An unmanned aerial vehicle beacon inspection multi-source information fusion and collaborative prediction method and system

Abstract

The application discloses a kind of unmanned aerial vehicle navigation beacon inspection near-field multi-source information fusion and collaborative prediction method and system, comprising: abnormal point detection and elimination to navigation beacon historical telemetry data sequence, establish trend degradation model;Unmanned aerial vehicle is collected navigation beacon state parameter as anchor point data in close range by at least two near-field communication links;Different link anchor point data is fused using D-S evidence theory, and high confidence anchor point estimated value is obtained;With anchor point data as correction basis, trend model parameters are corrected by weighted least square method, wherein anchor point data weight is higher than historical data weight;State parameter is predicted based on the model after correction, and the comprehensive health index is calculated to evaluate the health grade of navigation beacon and dynamically adjust the inspection interval.The application solves the problem of diagnostic blind area during link interruption by the collaborative mechanism of "telemetry continuous trend + unmanned aerial vehicle anchor calibration", and uses high-precision field data to correct long-term trend model, effectively improves the state prediction accuracy of navigation beacon and the utilization efficiency of inspection resources.

Description

Technical Field

[0001] This invention relates to the fields of maritime safety and UAV inspection technology, specifically to a method and system for collaboratively predicting the status of navigation marks by using near-field multi-source information fusion from UAVs. Background Technology

[0002] Maritime navigation aids are fundamental aids to ensure safe navigation for ships. Currently, telemetry and remote control systems based on satellite communication and IoT technologies are widely used in navigation aid management, enabling remote monitoring of the main operational status of navigation aids. However, in actual operation, telemetry and remote control links are frequently interrupted due to factors such as communication obstruction, electromagnetic interference, and terminal failures. When the link is interrupted, the monitoring center cannot obtain the real-time status of the navigation aid, and cannot determine whether the cause of the failure is hardware damage, power system failure, or simply a communication link problem, resulting in a lack of basis for emergency repair decisions.

[0003] On the other hand, existing navigational aid status trend models based on telemetry data (such as linear decay models of battery state of charge) have limited accuracy. Telemetry data itself may contain anomalies introduced by sensor false alarms and communication interference, and the telemetry sampling frequency is usually low, causing the prediction error of the trend model to increase over time.

[0004] The International Association of Lighthouse Authorities (IALA) has issued guidelines for using drones for visual inspection of navigation aids, and pilot applications of drone navigation aid inspection have been carried out in China. However, the existing drone inspection solutions have the following shortcomings: (1) The detection mode is one-way observation, which is limited to visual appearance inspection and cannot obtain in-depth information such as internal electrical parameters and fault codes of navigation aids; (2) The drone inspection and the telemetry and remote control system operate independently and lack a mechanism to use high-precision field data to correct long-term trend models; (3) There is a lack of a unified framework for multi-source information fusion, and it is impossible to comprehensively utilize data from multiple near-field communication links such as Bluetooth, Beidou, and AIS. Summary of the Invention

[0005] Technical problems to be solved

[0006] This invention aims to address the problems of insufficient accuracy of telemetry trend models, diagnostic blind spots during link interruptions, and the disconnect between UAV inspection and telemetry systems in existing navigation mark status monitoring. It provides a navigation mark status prediction method that integrates high-precision UAV field data with historical telemetry trend models.

[0007] Technical solution

[0008] (I) Overall Technical Solution

[0009] This invention provides a method and system for multi-source information fusion and collaborative prediction of UAV navigation mark inspection, comprising the following steps:

[0010] S1: Historical Telemetry Data Preprocessing and Trend Modeling

[0011] The historical telemetry data sequence of the target beacon is obtained, which includes beacon status parameters acquired through telemetry links at multiple historical moments; outlier detection and removal are performed on the historical telemetry data sequence to obtain a cleaned telemetry data sequence; a trend degradation model of the beacon status parameters over time is established based on the cleaned telemetry data sequence.

[0012] S2: UAV near-field multi-link anchor point acquisition

[0013] During the drone inspection process, at least two near-field communication modules mounted on the drone are used to collect the status parameters of the navigation mark at close range as anchor point data; the at least two near-field communication modules correspond to different physical layer communication links.

[0014] S3: Multi-source anchor point evidence fusion

[0015] Evidence fusion is performed on anchor point data collected through different communication links to handle inconsistencies between data from different links and obtain fused high-confidence anchor point estimates.

