Seabed equipment state self-monitoring and data quality dynamic evaluation method and system

Abstract

This invention relates to the field of self-monitoring technology for marine observation equipment, and discloses a method and system for self-monitoring the status of seabed-based equipment and dynamic evaluation of data quality. The method includes: S1, collecting key status parameters of the seabed-based equipment, including system power supply parameters, equipment attitude angle data, cabin temperature and humidity, cabin leakage, biological adhesion status on the equipment surface, and key operating parameters of scientific sensors; S2, using a hybrid evaluation mechanism combining a physical mechanism-based state-error correlation model and a data-driven MSTCN-Attention machine learning model to quantify the reliability of the key status parameters and generate a data quality index; S3, executing a preset control strategy based on the data quality index, etc. This method and system can achieve comprehensive perception of the status of seabed-based equipment, accurate evaluation of data quality, and deep integration of the two.

Description

Technical Field

[0001] This invention relates to the field of self-monitoring technology for marine observation equipment, specifically to a method and system for self-monitoring the status of seabed-based equipment and dynamically evaluating data quality. Background Technology

[0002] As a comprehensive observation platform deployed on the seabed, the quality of the data collected by the seabed-based equipment and the stability of its own condition are directly related to the accuracy of marine scientific research, resource exploration, and engineering safety decisions. However, the complex and harsh environment of the seabed (such as high pressure, low temperature, corrosion, and biological adhesion) and the long-term unattended operation mode of the equipment make achieving comprehensive perception of equipment status and reliable assurance of data quality a key technical bottleneck that urgently needs to be overcome in this field.

[0003] In the condition monitoring of seabed-based equipment, existing technologies are mostly limited to basic fault alarms such as power supply voltage and hull sealing. Effective real-time monitoring methods are lacking for critical conditions that directly affect data accuracy, such as gradual drift in sensor performance and subtle changes in equipment attitude. Regarding data quality, the industry generally relies on laboratory calibration and post-processing after observation, failing to conduct real-time quality assessment and control at the moment data is generated. Although some research has attempted to improve reliability through data fusion or redundant design, these methods have failed to fundamentally establish an intrinsic correlation model between the physical state of the equipment and data errors. This results in the condition monitoring system and the data quality control system being isolated from each other, making it difficult to form a collaborative optimization closed loop.

[0004] Taking the Acoustic Doppler Current Profiler (ADCP) as an example, in a panoramic sensing and monitoring project for offshore wind farms, the seabed is prone to tilting due to the impact of submarine currents, causing systematic deviations in the velocity data collected by the ADCP. However, due to the lack of effective attitude sensing and data correlation monitoring capabilities, this anomaly was not discovered until several months later when the equipment was retrieved, severely affecting the accuracy of the current marine hydrological analysis. This type of problem highlights the shortcomings of existing technologies in state perception and data quality correlation monitoring. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method and system for self-monitoring of the status of seabed-based equipment and dynamic evaluation of data quality. This method and system can realize comprehensive perception of the status of seabed-based equipment, accurate evaluation of data quality, and deep integration of the two.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Methods for self-monitoring of the condition and dynamic evaluation of data quality of seabed-based equipment include: S1. Collect key status parameters of the seabed-based equipment, including system power supply parameters, equipment attitude angle data, cabin temperature and humidity, cabin leakage, biofouling on the equipment surface, and key operating parameters of scientific sensors; S2. Employ a hybrid evaluation mechanism combining a physical mechanism-based state-error correlation model and a data-driven MSTCN-Attention machine learning model to quantify the reliability of the key status parameters and generate a data quality index; S3. Execute a preset control strategy based on the data quality index to achieve working mode switching and automatic generation of data quality labels, and output a standardized data product package; S2 includes: S21, standardizing key state parameters and generating a power health index, attitude stability index, and sensor performance degradation index based on the standardized key state parameters; S22, inputting the standardized key state parameters into a state-error correlation model and calculating the expected error introduced by the physical state based on the physical mechanism through the state-error correlation model; S23, inputting the standardized key state parameters into an MSTCN-Attention machine learning model and intelligently identifying abnormal patterns from historical and real-time data through the MSTCN-Attention machine learning model to generate anomaly probabilities; S24, generating a data quality index through weighted fusion based on the expected error and anomaly probabilities, and dynamically adjusting the weight coefficients of the state-error correlation model and the MSTCN-Attention machine learning model according to the application scenario and data characteristics during the fusion process.

[0007] In this invention, preferably, S3 includes: S31, executing multi-threshold judgment logic based on data quality index and key status parameters; S32, encapsulating all data, status reports and quality tags into a standardized data product package for output; The multi-threshold judgment logic includes: maintaining the current mode when the data quality index is greater than the excellent threshold; switching to a lower power consumption or safer working mode when the data quality index is less than or equal to the degradation threshold or a specific abnormal state is detected, while attaching a structured quality label to the output data; and triggering the highest priority emergency response when the data quality index is less than or equal to the danger threshold or a fatal fault signal is received, switching to the safe mode and sending a high-level warning simultaneously.

