An autonomous underwater vehicle risk coupling analysis method and system

Abstract

This invention discloses a risk coupling analysis method and system for autonomous underwater vehicles (AUVs), belonging to the field of AUVs. The invention involves offline preprocessing and data augmentation of historical multi-source heterogeneous data to generate an enhanced dataset and calibrate key model parameters; real-time acquisition and online fusion and preprocessing of sensor data to identify the current mission stage; construction of risk situation units based on the corresponding risk physical boundary thresholds according to the mission stage; cross-stage risk inheritance calculation using a risk inheritance matrix during mission stage switching; activation of the dynamic coupling matrix corresponding to the current mission stage for coupling evolution deduction to calculate the propagation and amplification effects of risk among multi-source subsystems; and comprehensive evaluation of the overall risk level and output of graded early warning information. This invention can effectively characterize the dynamic coupling relationship between multi-source risks, enabling real-time assessment and early warning of operational risks for AUVs.

Description

Technical Field

[0001] This invention belongs to the field of autonomous underwater vehicle technology, specifically relating to a risk coupling analysis method and system for autonomous underwater vehicles. Background Technology

[0002] Autonomous underwater vehicles (AUVs) are unmanned underwater platforms that integrate navigation, control, sensing, and mission payloads. They have been widely used in marine resource exploration, seabed topography mapping, and environmental monitoring. With the increasing demand for deep-sea development, AUVs are operating at greater depths and with significantly longer mission cycles. Their system structure and functions are becoming increasingly complex, and they face multiple challenges during operation, including high pressure, low temperature, weak communication, and unknown obstacles. This places high demands on the reliability and environmental adaptability of the system.

[0003] Due to the complex and highly uncertain operating environment of AUVs, significant dynamic interactions exist between their internal subsystems and between the system and the external environment, resulting in diverse and coupled sources of risk. On the one hand, marine environmental factors such as ocean current disturbances, density stratification, and sea ice cover directly affect the AUV's motion stability and navigation accuracy. On the other hand, equipment-level issues such as sensor drift, actuator performance degradation, energy depletion, and communication interruptions adversely affect the system's perception, control, and decision-making capabilities. Furthermore, inappropriate configuration of mission planning and control strategies can also induce system operational risks. These risk factors overlap and exhibit coupled evolutionary characteristics at different mission stages and operational states, thereby increasing the likelihood of mission failure or system malfunction. Therefore, a systematic coupled analysis of multi-source risks during AUV operation is crucial for improving its safety and mission success rate.

[0004] Existing research on risk analysis for underwater vehicles (AUVs) largely focuses on modeling single risk factors or analyzing local failures. These methods often rely on prior assumptions, making it difficult to fully characterize the dynamic relationships between multiple risk sources. Furthermore, they lack comprehensive utilization of mission phase changes, system state evolution, and multi-channel sensing data. Simultaneously, with the rapid growth in the scale and types of AUV operational data, traditional risk analysis methods neglect the dynamic characteristics of the state changes of each component over time during system operation, making it difficult to describe the complex nonlinear coupling relationships between subsystems. This results in analysis results often lagging behind the real-time evolution of the system state. Based on these findings, this invention discloses a risk coupling analysis method and system for autonomous underwater vehicles (AUVs). Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention proposes a risk coupling analysis method and system for autonomous underwater vehicles. The method and system are rationally designed, overcome the shortcomings of existing technologies, and have good results.

[0006] A risk coupling analysis method for autonomous underwater vehicles, characterized by the following steps: S1. Collect historical multi-source heterogeneous data of autonomous underwater vehicles, perform offline multi-channel data fusion and standardized preprocessing, and construct a standard dataset; S2. Perform physical constraint-based data augmentation on the scarce fault samples processed in S1 to generate an augmented dataset; S3. Use the augmented dataset to perform offline system identification and parameter calibration, calculate and store the key model parameters required for online analysis; the key model parameters include the risk physical boundary thresholds of each subsystem, the dynamic coupling matrix library for different task stages, and the risk inheritance matrix for cross-stage transition; S4. Real-time acquisition of sensor data from underwater vehicles, and online real-time data fusion and preprocessing that meet causal constraints; S5. Based on the current navigation status and control commands, identify the current mission phase of the vehicle in real time and generate a mission phase identifier; S6. Based on the task stage identifier, call the corresponding risk physical boundary threshold to construct a risk situation unit vector containing state deviation and trend change rate; S7. When a task phase switch is detected, perform cross-phase risk inheritance calculation using the risk inheritance matrix to determine the initial risk state of the new phase; S8. Activate the dynamic coupling matrix corresponding to the current task stage identifier, perform dynamic coupling evolution deduction on the risk situation unit vector, and calculate the propagation and amplification effect of risk among multi-source subsystems; S9. Comprehensively assess the overall risk level after evolution, and output graded early warning information accordingly.

[0007] Furthermore, the historical multi-source heterogeneous data of the autonomous underwater vehicle in S1 includes state observation data, which originates from multiple physical parameters corresponding to each subsystem. Among them, the energy and power subsystem includes voltage, current and temperature parameters; the propulsion subsystem includes motor speed, torque, power and thermal state parameters; the emergency safety subsystem includes cabin leakage sensor level parameters and cabin pressure parameters; the environmental perception system includes distance characteristic parameters of forward-looking / obstacle avoidance sonar; the communication subsystem includes underwater acoustic communication signal-to-noise ratio parameters and data transmission delay parameters; and the navigation and positioning subsystem includes three-axis velocity components of the Doppler velocities and angular velocity and linear acceleration components of the inertial measurement unit. The above six subsystems contain a total of 15 types of physical parameters.