[0016] S4: Anchor-based trend model correction

[0017] The high-confidence anchor point estimates are used as the basis for correction, and the parameters of the trend degradation model are corrected by a weighted regression method; wherein, the anchor point data has a higher weight in the weighted regression than the historical telemetry data points.

[0018] S5: Collaborative Prediction and Health Assessment

[0019] The modified trend degradation model is used to predict the values ​​of navigational aid status parameters at future times, and the health status level of the navigational aid is evaluated based on the prediction results.

[0020] (II) Detailed technical solutions for each step

[0021] Further options for step S1:

[0022] The historical telemetry data sequence is obtained through BeiDou short message service or mobile communication uplink of the navigation beacon telemetry and control system. The trend degradation model is a linear regression model or an exponential decay model.

[0023] Preferably, the outlier detection employs the isolated forest algorithm, which constructs a random binary search tree and calculates anomaly scores for each sample. Samples with anomaly scores exceeding a preset threshold are marked as outliers and removed. The preset threshold can be determined through cross-validation based on the statistical distribution characteristics of historical telemetry data.

[0024] Further options for step S2:

[0025] The at least two near-field communication links include at least two of the following links: Bluetooth link, BeiDou short message passive monitoring link, and AIS VHF receiving link.

[0026] The UAV adopts a progressive detection strategy from far to near: in the far field, it passively listens to the status parameters in the navigation beacon telemetry messages through Beidou short messages; in the medium field, it obtains the MMSI identifier and location information of the navigation beacon through AIS VHF reception; and in the near field, it actively connects to the built-in Bluetooth module of the navigation beacon through a Bluetooth link to read the full-function status parameters inside the navigation beacon's microcontroller.

[0027] The navigational aid status parameters include multiple electrical parameters (operating current, operating voltage, drive power, battery state of charge (SOC), battery health (SOH), optical parameters (luminous intensity, lamp quality cycle), communication parameters (AIS message update rate, MMSI consistency, coordinate consistency), and structural parameters (position deviation, structural integrity).

[0028] Further options for step S3:

[0029] The evidence fusion employs the Dempster-Shafer (DS) evidence theory. Specifically, it includes:

[0030] Establish an identification framework ,in Indicates the type of electrical system fault. Indicates the type of communication system fault. Representing location and structural anomaly categories; uncertainty is addressed by assigning basic probabilities to the entire set of the identification framework. To indicate, that is This indicates uncertainty as to which category the evidence belongs to.

[0031] A basic probability assignment function (BPA) is set for each communication link, with the mapping rule as follows: for links where electrical parameters are directly obtained, an assignment function is used. A higher BPA value; for links that primarily acquire communication status parameters, assign... A higher BPA value; for links that primarily acquire location information, assign... A high BPA value.

[0032] The Dempster composition rule is used to perform pairwise composition of the BPA of each link:

[0033]

[0034] in The conflict factor represents the degree of conflict between pieces of evidence, and is defined as the sum of the products assigned to disjoint propositions in the mass function of each piece of evidence:

[0035]

[0036] When K=0, the evidence is completely consistent; when K=1, the evidence is completely conflicting. When the conflict factor exceeds the preset conflict threshold, a conflicting evidence prompt is output.

[0037] The parameter estimate corresponding to the proposition with the highest confidence after fusion is taken as the high-confidence anchor point estimate.

[0038] Further options for step S4:

[0039] The weighted regression method is weighted least squares. Assuming there are N historical telemetry data points and M anchor data points (M≥1), the weighted objective function is constructed as follows:

[0040]

[0041] in Weights are assigned to historical data. For anchor data weights, and > .

[0042] Preferably, the weight ratio is dynamically determined based on the uncertainty of the anchor data. The uncertainty of the anchor data is calculated based on the trust function value after DS evidence fusion or the degree of consistency between the data from each link.

[0043] Further options for step S5:

[0044] The health status assessment uses the CHI (Comprehensive Health Index) model:

[0045]

[0046] in The number of state parameters participating in the evaluation. The weight of the i-th parameter is determined based on the Analytic Hierarchy Process (AHP). The standardized score for the i-th parameter (mapping the measured value to the interval [0,100]). It is the trend factor of the i-th parameter (reflecting the degradation rate of the parameter).