[0008] In this invention, preferably, the process of collecting the bio-attachment status on the equipment surface is as follows: an underwater camera is used to periodically image the equipment shell and the surface of the acoustic sensor, or an acoustic surface detector is used to periodically scan the equipment shell and the surface of the acoustic sensor, and the coverage and thickness of the attachments are evaluated by image recognition or echo intensity analysis.

[0009] In this invention, preferably, the system power supply related parameters include the actual monitored voltage value, average operating current, and remaining battery power percentage, and the formula for calculating the power health index is: Wherein, PHI is the power health index, α, β, γ are experimentally calibrated weighting coefficients that satisfy α+β+γ=1, V_actual is the actual monitored voltage value, V_nominal is the system rated operating voltage, I_avg is the average operating current, I_max is the maximum allowable operating current, and SOC is the remaining battery charge percentage. The equipment attitude angle data includes the roll angle and pitch angle within a set time window, and the formula for calculating the attitude stability index is: Wherein, ASI stands for Attitude Stability Index. The variance of the roll angle within a set time window. The variance of the pitch angle within the set time window; The sensor performance degradation index is calculated independently for each sensor. The performance degradation evaluation model for the acoustic Doppler current profiler is as follows: Wherein, SDI_ADCP is the sensor performance degradation index of the acoustic Doppler current profiler, I_actual is the current operating current, I_initial is the initial operating current; SNR_actual is the current signal-to-noise ratio, SNR_initial is the initial signal-to-noise ratio; BE_actual is the current echo intensity, BE_initial is the initial echo intensity; The performance degradation evaluation model for pressure-type tide gauges is as follows: Where I_actual is the current operating current, I_initial is the initial operating current, R_actual is the current pressure reading repeatability, R_initial is the initial pressure reading repeatability, S_actual is the current temperature drift stability coefficient, and S_initial is the initial temperature drift stability coefficient. The performance degradation evaluation model for the CTD water quality analyzer is as follows: Where R_actual is the current electrode response speed, R_initial is the initial electrode response speed, A_actual is the current optical signal attenuation rate, A_initial is the initial optical signal attenuation rate, T_actual is the current temperature drift compensation coefficient stability, and T_initial is the initial temperature drift compensation coefficient stability.

[0010] In this invention, preferably, in S22, the state-error correlation model includes: attitude-flow velocity error model, temperature-sensor drift model and bio-attachment-acoustic performance model; The mathematical expression for the attitude-flow velocity error model is: Where δV is the velocity measurement error vector, V_measured is the ADCP measured velocity value, θ is the pitch angle, and V_offset is the error offset calibrated based on the equipment installation position and local flow field characteristics; The mathematical expression for the temperature-sensor drift model is: Where δC is the sensor reading drift, T is the current temperature, T_ref is the sensor calibration reference temperature, and k1 and k2 are the temperature drift coefficients obtained through laboratory calibration; The mathematical expression for the bioattachment-acoustic performance model is: Where A_loss is the total attenuation of the acoustic signal, d is the thickness of the biological attachment layer, η is the attenuation coefficient per unit thickness, and A_base is the basic propagation loss of the acoustic system.

[0011] In this invention, preferably, in S23, the working process of the MSTCN-Attention machine learning model includes: S231, The input layer receives multidimensional time-series data of length L; S232, the convolutional layer adopts a parallel convolutional structure, using three different kernel sizes (3, 5, and 7) to synchronously extract temporal features from the standardized multidimensional temporal data. The expression is: Where H_i is the temporal feature, X is the input temporal data, k_i is the size of the i-th convolutional kernel, W_i is the weight matrix of the i-th convolutional kernel, and b_i is the bias vector of the i-th convolutional kernel. S233, the attention mechanism module weights the temporal features output by the convolutional layer based on their importance, and the calculation expression is as follows: Where Q is the query matrix, K is the key matrix, V is the value matrix, and d_k is the dimension of the key vector; S234, the output layer generates anomaly probabilities using a fully connected network and a sigmoid activation function, expressed as follows: Where P_anomaly is the anomaly probability, H_attention is the attention-weighted feature representation, W_o is the weight matrix of the output layer, b_o is the bias vector of the output layer, and σ is the sigmoid function.

