of shipboard indoor UWB positioning.

[0098] Table 1

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A shipboard indoor UWB multi-mode self-organizing network trusted positioning system, characterized in that, include: The raw data acquisition module is used to acquire observations and corresponding CIR sequences, wherein the observations include distance, horizontal angle, and pitch angle; A deep characterization extraction module is used to extract the deep characterization features of the CIR sequence; The credibility assessment module is used to assess the credibility of the depth representation features and obtain credibility scores corresponding to the distance observation, horizontal angle and pitch angle observations. The hierarchical distribution module is used to perform hierarchical distribution processing on the observations based on the confidence score, so as to obtain high confidence observations, medium confidence observations and low confidence observations; An error compensation module is used to compensate the medium-confidence observations based on the depth representation features and the confidence score of the medium-confidence observations. A candidate positioning solution construction module is used to construct candidate positioning solution models based on the number of base stations, base station combinations, and observation types. It then filters and constructs a set of candidate positioning solutions based on whether the rank of the Jacobian matrix corresponding to the observation equation of each candidate positioning solution model meets the requirements for three-dimensional positioning. The module also calculates the positioning reliability based on the reliability scores of the observations in the candidate positioning solution set. This reliability score is weighted, summed, and normalized by dividing by the weighted sum of the number of observations. The candidate positioning solution with the highest reliability is then selected as the optimal positioning solution. The positioning solution module is used to construct a weight matrix based on the confidence scores of each observation that actually participates in the positioning solution in the optimal positioning solution, and to obtain the three-dimensional positioning coordinates of the UWB tag by using the weighted least squares method based on the weight matrix.

2. The shipboard indoor UWB multi-mode self-organizing network trusted positioning system according to claim 1, characterized in that, The raw data acquisition module includes UWB base stations that measure UWB tags, forming a UWB base station list. The UWB base station list includes the observations of each UWB base station and the corresponding CIR sequences.

3. The shipboard indoor UWB multi-mode self-organizing network trusted positioning system according to claim 1, characterized in that, The deep representation extraction module is a Transformer encoder, which is obtained through self-supervised pre-training based on unlabeled CIR sequences. The pre-trained Transformer encoder encodes the CIR sequences to obtain deep representations. The self-supervised pre-training uses several pre-training tasks to learn the representations of unlabeled CIR sequences, including CIR sequence mask reconstruction and contrastive learning tasks.

4. The shipboard indoor UWB multi-mode self-organizing network trusted positioning system according to claim 1, characterized in that, The credibility assessment module inputs the deep representation features into the credibility assessment network that has been fine-tuned and trained, and outputs a credibility score for the observed data.

5. The shipboard indoor UWB multi-mode self-organizing network trusted positioning system according to claim 1, characterized in that, The hierarchical distribution module classifies observations with a confidence level below a first threshold as low-confidence observations, observations with a confidence level greater than or equal to a second threshold as high-confidence observations, and observations with a confidence level greater than or equal to the first threshold and lower than the second threshold as medium-confidence observations, wherein the first threshold is less than the second threshold.

6. The shipboard indoor UWB multi-mode self-organizing network trusted positioning system according to claim 1, characterized in that, The error compensation module inputs the depth characterization features, the confidence score of the medium-confidence observations, and the medium-confidence observations into the error compensation network, outputs the observation error estimate, and compensates the medium-confidence observations based on the observation error estimate.

7. The shipboard indoor UWB multi-mode self-organizing network trusted positioning system according to claim 1, characterized in that, The candidate localization solution construction module constructs observation equations based on the observations in the candidate localization solution mode, constructs observation functions based on all observation equations, and constructs the Jacobian matrix based on the observation functions and the three-dimensional coordinates of the target to be localized. Singular value decomposition is performed on the Jacobian matrix to obtain singular values. The number of singular values ​​greater than a third threshold is counted. When the number is greater than or equal to 3, the current candidate positioning solution mode meets the requirements of three-dimensional positioning solution. When the number is less than 3, it is eliminated.

8. The shipboard indoor UWB multi-mode self-organizing network trusted positioning system according to claim 1, characterized in that, The formula for calculating UWB positioning coordinates is: in, Locate coordinates for UWB. Let the label coordinates be the variables to be estimated. For the residual vector, The weight is denoted as .

Images