[0047] The health status of navigational aids is divided into multiple levels (such as healthy, good, warning, and dangerous) based on the CHI value, and the inspection interval of drones is dynamically adjusted according to the level and trend of CHI: when the rate of CHI decline exceeds the preset rate threshold, the inspection interval is shortened; when CHI remains stable, the inspection interval is maintained or extended.

[0048] (III) System Solution

[0049] This invention also provides a multi-source information fusion and collaborative prediction system for UAV navigation mark inspection, comprising:

[0050] Trend modeling module: used to acquire historical telemetry data sequences of target navigation marks, perform outlier detection and removal, and establish a trend degradation model of navigation mark status parameters over time;

[0051] Multi-link anchor point acquisition module: Deployed on the UAV, it includes at least two near-field communication modules, used to acquire the status parameters of the beacon as anchor point data within a short range;

[0052] Evidence fusion module: used to fuse anchor data collected through different communication links and output high-confidence anchor estimates;

[0053] Model correction module: used to inject high-confidence anchor point estimates into the trend degradation model and correct the model parameters through weighted regression;

[0054] Prediction and Assessment Module: Used to predict the status of navigation marks and assess their health level based on the modified model.

[0055] Beneficial effects

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] (1) Filling the diagnostic blind spot: UAVs can directly obtain the internal status parameters of the beacon when the telemetry link is interrupted by multi-link near-field acquisition, eliminating the diagnostic blind spot during the link interruption period and enabling maintenance personnel to accurately determine the cause of the fault.

[0058] (2) Improve prediction accuracy: By injecting high-precision anchor point data collected by UAVs on-site into the historical telemetry trend model and using weighted regression to correct the model parameters, the influence of noise and outliers in the telemetry data on the trend model is effectively suppressed, and the accuracy of long-term state prediction is significantly improved.

[0059] (3) Multi-source redundancy verification: Bluetooth, Beidou and AIS physical layer links serve as backups for each other. The inconsistency of multi-link data is processed by DS evidence theory, which improves data credibility and system robustness.

[0060] (4) Closed-loop continuous optimization: After each UAV inspection, the anchor point data is archived to the historical database, and the trend model is continuously updated to achieve a collaborative prediction closed loop of "continuous telemetry trend + sparse anchor point calibration".

[0061] (5) Resource optimization scheduling: Based on the dynamic inspection interval adjustment mechanism of comprehensive health index, the inspection frequency of UAVs is optimized and the operation and maintenance costs are reduced while ensuring safety. Attached Figure Description

[0062] Figure 1 Overall Flowchart (showing the complete execution flow and data flow of S1~S5)

[0063] Figure 2 Schematic diagram of a three-layer information interaction architecture (physical layer (Bluetooth / BeiDou / AIS) → protocol layer (JT / T 788 adaptation) → application layer (fusion / correction / prediction))

[0064] Figure 3 : Schematic diagram of progressive detection strategy (progressive relationship of far-field Beidou monitoring → mid-field AIS reception → near-field Bluetooth connection)

[0065] Figure 4 : Schematic diagram of the collaborative prediction framework (comparison of historical telemetry trend lines and trend lines after anchor point correction)

[0066] Figure 5 Closed-loop operation and maintenance process (from historical data cleaning → modeling → monitoring → inspection triggering → anchor point collection → fusion correction → prediction and evaluation → feedback archiving) Detailed Implementation

[0067] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0068] [Example 1: Method Flow]

[0069] See Figure 1 The method and system for multi-source information fusion and collaborative prediction of UAV navigation mark inspection provided in this embodiment includes the following steps:

[0070] Step 1: Historical Telemetry Data Preprocessing and Trend Modeling

[0071] Extract historical telemetry data sequences of target navigation marks from the navigation mark telemetry and remote control system database, including recorded values ​​of telemetry parameters such as battery state of charge (SOC), battery voltage, operating current, and position offset at each time point.

[0072] An isolated forest algorithm is used to detect outliers in historical telemetry data sequences. This algorithm constructs multiple binary search trees (iTrees) by randomly selecting features and segmentation values. For each sample, the average path length across all trees is calculated to obtain an anomaly score. A score closer to 1 indicates a higher likelihood of the sample being an outlier. Samples with anomaly scores exceeding a threshold are marked as anomalies and removed from the sequence. This threshold can be determined using hold-out or cross-validation based on the statistical distribution characteristics of the historical telemetry data.