[0008] Furthermore, S2 specifically includes: S2.1. Fault injection based on the AUV six-degree-of-freedom dynamic model, using physical equations to generate sudden fault samples, i.e., hard fault samples, including: Establish the six-degree-of-freedom motion equations of the AUV in volume coordinates: ; in, This is the velocity vector in the carrier coordinate system; This is the pose vector in the geodetic coordinate system; The inertia matrix; For the Coriolis and centripetal force matrix; This is the hydrodynamic resistance matrix; The restoring force vector; The vector of total thrust and moment acting on the hull Establish a fault injection model: ; in, This represents the actual thrust vector acting on the hull after the failure occurs. The desired thrust vector emitted by the control system; It is the identity matrix; This is the failure performance loss matrix; This represents the environmental disturbance vector. S2.2. Standard dataset constructed using S1 The system utilizes real-world, historically scarce fault samples and employs a physical information generative adversarial network to generate gradually varying, nonlinear sensor faults and complexly coupled fault samples, i.e., soft fault samples, including: Construct a generative adversarial network architecture and build a system containing generators. and discriminator Time series generation model; The input uses real historical rare fault samples as seed data. Generator input random noise and seed data Output the generated virtual data sequence : ; Discriminator Automatically distinguishes between extracted real historical scarce fault samples and virtual fault data sequences. ; In the network loss function Add physical constraint terms: ; in, To generate the original adversarial loss for the adversarial network; These are the weighting coefficients for physical constraints; For physical constraint terms, the expression is: ; in, , These are the virtual velocities and accelerations generated by the generative adversarial network, respectively. , , These represent the virtual power, voltage, and current generated by the generative adversarial network, respectively. S2.3. Perform physical feasibility filtering on the hard fault samples and soft fault samples generated in S2.1 and S2.2 respectively. If the physical feasibility judgment condition is met, it is considered a valid sample. Valid hard fault samples are denoted as... Valid samples of soft faults are denoted as Store in augmented dataset If the conditions are not met, the data is discarded and regenerated. The physical feasibility assessment criteria include: The state continuity test is expressed as follows: ; In the formula, for The Middle One element, The smoothness threshold is calculated using the following formula: ; In the formula, for The arithmetic mean of the first-order difference sequence of normal standardized data; This is the standard deviation of the first-order difference sequence; This is the preset tolerance coefficient; The physical limit threshold test is expressed as follows: ; In the formula, For the generated virtual state sample values, , The set physical limit threshold; The fault feature identifiability test is expressed as follows: ; In the formula, The preset system noise threshold, To obtain from standard datasets The normal baseline data sequence extracted and aligned with the length of the generated virtual fault data sequence.

[0009] Furthermore, S3 specifically includes: S3.1. Enhance the dataset The unordered time-series data was divided into four mission phases: dive, cruise, near-bottom operation, and surfacing recovery. The K-Means clustering algorithm was used to partition the data based on its features, thus dividing the entire dataset... Divided into The clustering objective function for subsets is: ; in, This is a measure of the total bias in the clustering partitions; For the first The data cluster set for each task phase; To enhance the dataset A single state sample vector in; For the first The centroid vector of each cluster center; S3.2. Calibrate the risk physical boundary thresholds of the energy and power subsystem, propulsion subsystem, emergency safety subsystem, environmental perception subsystem, communication subsystem, and navigation and positioning subsystem of the underwater vehicle. Using the fault samples and extreme condition samples contained in the enhanced dataset, determine the safety envelope of each physical parameter corresponding to each subsystem using the following formula: The upper threshold is calculated using the following formula: ; The lower bound threshold is calculated using the following formula: ; in, , The first The upper and lower limits of the risk physical boundary for each physical parameter; This is a subset of normal states from the standard dataset after removing historical faults and abnormal records. All samples included in this subset are from the first set of samples. A set of numerical values ​​for physical parameters; and These are the arithmetic mean operator and the standard deviation operator, respectively; This refers to the safety margin factor. S3.3. The degree of mutual influence between subsystems varies in different task phases, so an independent dynamic coupling matrix is ​​constructed for each task phase. , The coupling strength was quantified using the Pearson correlation coefficient matrix. elements in Indicates the first Under the first task phase, the first The physical parameter and the first The dynamic coupling strength between the physical parameters is expressed as: ; in, For the first The set of all moments corresponding to each task phase; , for The first moment and the Data values ​​for each physical parameter; S3.4. To address the risk transmission issue during task switching, time-delay cross-correlation analysis is used to analyze the data characteristics during phase transitions, ultimately forming a cross-phase risk inheritance matrix. Used for initial state correction during online analysis; The risk inheritance factor is calculated using the following formula: ; in, As a risk inheritance factor, it represents the risk inheritance rate of different subsystems of the AUV from the previous mission phase. The physical parameters affect the current mission phase. The influence weight of each physical parameter; and These are the end time of the previous task phase and the start time of the current task phase, respectively. The normalized state data values ​​of the j-th physical parameter and the j-th physical parameter are directly obtained from the augmented dataset D. aug Read from; For mathematical expectation operators; This corresponds to the normalized standard deviation; This indicates the end time of the previous task phase, and the different subsystems at the [missing information]. Historical statistical average values ​​of several physical parameters; Indicates the start time of the current task phase, and the different subsystems' first... The historical statistical average of each physical parameter.

[0010] Further, S4 specifically includes: cleaning the sensor data of the physical parameters of each subsystem acquired in real time using a moving average filtering method; calculating the local window mean by constructing a sliding window and comparing it with the current acquired value to identify and remove instantaneous noise; performing online multi-channel data fusion on the cleaned real-time data; using the current system master clock as a reference, mapping the data of each sensor channel to the same moment using first-order linear extrapolation; and using extended Kalman filtering for state prediction and correction; performing real-time normalization processing on the fused data; and calling the historical extreme values ​​stored offline to map the real-time state vector to the standard space to generate a standardized holographic state vector. ; S5 specifically includes: S5.1. Extract key variables from the real-time fused state data and control system output in step S4, and construct a state discrimination vector for task phase identification. : ; In the formula, For the spacecraft at time The depth of navigation; The rate of change of the vehicle's depth; For sailing speed; The rate of change of the vehicle's attitude angle; It represents the task control command quantity, indicating the current low-level control mode such as fixed depth mode and fixed height mode; S5.2. Based on the preset stage discrimination threshold parameter, for Perform the following logical checks to generate the task stage representation for the current moment. : When the following formula is satisfied, it is determined to be in the diving phase. : > and ; In the formula, The preset threshold for the rate of change of depth during the descent phase. For relay depth; When the following formula is met, it is determined to be in the cruise phase. : ≤ and ; In the formula, The preset depth change rate threshold for the cruise phase. The cruise speed threshold; When the following formula is met, it is determined to be in the near-bottom operation phase. : and ; In the formula, For the target operating depth, To allow for error, Low speed threshold; When the following formula is met, it is determined to be in the buoyancy recovery phase. : ; S5.3. Based on the above judgment results, output the task stage identifier at the current moment: .

[0011] Furthermore, S6 specifically includes: S6.1. Calculate the state deviation: ; In the formula, For the first A physical parameter in Degree of deviation from state at any given moment; For the currently identified task stage The mean of the cluster centers of this physical parameter; For the risk physical boundary threshold determined in step S3.2, if Then take the upper bound. Otherwise, take the lower bound. ; S6.2. Calculate the risk of trend change rate: ; In the formula, Risk value for the rate of change of trend; The sampling time interval; The smoothness threshold calculated in step S2.3; S6.3. Constructing the risk posture unit vector: ; ; in, For the first Independent risk values ​​for each physical parameter; This is the deviation weighting coefficient; For the current moment An initial risk situation unit vector composed of physical parameters.

[0012] Furthermore, S7 specifically includes: when the system detects a change in the task phase, it uses the risk inheritance matrix to correct the current risk situation, expressed as: ; In the formula, This is the risk situation vector after inheritance and modification; This represents the risk situation vector from the previous moment. This is the time decay factor.