[0073] Based on the cleaned telemetry data sequence, a trend degradation model is established using linear regression or exponential regression methods. Taking battery SOC as an example, assume the cleaned telemetry sequence contains N data points. Establish a linear model The initial trend degradation model is obtained by estimating parameters a and b using the least squares method.

[0074] Step 2: UAV near-field multi-link anchor point acquisition

[0075] When the drone arrives near the navigation mark during its inspection mission, it performs multi-link data collection according to a progressive detection strategy from far to near:

[0076] Within the far-field range, the BeiDou short message receiving module carried by the UAV passively listens to the uplink telemetry messages transmitted by the beacon at the regular frequency, and analyzes and obtains location information, voltage information, light status information, etc.

[0077] Within a medium distance range, the VHF receiver module on the UAV listens for AtoN messages transmitted by navigation marks on the AIS standard frequency, and decodes them to obtain the MMSI identifier, coordinate information and navigation mark type;

[0078] Within the near-field range, the drone actively scans and connects to the Bluetooth slave module built into the beacon light via its onboard Bluetooth host module. It reads the full-function status parameters inside the beacon microcontroller (MCU) through the standard service protocol, including but not limited to: real-time SOC, battery health SOH, charging current, operating voltage, operating current, drive power, fault codes, etc.

[0079] The parameters collected via different links during the same inspection period constitute the anchor point dataset for this inspection. Because the Bluetooth link directly reads the MCU's internal registers, its data accuracy and integrity are superior to the passive monitoring link, and it is unaffected by interruptions in the telemetry communication link.

[0080] Step 3: Multi-source anchor point evidence fusion

[0081] Since different communication links may provide inconsistent parameter estimates, it is necessary to fuse the anchor data from multiple links.

[0082] This embodiment uses the DS evidence theory for data fusion. First, an identification framework is established. ,in This indicates an electrical system fault (including battery, drive power supply, and light source module). This indicates a communication system malfunction (including BeiDou module and AIS module). The location and structural anomalies (including drift, collision, and lampshade damage) are represented by m(Θ). When the evidence cannot determine the specific fault category, BPA is assigned to the entire set Θ.

[0083] Based on the physical characteristics of each communication link, a basic probability assignment (BPA) is set: the Bluetooth link directly obtains the internal electrical parameters of the MCU, and... Assigning a high level of trust; BeiDou telemetry messages mainly reflect communication and location status, for Assign a high level of trust; the core content of the AIS message is MMSI and coordinates, for Give it a high degree of trust.

[0084] The Dempster synthesis rule is used to synthesize the BPAs of the Bluetooth, BeiDou, and AIS links pairwise, and the mass function of the fused data is calculated. When the conflict factor K is large (indicating serious contradictions between the evidence), a discount factor can be introduced to correct the BPAs of each link before synthesis, or a contradiction warning can be output for manual review.

[0085] After fusion, the fault category with the highest confidence is selected, and its corresponding parameter estimate (such as the fused SOC value) is used as the high-confidence anchor point estimate for subsequent model correction.

[0086] Step 4: Trend Model Correction Based on Anchor Points

[0087] Let the initial trend model established in step one be... Where a and b are the model parameters to be corrected. Assume there are N historical telemetry data points. The inspection yielded M high-confidence anchor points obtained through DS fusion. The weights of historical data points are taken as follows: The weight of the anchor data point is taken as .

[0088] Construct the weighted least squares objective function:

[0089]

[0090] in > This reflects the high reliability of the anchor data. The weight ratio can be dynamically determined based on the uncertainty of the anchor data: if the consistency of each link after DS fusion is high and the trust function value is large, a larger weight ratio can be taken; if the data of each link differs greatly, the weight ratio should be appropriately reduced.

[0091] By minimizing the objective function The corrected model parameters are obtained by solving. and The corrected trend degradation model was obtained. This model retains long-term trend information from historical telemetry data while incorporating high-precision anchor points collected by UAVs in the field, achieving collaborative prediction through "continuous trend + sparse calibration".

[0092] Step 5: Collaborative Prediction and Health Assessment

[0093] Based on the modified trend degradation model, the values ​​of navigational aid state parameters are predicted at future target times. For example, based on the modified SOC linear decay model, the SOC value of the navigational aid on a future day is predicted and compared with a preset low battery alarm threshold to determine in advance whether maintenance needs to be arranged.

[0094] Meanwhile, the Comprehensive Health Index (CHI) is used to quantitatively assess the overall health status of navigation aids. The CHI calculation considers multiple core status parameters (SOC, SOH, operating voltage, drive power, luminous intensity, lamp quality cycle, AIS update rate, position deviation, etc.). Each parameter is converted into a score according to a preset weight and standardized mapping function, and the weighted sum is used to obtain the CHI value.

[0095] CHI adopts a grading standard that intuitively corresponds to the health status of navigational aids, such as dividing it into four levels: healthy, good, warning, and dangerous. When CHI enters the warning level or the rate of decline of CHI exceeds a preset threshold, the interval between the next UAV inspection is automatically shortened; when CHI maintains a healthy level for a long period, the inspection interval can be appropriately extended to achieve optimal resource allocation.

[0096]

Example 2: System Composition

[0097] See Figure 2 This embodiment provides a method and system for multi-source information fusion and collaborative prediction of UAV navigation mark inspection, including the following modules:

[0098] Trend Modeling Module: Deployed on a shore-based server or cloud computing platform, this module acquires historical telemetry data from the navigation beacon telemetry and remote control system, performs isolated forest anomaly detection and data cleaning, and establishes and maintains trend degradation models for each navigation beacon. This module also manages the product database (storing factory nominal parameters), historical inspection database (storing inspection records), and fault knowledge base (storing the mapping relationship between fault modes and anomaly characteristics) for each navigation beacon.

[0099] Multi-link anchor point acquisition module: See Figure 3Deployed on an unmanned aerial vehicle (UAV) platform, the module includes a Bluetooth communication module, a BeiDou short message receiving module, and an AIS VHF receiving module. Following a progressive detection strategy, this module sequentially activates BeiDou monitoring, AIS reception, and Bluetooth connection as the UAV approaches the beacon, collecting full-function status parameters of the beacon. The collected parameters include telemetry message content defined in the appendix of the JT / T 788-2023 standard and MCU internal parameters read directly from the Bluetooth link.

[0100] Evidence fusion module: See Figure 4 This module performs time alignment and evidence fusion on data from different links acquired by the multi-link anchor point acquisition module. It executes BPA assignment, Dempster synthesis, and conflict detection according to DS evidence theory, outputting high-confidence anchor point estimates. When the conflict factor exceeds a threshold, an evidence conflict alarm is generated.

[0101] Model Correction Module: Receives the high-confidence anchor point estimate output by the evidence fusion module, reads the current trend degradation model parameters from the trend modeling module, performs weighted least squares correction calculation, updates the model parameters, and writes them back to the trend modeling module.

[0102] Prediction and Assessment Module: Based on a modified trend degradation model, this module predicts future conditions, calculates the Comprehensive Health Index (CHI), and outputs health level assessment results and maintenance recommendations. This module is also responsible for dynamically adjusting inspection plans.

[0103] Optional knowledge graph reasoning module: See Figure 5 A knowledge graph of navigational aid faults is constructed, storing fault cases in the form of triples (fault mode - association - abnormal features or handling suggestions). When an abnormal parameter combination is detected, reasoning is performed on the knowledge graph using a graph attention network to locate the most likely root cause of the fault and recommend handling measures. After each inspection, the diagnostic conclusions and actual maintenance results are added to the knowledge graph as new samples to continuously optimize the reasoning ability.

[0104] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural modifications made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method and system for multi-source information fusion and collaborative prediction in UAV navigation mark inspection, characterized in that, Includes the following steps: S1: Obtain the historical telemetry data sequence of the target beacon, which includes beacon status parameters acquired through telemetry links at multiple historical moments; perform outlier detection and removal on the historical telemetry data sequence to obtain a cleaned telemetry data sequence; establish a trend degradation model of the beacon status parameters over time based on the cleaned telemetry data sequence. S2: During the UAV inspection process, at least two near-field communication modules mounted on the UAV are used to collect the status parameters of the beacon at close range as anchor point data; the at least two near-field communication modules correspond to different physical layer communication links. S3: Perform evidence fusion on anchor point data collected through different communication links to obtain the fused high-confidence anchor point estimate; S4: Using the high-confidence anchor point estimate as the basis for correction, the parameters of the trend degradation model are corrected by a weighted regression method to obtain the corrected trend degradation model; wherein, the weight of the anchor point data in the weighted regression is greater than the weight of the historical telemetry data points; S5: Based on the modified trend degradation model, predict the values ​​of the navigation mark state parameters at future times.