[0013] Furthermore, S8 specifically includes: the dynamic coupling evolution calculation method for the risk situation unit vector is as follows: ; In the formula, Let M be the evolved global coupling risk vector, with dimension M. The corrected risk vector is output from step S7; It is the identity matrix; This is the coupling gain coefficient; For the construction in step S3.3, which is related to the current task stage The corresponding dynamic coupling matrix.

[0014] Furthermore, S9 specifically includes: S9.1. Calculate the overall risk level of AUVs: ; In the formula, The overall risk level of the AUV at the current moment; Global coupling risk vector The first in Risk elements of each physical parameter; S9.2. Output a level three warning signal : ; in, and The pre-defined risk level warning line; S9.3. Based on the comprehensive risk level of AUVs output in S9.1, predict the risk growth rate based on a sliding time window, specifically as follows: Set the prediction window length to Constructing historical risk sequences : ; in, The sampling time within the sliding time window; The sampling time calculated from S9.1 Overall risk level value; Calculate the risk growth rate: ; In the formula, Given the current rate of risk growth, if A value greater than 0 indicates that the risk is worsening. ≤0 indicates that the risk tends to converge or decrease; It is the arithmetic mean of the time series within the time window; It is the arithmetic average of the overall risk level within the time window; S9.4. Predict the remaining time required for the AUV to reach its physical failure limit, specifically: Define the physical failure limit threshold of AUV The threshold is higher than ; Calculate the remaining safe time window: ; In the formula, The remaining safe time window as assessed; This is a preset system failure limit threshold; The current rate of risk growth; It is a safe and conservative factor.

[0015] An autonomous underwater vehicle (AUV) risk coupling analysis system is provided to implement the AUV risk coupling analysis method described above, comprising: The data acquisition unit is used to collect historical multi-source heterogeneous data of autonomous underwater vehicles, as well as real-time sensor data of underwater vehicles; The data preprocessing unit is used to perform offline multi-channel data fusion and standardization preprocessing on the historical multi-source heterogeneous data, and to perform online real-time data fusion and preprocessing on the real-time sensor data to meet causal constraints. The data augmentation unit is used to perform physical constraint-based data augmentation on preprocessed scarce fault samples to generate an augmented dataset. The parameter calibration unit is used to perform offline system identification and parameter calibration using the augmented dataset, calculate and store the key model parameters required for online analysis; the key model parameters include the risk physical boundary thresholds of each subsystem, a dynamic coupling matrix library for different task stages, and a risk inheritance matrix for cross-stage transition; The mission phase identification unit is used to identify the current mission phase of the aircraft in real time based on the current navigation status and control commands, and generate a mission phase identifier. The risk situation construction unit is used to construct a risk situation unit containing state deviation and trend change rate by calling the corresponding risk physical boundary threshold according to the task stage identifier. The risk inheritance calculation unit is used to perform cross-stage risk inheritance calculation using the risk inheritance matrix when a task stage switch is detected, and to determine the initial risk state of the new stage. The coupled evolution and deduction unit is used to activate the dynamic coupling matrix corresponding to the current task stage identifier, perform dynamic coupled evolution and deduction on the risk situation unit, and calculate the propagation and amplification effect of risk among multi-source subsystems. The risk assessment output unit is used to comprehensively assess the overall risk level after evolution and output graded early warning information accordingly.

[0016] The beneficial technical effects of this invention are as follows: This invention addresses the bottleneck of sufficient normal samples but scarce fault samples in autonomous underwater vehicle (AUV) risk analysis by constructing a risk coupling analysis framework that combines offline and online methods. By introducing a fault injection method based on a six-degree-of-freedom dynamic model and a data augmentation mechanism combining physical constraint generative adversarial networks, the coverage of fault samples can be expanded without relying on a large number of real fault tests. This provides a sufficient data foundation for subsequent system identification, threshold calibration, and risk modeling, thereby improving the stability and reliability of risk analysis results. Secondly, a mission phase constraint mechanism is introduced into risk modeling, dividing the complete operation process of the AUV into diving, cruising, near-bottom operation, and surfacing recovery phases. For each phase, risk physical boundary thresholds, dynamic coupling matrices, and risk inheritance matrices are constructed, effectively addressing the phase mismatch problem caused by the use of uniform thresholds and fixed models in traditional risk analysis methods. This allows risk assessment to more accurately reflect the true risk characteristics of the vehicle under different operating conditions. Finally, this invention proposes a risk situation unit and a dynamic coupling evolution mechanism, breaking through the approach of providing independent early warning for only a single subsystem. By quantifying the propagation, amplification, and suppression relationships of risks among subsystems, this invention effectively reveals the coupling effects between different subsystems such as energy power, propulsion, environmental perception, and communication and navigation. This allows for the identification of potential paths from local anomalies to system-level risks, providing a more comprehensive basis for integrated safety assessments. The invention introduces risk inheritance calculations across mission phases, addressing the issue of discontinuous risk states during mission switching and avoiding misjudgments caused by phase resets or information loss. This ensures the continuity and physical rationality of risk assessment results across time. Finally, at the risk output level, this invention not only provides tiered early warning results but also combines risk growth rate and remaining safety time predictions to provide quantitative and interpretable decision-making basis for the mother ship or airborne decision-making system, facilitating emergency intervention, mission adjustment, and safe recovery operations. Attached Figure Description

[0017] Figure 1 This is a flowchart of a risk coupling analysis method for autonomous underwater vehicles. Detailed Implementation

[0018] The specific embodiments of the present invention will be further described below with reference to specific examples: A risk coupling analysis method for autonomous underwater vehicles, such as Figure 1 As shown, it includes the following steps: S1. Collect historical multi-source heterogeneous data from autonomous underwater vehicles, perform offline multi-channel data fusion and standardized preprocessing, and construct a standard dataset; S1 specifically includes: S1.1: Collect historical multi-source heterogeneous data of autonomous underwater vehicles and classify it into the following three categories: The first category is offline knowledge base data, including equipment design parameters, theoretical performance models, expert experience rules, typical scenario patterns, and preset physical thresholds for mission stage discrimination; the second category is motion and physical constraint benchmark data, which is not directly used as physical parameters for risk monitoring, but as the basis for calculating the physical nature of AUVs and as input to the model, including navigation depth, rate of change of depth, navigation speed, six-degree-of-freedom attitude angles and their rate of change; the third category is state observation data, which comes from multiple physical parameters corresponding to each subsystem. The energy and power subsystem includes voltage, current, and temperature parameters; the propulsion subsystem includes motor speed, torque, power, and thermal state parameters; the emergency safety subsystem includes the level parameters of the hull leakage sensor and the internal pressure parameters; the environmental perception subsystem includes the distance characteristic parameters of the forward-looking / obstacle avoidance sonar; the communication subsystem includes the signal-to-noise ratio parameters of underwater acoustic communication and the data transmission delay parameters; and the navigation and positioning subsystem includes the three-axis velocity components of the Doppler velocities and the angular velocity and linear acceleration components of the inertial measurement unit. These six subsystems contain a total of 15 types of physical parameters.

[0019] S1.2: Perform global cleaning on the collected state observation data to remove abnormal noise points; Let the historical dataset be Calculate the global mean: ; Calculate the global standard deviation: ; Perform outlier detection and cleanup: ; In the formula, For the first in the historical dataset One original sampling point; The total number of samples in the historical data; This is the global mean of historical data; The global standard deviation of historical data; If the above formula is true, then determine Abnormal noise was removed from the historical dataset.

[0020] S1.3: Perform multi-channel data fusion on the cleaned historical data; Mapping data from different frequencies to a unified reference time axis Above, complete time alignment within any region / interval. Built-in cubic spline function: ; In the formula, These are the interpolated and aligned continuous-time data values. , , , These are the polynomial coefficients; Even after time alignment, the data may still contain random measurement noise. A fixed interval smoothing method is adopted, which uses the Kalman filter concept and combines forward filtering and reverse smoothing. The global optimal estimate of the historical state is made using observation information from all time periods, so as to eliminate historical noise to the greatest extent and reduce the sensor zero bias error to the minimum.

[0021] Perform forward filtering state update: ; Perform backward filtering for smooth updates: ; In the formula, This is the state estimate at time k obtained from the forward filtering process; For the observed values, ; The Kalman gain matrix; This is the observation matrix, used to map the high-dimensional system state space to the sensor observation space and construct the predicted observations. This is the final output smooth state vector; This is the smoothing gain matrix.

[0022] The smoothed state vectors are normalized using feature normalization to construct a standard dataset: ; In the formula, The first part of the standardized dataset formed after offline multi-channel data fusion. Data for each physical parameter, Smooth state vector The Middle The components of a physical parameter; ; and The first The historical maximum and minimum values ​​of each physical parameter. The final standard dataset is as follows: , .

[0023] S2. Perform physical constraint-based data augmentation on the scarce fault samples processed in S1 to generate an augmented dataset; S2 specifically includes: S2.1. Fault injection (hard enhancement) based on the AUV six-degree-of-freedom dynamic model, using physical equations to generate sudden fault samples, i.e., hard fault samples, including: Establish the six-degree-of-freedom motion equations of the AUV in volume coordinates: ; In the formula, The velocity vector in the carrier coordinate system. , representing the three-axis velocities and angular velocities, respectively; This is the pose vector in the geodetic coordinate system. , used to calculate the buoyancy force affected by attitude; The inertial matrix includes the rigid body mass matrix and the hydrodynamic added mass matrix; Let the Coriolis and centripetal force matrices describe the nonlinear forces generated by rotational motion; This is the hydrodynamic drag matrix, which includes linear damping terms and second-order nonlinear damping terms; The restoring force vector describes the static forces and moments generated by gravity and buoyancy; This is the vector of total thrust and torque acting on the hull; Establish a fault injection model: ; In the formula, This represents the actual thrust vector acting on the hull after the failure occurs. The desired thrust vector emitted by the control system; A unit matrix represents the actuator operating in a completely fault-free state; The fault performance loss matrix is ​​a diagonal matrix with diagonal elements. Indicates the first The failure rate of each channel; This is the environmental disturbance vector, representing the external disturbance force generated by simulated ocean currents or waves; S2.2. Standard dataset constructed using S1 The system utilizes real-world, historically scarce fault samples and employs a physical information generative adversarial network to generate gradually varying, nonlinear sensor faults and complexly coupled fault samples, i.e., soft fault samples, including: Construct a generative adversarial network architecture and build a system containing generators. (Generator) and Discriminator (Discriminator) time series generation model; The input uses real historical rare fault samples as seed data. Generator input random noise and seed data Output the generated virtual data sequence : ; Discriminator Automatically distinguishes between extracted real historical scarce fault samples and virtual fault data sequences. ; To ensure that the data generated by generative adversarial networks conforms to physical laws, the network's loss function... Add physical constraint terms: ; In the formula, To generate the original adversarial loss for the adversarial network; These are the weighting coefficients for physical constraints; For physical constraint terms, the expression is: ; In the formula, , These are the virtual velocities and accelerations generated by the generative adversarial network, respectively. , , These represent the virtual power, voltage, and current generated by the generative adversarial network, respectively. S2.3. Perform physical feasibility filtering on the hard fault samples and soft fault samples generated in S2.1 and S2.2 respectively. If the physical feasibility judgment condition is met, it is considered a valid sample. Valid hard fault samples are denoted as... Valid samples of soft faults are denoted as Store in augmented dataset If the conditions are not met, the data is discarded and regenerated. In the formula, the physical feasibility judgment conditions include: State continuity test: Check the first difference of the generated data to ensure there are no mathematical breakpoints. The expression is: ; In the formula, for The Middle One element, The smoothness threshold is calculated using the following formula: Based on the output of step S1, a normalized dataset is generated. (Assume the total duration is) Construct a set of first-order difference sequences. ; ; In the formula, for The arithmetic mean of the first-difference sequence of normal standardized data. ; First-order difference sequence standard deviation ; This is the preset tolerance coefficient; A physical limit threshold test is performed to ensure that the generated data are all within a reasonable range. The expression is: ; In the formula, For the generated virtual state sample values, , The set physical limit threshold; Fault feature identifiability verification ensures that the intensity of the generated fault features can penetrate normal measurement noise without being drowned out. The expression is: ; In the formula, The preset system noise threshold, To obtain from standard datasets The normal baseline data sequence extracted and aligned with the length of the generated virtual fault data sequence.