2. The method according to claim 1, characterized in that, The at least two near-field communication links include at least two of the following links: Bluetooth link, BeiDou short message passive monitoring link, and AIS VHF receiving link.

3. The method according to claim 2, characterized in that, In step S2, the UAV adopts a progressive detection strategy from far to near: in the far field, it passively listens to the Beidou short message to obtain the status parameters in the navigation beacon telemetry message; in the medium field, it obtains the MMSI identifier and location information of the navigation beacon through AIS VHF reception; and in the near field, it actively connects to the built-in Bluetooth module of the navigation beacon through a Bluetooth link to read the status parameters inside the navigation beacon.

4. The method according to claim 1, characterized in that, The outlier detection in step S1 uses the isolated forest algorithm, which constructs a random binary search tree and calculates the outlier score for each sample. Samples with outlier scores exceeding a preset threshold are marked as outliers and removed.

5. The method according to claim 1, characterized in that, The evidence fusion in step S3 adopts the DS evidence theory, specifically including: An identification framework is established, which includes categories of electrical system faults, communication system faults, location and structural anomalies, and uncertainties. A basic probability assignment function (BPA) is set for each communication link. The mapping rules are as follows: for links that mainly acquire electrical parameters, a higher BPA value is assigned to the electrical system fault category; for links that mainly acquire communication status parameters, a higher BPA value is assigned to the communication system fault category; for links that mainly acquire location information, a higher BPA value is assigned to the location and structural anomaly category. The BPA of each link is synthesized using Dempster synthesis rules; The parameter estimate corresponding to the proposition with the highest confidence level after fusion is taken as the high-confidence anchor point estimate.

6. The method according to claim 5, characterized in that, When the conflict factor in the Dempster synthesis rule exceeds the preset conflict threshold, an evidence contradiction prompt is output and manual review is triggered.

7. The method according to claim 1, characterized in that, In step S4, the weighted regression method is weighted least squares, and the ratio of historical data weight to anchor data weight is dynamically determined based on the uncertainty of the anchor data; the uncertainty of the anchor data is calculated based on the trust function value after evidence fusion or the degree of consistency between the data in each link.

8. The method according to claim 1, characterized in that, Following step S5, the following is also included: S6: Calculate the Comprehensive Health Index (CHI) based on multiple status parameters of the navigation beacon, wherein the CHI is the weighted sum of the standardized scores of each status parameter and their corresponding weights; S7: Dynamically adjust the UAV inspection interval according to the value and trend of CHI. When the CHI descent rate exceeds the preset rate threshold, shorten the inspection interval.

9. A method and system for multi-source information fusion and collaborative prediction in UAV navigation mark inspection, characterized in that, include: The trend modeling module is used to acquire historical telemetry data sequences of target navigation marks, perform outlier detection and removal, and establish a trend degradation model of navigation mark status parameters over time. A multi-link anchor point acquisition module, deployed on a drone, includes at least two near-field communication modules for acquiring the status parameters of navigation marks as anchor point data within a short range; The evidence fusion module is used to fuse anchor data collected through different communication links and output high-confidence anchor estimates. The model correction module is used to inject the high-confidence anchor point estimate into the trend degradation model and correct the model parameters through weighted regression, wherein the weight of the anchor point data in the weighted regression is greater than the weight of the historical telemetry data points. The prediction and assessment module is used to predict the status of navigation marks and assess their health level based on the modified model.

10. The system according to claim 9, characterized in that, The multi-link anchor point acquisition module includes at least two of the following: a Bluetooth communication module, a Beidou short message receiving module, and an AIS VHF receiving module. The system also includes a knowledge graph reasoning module, which is used to construct a navigation beacon fault knowledge graph based on historical fault cases, and to perform fault root cause reasoning on the knowledge graph through a graph neural network when abnormal parameters are detected. After each inspection, the diagnostic conclusion is added to the knowledge graph as a new sample, and the node association weights are updated to achieve continuous learning.

Images