[0024] S3. Use the augmented dataset to perform offline system identification and parameter calibration, calculate and store the key model parameters required for online analysis; the key model parameters include the risk physical boundary thresholds of each subsystem, the dynamic coupling matrix library for different task stages, and the risk inheritance matrix for cross-stage transition; S3 specifically includes: S3.1. To achieve risk coupling analysis at different task stages, the dataset will be enhanced. The unordered time-series data was divided into four mission phases: dive, cruise, near-bottom operation, and surfacing recovery. The K-Means clustering algorithm was used to partition the data based on its features, thus dividing the entire dataset... Divided into Subset The clustering objective function is: ; In the formula, This is a measure of the total bias in the clustering partitions; For the first The data cluster set for each task phase; To enhance the dataset A single state sample vector in; For the first The centroid vector of each cluster center; S3.2. Calibrate the risk physical boundary thresholds for the energy and power subsystem, propulsion subsystem, emergency safety subsystem, environmental perception subsystem, communication subsystem, and navigation and positioning subsystem of the underwater vehicle. Using the fault samples and extreme condition samples contained in the enhanced dataset, determine the safety envelope of each physical parameter corresponding to each subsystem using the following formula: The upper threshold is calculated using the following formula: ; The lower bound threshold is calculated using the following formula: ; In the formula, , The first The upper and lower limits of the risk physical boundary for each physical parameter; This is a subset of normal states from the standard dataset after removing historical faults and abnormal records. All samples included in this subset are from the first set of samples. A set of numerical values ​​for physical parameters; and These are the arithmetic mean operator and the standard deviation operator, respectively; This refers to the safety margin factor. S3.3. The degree of mutual influence between subsystems varies in different task phases, so an independent dynamic coupling matrix is ​​constructed for each task phase. , The coupling strength was quantified using the Pearson correlation coefficient matrix. elements in Indicates the first Under the first task phase, the first The physical parameter and the first The dynamic coupling strength between physical parameters: ; In the formula, For the first The data cluster set for each task phase; , for The first moment and the Data values ​​for each physical parameter; S3.4. To address the risk transmission issue during task switching, time-delay cross-correlation analysis is used to analyze the data characteristics during phase transitions, ultimately forming a cross-phase risk inheritance matrix. Used for initial state correction during online analysis; The risk inheritance factor is calculated using the following formula: ; In the formula, As a risk inheritance factor, it represents the risk inheritance rate of different subsystems of the AUV from the previous mission phase. The physical parameters affect the current mission phase. The influence weight of each physical parameter; and These are the end time of the previous task phase (Δt before the switch point) and the start time of the current task phase (Δt after the switch point), respectively. The physical parameter and the first Normalized state data values ​​of each physical parameter, directly from the augmented dataset. Read from; The mathematical expectation operator (averaging the sample segments in the augmented dataset where phase switching occurs); This corresponds to the normalized standard deviation; This indicates the end time of the previous task phase, and the different subsystems at the [missing information]. Historical statistical average values ​​of several physical parameters; Indicates the start time of the current task phase, and the different subsystems' first... Historical statistical averages of several physical parameters. This indicates the general category of the preset task stages. In the first The last moment before the task transition occurred The specific values ​​of each physical parameter In the In the very first moment after the task transition, the first The specific numerical value of each feature.

[0025] S4. Real-time acquisition of sensor data from underwater vehicles, and online real-time data fusion and preprocessing that meet causal constraints; S4 specifically includes: S4.1. When the AUV performs underwater missions, the onboard computer uses a moving average filtering method to clean the sensor data of the physical parameters of each subsystem acquired in real time, preventing erroneous alarms caused by transient noise. (1) Construct a sliding window and set its length to be [value missing]. The window should only retain the most recent one. Data at each moment: ; (2) Calculate the local window mean: ; In the formula, For the present The moving average over time; The length of the sliding window; For the current and past Real-time measurement values ​​at each moment; (3) Real-time abnormal cleaning: ; In the formula, This is the latest raw data collected at the current moment; This is the moving average value calculated at the previous time step; The preset deviation threshold is set according to the characteristics of AUV; The real-time anomaly cleanup logic is as follows: if the inequality is true, then it means... To eliminate sudden noise, a cleaning operation is performed using... replace Input into the risk coupling analysis unit; if the inequality does not hold, retain it. The final dataset after data cleaning is denoted as ; S4.2. Perform online multi-channel data fusion on the cleaned real-time data, using an alignment and extended Kalman filter fusion strategy based on linear extrapolation.

[0026] (1) Since the sampling frequencies of the sensors in each subsystem are inconsistent, the data of each channel is mapped to the same time by using the current system master clock time t as the reference and performing a first-order linear extrapolation using the current time acquisition value and the previous time acquisition value: ; In the formula, For the first Aligned estimates of each sensor channel at the current system time t; For the first The latest measurement value received by the sensor after S4.1 cleaning. ; The first The measurement value of the previous frame from each sensor; and These are the physical timestamps corresponding to these two data frames.

[0027] (2) Extended Kalman filtering is used for fusion, and the state is further predicted based on the six-degree-of-freedom motion equations of the underwater vehicle constructed in step S2.1: ; And then correct it using the aligned observation data: ; In the formula, This is the estimated value of the fused holographic state vector at the current moment; This is the nonlinear state transition function derived in S2.1; This is the current control command; The Kalman gain matrix is ​​calculated online. The observation vector consists of all channels; For observation functions; S4.3. Perform real-time normalization on the data processed in S4.2, call the historical extreme values ​​stored offline in step S1.3, map the real-time state vector to the standard space, and generate a standardized holographic state vector. The expression is: ; In the formula, For the current moment The standardized values ​​of each feature channel constitute the standardized holographic state vector. ; The first in the fused state vector One portion, and This refers to the data obtained in step S1.

[0028] S5. Based on the current navigation status and control commands, identify the current mission phase of the vehicle in real time and generate a mission phase identifier; During the execution of a complete mission by an autonomous underwater vehicle, the dominant risk types, risk evolution rates, and risk coupling relationships differ significantly across different mission phases. Therefore, before constructing a risk situation unit, it is necessary to first identify the current mission phase, thereby providing phase constraints for subsequent risk threshold invocation (step S6) and coupling matrix activation (step S8).

[0029] S5 specifically includes: S5.1. Extract key variables from the real-time fused state data and control system output in step S4, and construct a state discrimination vector for task phase identification. : ; In the formula, For the spacecraft at time The navigation depth is used to extract the position component from the fused state vector; For the vehicle's depth change rate, extract the velocity component from the fused state vector; For sailing speed; The rate of change of the vehicle's attitude angle; It is a task control instruction representation (dimensionless integer), directly read from the AUV main control computer, representing the current low-level control modes such as fixed depth mode and fixed altitude mode; S5.2. Based on the stage discrimination threshold parameters preset in S1.1, for Perform the following logical checks to generate the task stage representation for the current moment. : When the following formula is satisfied, it is determined to be in the diving phase. : > and ; In the formula, The preset threshold for the rate of change of depth during the descent phase. Relay depth; Logic: Vertical velocity downwards and exceeds the threshold. Meanwhile, the current depth has not reached the relay depth. ; When the following formula is met, it is determined to be in the cruise phase. : ≤ and ; In the formula, The preset depth change rate threshold for the cruise phase. Cruise speed threshold; Logic: Depth remains stable (absolute value of rate of change less than threshold). And the sailing speed is greater than the cruising speed threshold. ; When the following formula is met, it is determined to be in the near-bottom operation phase. : and ; In the formula, For the target operating depth, To allow for error, Low-speed threshold; Logic: Current depth and target job depth The deviation is less than the allowable error And the speed is below the low speed threshold ; When the following formula is met, it is determined to be in the buoyancy recovery phase. : ; Logic: Vertical velocity increases and the rate exceeds the floating threshold - ; S5.3. Based on the above judgment results, output the task stage identifier at the current moment: .

[0030] The logo This will be directly used as an index and input into step S6 to call the corresponding risk physical boundary threshold. And input it into S8 to activate the corresponding dynamic coupling matrix. .

[0031] S6. Based on the task stage identifier, call the corresponding risk physical boundary threshold to construct a risk situation unit vector containing state deviation and trend change rate; the purpose of this step is to transform the original state data into a dimensionless risk metric.

[0032] S6 specifically includes: S6.1. Calculate the state deviation: ; In the formula, For the first A physical parameter in Degree of deviation from state at any given moment; For the currently identified task stage Below, the cluster center mean of this physical parameter, i.e., in step S3.1 The corresponding components; For the risk physical boundary threshold determined in step S3.2, if > Then take the upper bound. Otherwise, take the lower bound. ; S6.2. Calculate the risk of trend change rate: ; In the formula, The risk value is the rate of change of trend; Δt is the sampling time interval. The smoothness threshold calculated in step S2.3; S6.3. Constructing the risk posture unit vector: ; ; in, For the first Independent risk values ​​for each physical parameter; The deviation weighting coefficient is set based on empirical values. For the current moment An initial risk situation unit vector composed of physical parameters.

[0033] S7. When a task phase switch is detected, perform cross-phase risk inheritance calculation using the risk inheritance matrix to determine the initial risk state of the new phase; When the system detects a change in the task phase, it corrects the current risk posture using the risk inheritance matrix, expressed as follows: ; In the formula, This is the risk situation vector after inheritance and modification; This represents the risk situation vector from the previous moment. This is a time decay factor used to simulate the natural dissipation of risk over time.

[0034] S8. Activate the dynamic coupling matrix corresponding to the current task stage identifier, perform dynamic coupling evolution deduction on the risk situation unit vector, and calculate the propagation and amplification effect of risk among multi-source subsystems; Unlike the independent early warning systems of each subsystem, this step quantifies the propagation effect of risk. The dynamic coupling evolution calculation method for the risk situation unit vector is as follows: ; In the formula, The evolved global coupling risk vector has the following dimensions: This corresponds to the total number of physical parameters; The corrected risk vector is output from step S7; It is the identity matrix; This is the coupling gain coefficient; For the construction in step S3.3, and the current task stage The corresponding dynamic coupling matrix.

[0035] S9. Comprehensively assess the overall risk level after evolution, and output graded early warning information accordingly.

[0036] S9 specifically includes: S9.1. Calculate the overall risk level of AUVs: ; In the formula, The overall risk level of the AUV at the current moment; Global coupling risk vector The first in Risk elements of each physical parameter; S9.2. Output a level three warning signal : ; in, and The preset risk classification warning line is determined based on the 90th and 95th percentiles of the risk distribution of normal data obtained from offline statistics of S3. S9.3. Based on the comprehensive risk level of AUV output in S9.1, predict the risk growth rate based on a sliding time window, eliminate instantaneous fluctuations, and extract the long-term trend change rate of risk evolution, specifically as follows: Set the prediction window length to Constructing historical risk sequences : ; in, The sampling time within the sliding time window; The sampling time calculated from S9.1 Overall risk level value; Calculate the risk growth rate: ; In the formula, Given the current rate of risk growth, if This indicates that the risk is worsening, if ≤0 indicates that the risk tends to converge or decrease; It is the arithmetic mean of the time series within the time window; It is the arithmetic average of the overall risk level within the time window; S9.4. Based on the calculated risk value and risk growth rate, predict the remaining time required for the AUV to reach its physical failure limit, providing a quantitative basis for emergency decision-making. Specifically: Define the physical failure limit threshold of AUV The threshold is higher than This represents a critical point at which the system will experience irreversible physical damage, including the rupture of the pressure-resistant housing and the burnout of the motor. Calculate the remaining safe time window: ; In the formula, The remaining safe time window as assessed; This is a preset system failure limit threshold; The current rate of risk growth; This is a safety conservatism factor, representing a human-increased deviation, to ensure the safety margin of the project. The judgment logic is as follows: the first line in the above formula represents linear extrapolation prediction when the risk is increasing but has not yet failed; the second line represents the remaining safe time as 0 when the risk has exceeded the failure limit.

[0037] By conducting risk coupling analysis on autonomous underwater vehicles (AUVs), an objective understanding of the risk level of the vehicle during mission execution can be formed, allowing for a determination of whether the conditions for continuing the predetermined mission are met. When the risk analysis results show that the risk level of a local subsystem or the overall system is continuously increasing, the AUV can trigger or execute degraded operation strategies or safety control logic in advance, such as reducing speed, adjusting course, restricting high-maneuverability operations, or entering standby or return-to-base modes, thereby preventing further risk accumulation that could lead to mission failure or equipment loss. Simultaneously, based on the risk coupling analysis results, the platform or command system can conduct more in-depth risk assessment and decision support. For identified potential failure modes or high-risk scenarios, targeted emergency response plans and operational adjustment strategies can be developed to improve the ability to respond to complex marine environments and system anomalies. Furthermore, the risk coupling analysis results can provide a basis for vehicle structural design, system integration, and functional optimization, supporting continuous improvement of the reliability, redundancy configuration, and control strategies of key components.

[0038] An autonomous underwater vehicle (AUV) risk coupling analysis system, used to implement the aforementioned AUV risk coupling analysis method, includes: The data acquisition unit is used to collect historical multi-source heterogeneous data of autonomous underwater vehicles, as well as real-time sensor data of underwater vehicles; The data preprocessing unit is used to perform offline multi-channel data fusion and standardization preprocessing on the historical multi-source heterogeneous data, and to perform online real-time data fusion and preprocessing on the real-time sensor data to meet causal constraints. The data augmentation unit is used to perform physical constraint-based data augmentation on preprocessed scarce fault samples to generate an augmented dataset. The parameter calibration unit is used to perform offline system identification and parameter calibration using the augmented dataset, calculate and store the key model parameters required for online analysis; the key model parameters include the risk physical boundary thresholds of each subsystem, a dynamic coupling matrix library for different task stages, and a risk inheritance matrix for cross-stage transition; The mission phase identification unit is used to identify the current mission phase of the aircraft in real time based on the current navigation status and control commands, and generate a mission phase identifier. The risk situation construction unit is used to construct a risk situation unit containing state deviation and trend change rate by calling the corresponding risk physical boundary threshold according to the task stage identifier. The risk inheritance calculation unit is used to perform cross-stage risk inheritance calculation using the risk inheritance matrix when a task stage switch is detected, and to determine the initial risk state of the new stage. The coupled evolution and deduction unit is used to activate the dynamic coupling matrix corresponding to the current task stage identifier, perform dynamic coupled evolution and deduction on the risk situation unit, and calculate the propagation and amplification effect of risk among multi-source subsystems. The risk assessment output unit is used to comprehensively assess the overall risk level after evolution and output graded early warning information accordingly.

[0039] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A risk coupling analysis method for autonomous underwater vehicles, characterized in that, Includes the following steps: S1. Collect historical multi-source heterogeneous data of autonomous underwater vehicles, perform offline multi-channel data fusion and standardized preprocessing, and construct a standard dataset; S2. Perform physical constraint-based data augmentation on the scarce fault samples processed in S1 to generate an augmented dataset; S3. Use the augmented dataset to perform offline system identification and parameter calibration, calculate and store the key model parameters required for online analysis; the key model parameters include the risk physical boundary thresholds of each subsystem, the dynamic coupling matrix library for different task stages, and the risk inheritance matrix for cross-stage transition; S4. Real-time acquisition of sensor data from underwater vehicles, and online real-time data fusion and preprocessing that meet causal constraints; S5. Based on the current navigation status and control commands, identify the current mission phase of the vehicle in real time and generate a mission phase identifier; S6. Based on the task stage identifier, call the corresponding risk physical boundary threshold to construct a risk situation unit vector containing state deviation and trend change rate; S7. When a task phase switch is detected, perform cross-phase risk inheritance calculation using the risk inheritance matrix to determine the initial risk state of the new phase; S8. Activate the dynamic coupling matrix corresponding to the current task stage identifier, perform dynamic coupling evolution deduction on the risk situation unit vector, and calculate the propagation and amplification effect of risk among multi-source subsystems; S9. Comprehensively assess the overall risk level after evolution, and output graded early warning information accordingly.

2. The risk coupling analysis method for autonomous underwater vehicles according to claim 1, characterized in that, The historical multi-source heterogeneous data of the autonomous underwater vehicle in S1 includes state observation data, which comes from multiple physical parameters corresponding to each subsystem. Among them, the energy and power subsystem includes voltage, current and temperature parameters; the propulsion subsystem includes motor speed, torque, power and thermal state parameters; the emergency safety subsystem includes cabin leakage sensor level parameters and cabin pressure parameters; the environmental perception system includes distance characteristic parameters of forward-looking / obstacle avoidance sonar; the communication subsystem includes underwater acoustic communication signal-to-noise ratio parameters and data transmission delay parameters; and the navigation and positioning subsystem includes three-axis velocity components of Doppler velocities and angular velocity and linear acceleration components of the inertial measurement unit. The above 6 subsystems have a total of 15 types of physical parameters.

3. The risk coupling analysis method for autonomous underwater vehicles according to claim 1, characterized in that, S2 specifically includes: S2.

1. Fault injection based on the AUV six-degree-of-freedom dynamic model, using physical equations to generate sudden fault samples, i.e., hard fault samples, including: Establish the six-degree-of-freedom motion equations of the AUV in volume coordinates: ； in, This is the velocity vector in the carrier coordinate system; This is the pose vector in the geodetic coordinate system; The inertia matrix; For the Coriolis and centripetal force matrix; This is the hydrodynamic resistance matrix; The restoring force vector; The vector of total thrust and moment acting on the hull Establish a fault injection model: ； in, This represents the actual thrust vector acting on the hull after the failure occurs. The desired thrust vector emitted by the control system; It is the identity matrix; This is the failure performance loss matrix; This represents the environmental disturbance vector. S2.

2. Standard dataset constructed using S1 The system utilizes real-world, historically scarce fault samples and employs a physical information generative adversarial network to generate gradually varying, nonlinear sensor faults and complexly coupled fault samples, i.e., soft fault samples, including: Construct a generative adversarial network architecture and build a system containing generators. and discriminator Time series generation model; The input uses real historical rare fault samples as seed data. Generator input random noise and seed data Output the generated virtual data sequence : ； Discriminator Automatically distinguishes between extracted real historical scarce fault samples and virtual fault data sequences. ; In the network loss function Add physical constraint terms: ； in, To generate the original adversarial loss for the adversarial network; These are the weighting coefficients for physical constraints; For physical constraint terms, the expression is: ； in, , These are the virtual velocities and accelerations generated by the generative adversarial network, respectively. , , These represent the virtual power, voltage, and current generated by the generative adversarial network, respectively. S2.

3. Perform physical feasibility filtering on the hard fault samples and soft fault samples generated in S2.1 and S2.2 respectively. If the physical feasibility judgment condition is met, it is considered a valid sample. Valid hard fault samples are denoted as... Valid samples of soft faults are denoted as Store in augmented dataset If the conditions are not met, the data is discarded and regenerated. The physical feasibility assessment criteria include: The state continuity test is expressed as follows: ； In the formula, for The Middle One element, The smoothness threshold is calculated using the following formula: ； In the formula, for The arithmetic mean of the first-order difference sequence of normal standardized data; Let be the standard deviation of this first-order difference sequence; This is the preset tolerance coefficient; The physical limit threshold test is expressed as follows: ； In the formula, For the generated virtual state sample values, , The set physical limit threshold; The fault feature identifiability test is expressed as follows: ； In the formula, The preset system noise threshold, To obtain from standard datasets The normal baseline data sequence extracted and aligned with the length of the generated virtual fault data sequence.

4. The risk coupling analysis method for autonomous underwater vehicles according to claim 1, characterized in that, S3 specifically includes: S3.

1. Enhance the dataset The unordered time-series data was divided into four mission phases: dive, cruise, near-bottom operation, and surfacing recovery. The K-Means clustering algorithm was used to partition the data based on its features, thus dividing the entire dataset... Divided into The subsets are clustered with the following objective function: ； in, This is a measure of the total bias in the clustering partitions; For the first The data cluster set for each task phase; To enhance the dataset A single state sample vector in; For the first The centroid vector of each cluster center; S3.

2. Calibrate the risk physical boundary thresholds of the energy and power subsystem, propulsion subsystem, emergency safety subsystem, environmental perception subsystem, communication subsystem, and navigation and positioning subsystem of the underwater vehicle. Using the fault samples and extreme condition samples contained in the enhanced dataset, determine the safety envelope of each physical parameter corresponding to each subsystem using the following formula: The upper threshold is calculated using the following formula: ； The lower bound threshold is calculated using the following formula: ； in, , The first The upper and lower limits of the risk physical boundary for each physical parameter; This is a subset of normal states from the standard dataset after removing historical faults and abnormal records. All samples included in this subset are from the first set of samples. A set of numerical values ​​for physical parameters; and These are the arithmetic mean operator and the standard deviation operator, respectively; This refers to the safety margin factor. S3.

3. The degree of mutual influence between subsystems varies in different task phases, so an independent dynamic coupling matrix is ​​constructed for each task phase. , The coupling strength was quantified using the Pearson correlation coefficient matrix. elements in Indicates the first Under the first task phase, the first The physical parameter and the first The dynamic coupling strength between the physical parameters is expressed as: ； in, For the first The set of all moments corresponding to each task phase; , for The first moment and the Data values ​​for each physical parameter; S3.

4. To address the risk transmission issue during task switching, time-delay cross-correlation analysis is used to analyze the data characteristics during phase transitions, ultimately forming a cross-phase risk inheritance matrix. Used for initial state correction during online analysis; The risk inheritance factor is calculated using the following formula: ； in, As a risk inheritance factor, it represents the risk inheritance rate of different subsystems of the AUV from the previous mission phase. The physical parameters affect the current mission phase. The influence weight of each physical parameter; and These are the end time of the previous task phase and the start time of the current task phase, respectively. The normalized state data values ​​of the j-th physical parameter and the j-th physical parameter are directly obtained from the augmented dataset D. aug Read from; For mathematical expectation operators; This corresponds to the normalized standard deviation; This indicates the end time of the previous task phase, and the different subsystems at the [missing information]. Historical statistical average values ​​of several physical parameters; Indicates the start time of the current task phase, and the different subsystems' first... The historical statistical average of each physical parameter.

5. The risk coupling analysis method for autonomous underwater vehicles according to claim 1, characterized in that, S4 specifically includes: cleaning the sensor data of physical parameters of each subsystem acquired in real time using a moving average filtering method; calculating the local window mean by constructing a sliding window and comparing it with the current acquired value to identify and remove instantaneous noise; performing online multi-channel data fusion on the cleaned real-time data; using the current system master clock as a reference, mapping the data of each sensor channel to the same moment using first-order linear extrapolation; and using extended Kalman filtering for state prediction and correction; performing real-time normalization processing on the fused data; and calling the historical extreme values ​​stored offline to map the real-time state vector to the standard space to generate a standardized holographic state vector. ; S5 specifically includes: S5.

1. Extract key variables from the real-time fused state data and control system output in step S4, and construct a state discrimination vector for task phase identification. : ； In the formula, For the spacecraft at time The depth of navigation; The rate of change of the vehicle's depth; For sailing speed; The rate of change of the vehicle's attitude angle; This is a representation of task control instructions, indicating the current underlying control mode; S5.

2. Based on the preset stage discrimination threshold parameter, for Perform the following logical checks to generate the task stage representation for the current moment. : When the following formula is satisfied, it is determined to be in the diving phase. : > and ; In the formula, The preset threshold for the rate of change of depth during the descent phase. For relay depth; When the following formula is satisfied, it is determined to be in the cruise phase. : ≤ and ; In the formula, The preset depth change rate threshold for the cruise phase. The cruise speed threshold; When the following formula is met, it is determined to be in the near-bottom operation phase. : and ; In the formula, For the target operating depth, To allow for error, Low speed threshold; When the following formula is met, it is determined to be in the buoyancy recovery phase. : ； S5.

3. Based on the above judgment results, output the task stage identifier at the current moment: 。 6. The risk coupling analysis method for autonomous underwater vehicles according to claim 1, characterized in that, S6 specifically includes: S6.

1. Calculate the state deviation: ； In the formula, For the first A physical parameter in Degree of deviation from state at any given moment; For the currently identified task stage The mean of the cluster centers of this physical parameter; For the risk physical boundary threshold determined in step S3.2, if Then take the upper bound. Otherwise, take the lower bound. ; S6.

2. Calculate the risk of trend change rate: ； In the formula, Risk value for the rate of change of trend; The sampling time interval; The smoothness threshold calculated in step S2.3; S6.

3. Constructing the risk situation unit vector: ； ； in, For the first Independent risk values ​​for each physical parameter; This is the deviation weighting coefficient; For the current moment An initial risk situation unit vector composed of physical parameters.

7. The risk coupling analysis method for autonomous underwater vehicles according to claim 1, characterized in that, S7 specifically includes: when the system detects a change in the task phase, it uses the risk inheritance matrix to correct the current risk situation, the expression of which is: ； In the formula, This is the risk situation vector after inheritance and modification; This represents the risk situation vector from the previous moment. This is the time decay factor.

8. The risk coupling analysis method for autonomous underwater vehicles according to claim 1, characterized in that, S8 specifically includes: a dynamic coupling evolution calculation method for the risk situation unit vector, which is as follows: ； In the formula, Let M be the evolved global coupling risk vector, with dimension M. The corrected risk vector is output from step S7; It is the identity matrix; This is the coupling gain coefficient; For the construction in step S3.3, and the current task stage The corresponding dynamic coupling matrix.

9. The risk coupling analysis method for autonomous underwater vehicles according to claim 1, characterized in that, S9 specifically includes: S9.

1. Calculate the overall risk level of AUVs: ； In the formula, The overall risk level of the AUV at the current moment; Global coupling risk vector The first in Risk elements of each physical parameter; S9.

2. Output a level three warning signal : ； in, and The preset risk level warning line; S9.

3. Based on the comprehensive risk level of AUVs output in S9.1, predict the risk growth rate based on a sliding time window, specifically as follows: Set the prediction window length to Constructing historical risk sequences : ； in, The sampling time within the sliding time window; The sampling time calculated by S9.1 Overall risk level value; Calculate the risk growth rate: ； In the formula, Given the current rate of risk growth, if A value greater than 0 indicates that the risk is worsening. ≤0 indicates that the risk tends to converge or decrease; It is the arithmetic mean of the time series within the time window; It is the arithmetic average of the overall risk level within the time window; S9.

4. Predict the remaining time required for the AUV to reach its physical failure limit, specifically: Define the physical failure limit threshold of AUV The threshold is higher than ; Calculate the remaining safe time window: ； In the formula, The remaining safe time window as assessed; This is a preset system failure limit threshold; The current rate of risk growth; It is a safe and conservative factor.

10. A risk coupling analysis system for autonomous underwater vehicles, used to implement the risk coupling analysis method for autonomous underwater vehicles as described in any one of claims 1-9, characterized in that, include: The data acquisition unit is used to collect historical multi-source heterogeneous data of autonomous underwater vehicles, as well as real-time sensor data of underwater vehicles; The data preprocessing unit is used to perform offline multi-channel data fusion and standardization preprocessing on the historical multi-source heterogeneous data, and to perform online real-time data fusion and preprocessing on the real-time sensor data to meet causal constraints. The data augmentation unit is used to perform physical constraint-based data augmentation on preprocessed scarce fault samples to generate an augmented dataset. The parameter calibration unit is used to perform offline system identification and parameter calibration using the augmented dataset, calculate and store the key model parameters required for online analysis; the key model parameters include the risk physical boundary thresholds of each subsystem, a dynamic coupling matrix library for different task stages, and a risk inheritance matrix for cross-stage transition; The mission phase identification unit is used to identify the current mission phase of the aircraft in real time based on the current navigation status and control commands, and generate a mission phase identifier. The risk situation construction unit is used to construct a risk situation unit containing state deviation and trend change rate by calling the corresponding risk physical boundary threshold according to the task stage identifier. The risk inheritance calculation unit is used to perform cross-stage risk inheritance calculation using the risk inheritance matrix when a task stage switch is detected, and to determine the initial risk state of the new stage. The coupled evolution and deduction unit is used to activate the dynamic coupling matrix corresponding to the current task stage identifier, perform dynamic coupled evolution and deduction on the risk situation unit, and calculate the propagation and amplification effect of risk among multi-source subsystems. The risk assessment output unit is used to comprehensively assess the overall risk level after evolution and output graded early warning information accordingly.

Images