[0012] The seabed-based equipment condition self-monitoring and data quality dynamic evaluation system includes a condition self-monitoring module, a data quality dynamic evaluation module, and an intelligent decision-making and control module connected in sequence. The self-monitoring module is used to collect key status parameters of the seabed-based equipment. These key status parameters include system power supply parameters, equipment attitude angle data, cabin temperature and humidity, cabin leakage, biological adhesion status on the equipment surface, and key operating parameters of scientific sensors. The data quality dynamic evaluation module is used to quantify the credibility of key state parameters and generate a data quality index by employing a hybrid evaluation mechanism that combines a state-error correlation model based on physical mechanisms with a data-driven MSTCN-Attention machine learning model. The intelligent decision-making and control module is used to execute preset control strategies based on the data quality index, realize the switching of working modes and the automatic generation of data quality labels, and output standardized data product packages. The status self-monitoring module includes: The power monitoring unit consists of a voltage sensor, a current sensor, and a smart fuel meter chip, and is used to collect system power supply related parameters. The physical attitude monitoring unit adopts an integrated nine-axis inertial measurement unit, including a three-axis MEMS accelerometer, a three-axis MEMS gyroscope and a three-axis magnetometer, to collect equipment attitude angle data; The internal environment monitoring unit includes a digital temperature and humidity sensor, a resistive leak detection strip, and a biofouling monitoring device, which are used to collect data on the temperature and humidity inside the cabin, the cabin leakage situation, and the biofouling status on the equipment surface. Sensor health monitoring unit, used to collect key operating parameters of scientific sensors; The bio-attachment monitoring device uses an underwater camera or an acoustic surface detector to assess the coverage and thickness of the attached material through image recognition or echo intensity analysis. The scientific sensors include at least an acoustic Doppler current profiler, a pressure-type tide gauge, and a CTD water quality meter. The data quality dynamic assessment module includes: The state parameter generation layer is used to standardize key state parameters and generate power health index, attitude stability index and sensor performance degradation index based on the standardized key state parameters. State-error correlation model, used to calculate the expected error introduced by physical state based on physical mechanism; The MSTCN-Attention machine learning model is used to intelligently identify abnormal patterns from historical and real-time data and generate anomaly probabilities. The quality index fusion layer is used to generate a data quality index by weighted fusion based on expected error and anomaly probability. During the fusion process, the weight coefficients of the state-error association model and the MSTCN-Attention machine learning model are dynamically adjusted according to the application scenario and data characteristics. The intelligent decision-making and control module includes: The execution unit is used to execute multi-threshold judgment logic based on the data quality index and key status parameters; The output unit is used to encapsulate all data, status reports, and quality labels into a standardized data product package for output. The multi-threshold judgment logic includes: maintaining the current mode when the data quality index is greater than the excellent threshold; switching to a lower power consumption or safer working mode when the data quality index is less than or equal to the degradation threshold or a specific abnormal state is detected, while attaching a structured quality label to the output data; and triggering the highest priority emergency response when the data quality index is less than or equal to the danger threshold or a fatal fault signal is received, switching to the safe mode and sending a high-level warning simultaneously.

[0013] In this invention, preferably, the system power supply related parameters include the actual monitored voltage value, average operating current, and remaining battery power percentage, and the formula for calculating the power health index is: Wherein, PHI is the power health index, α, β, γ are experimentally calibrated weighting coefficients that satisfy α+β+γ=1, V_actual is the actual monitored voltage value, V_nominal is the system rated operating voltage, I_avg is the average operating current, I_max is the maximum allowable operating current, and SOC is the remaining battery charge percentage. The equipment attitude angle data includes the roll angle and pitch angle within a set time window, and the formula for calculating the attitude stability index is: Wherein, ASI stands for Attitude Stability Index. The variance of the roll angle within a set time window. The variance of the pitch angle within the set time window; The sensor performance degradation index is calculated independently for each sensor. The performance degradation evaluation model for the acoustic Doppler current profiler is as follows: Wherein, SDI_ADCP is the sensor performance degradation index of the acoustic Doppler current profiler, I_actual is the current operating current, I_initial is the initial operating current; SNR_actual is the current signal-to-noise ratio, SNR_initial is the initial signal-to-noise ratio; BE_actual is the current echo intensity, BE_initial is the initial echo intensity; The performance degradation evaluation model for pressure-type tide gauges is as follows: Where I_actual is the current operating current, I_initial is the initial operating current, R_actual is the current pressure reading repeatability, R_initial is the initial pressure reading repeatability, S_actual is the current temperature drift stability coefficient, and S_initial is the initial temperature drift stability coefficient. The performance degradation evaluation model for the CTD water quality analyzer is as follows: Where R_actual is the current electrode response speed, R_initial is the initial electrode response speed, A_actual is the current optical signal attenuation rate, A_initial is the initial optical signal attenuation rate, T_actual is the current temperature drift compensation coefficient stability, and T_initial is the initial temperature drift compensation coefficient stability.

[0014] In this invention, preferably, the state-error correlation model includes: attitude-flow velocity error model, temperature-sensor drift model and bio-attachment-acoustic performance model; The mathematical expression for the attitude-flow velocity error model is: Where δV is the velocity measurement error vector, V_measured is the ADCP measured velocity value, θ is the pitch angle, and V_offset is the error offset calibrated based on the equipment installation position and local flow field characteristics; The mathematical expression for the temperature-sensor drift model is: Where δC is the sensor reading drift, T is the current temperature, T_ref is the sensor calibration reference temperature, and k1 and k2 are the temperature drift coefficients obtained through laboratory calibration; The mathematical expression for the bioattachment-acoustic performance model is: Where A_loss is the total attenuation of the acoustic signal, d is the thickness of the biological attachment layer, η is the attenuation coefficient per unit thickness, and A_base is the basic propagation loss of the acoustic system.

[0015] In this invention, preferably, the working process of the MSTCN-Attention machine learning model includes: The input layer receives multi-dimensional time-series data of length L and performs standardization processing on the multi-dimensional time-series data. The convolutional layers employ a parallel convolutional structure, using convolutional kernels of 3, 5, and 7 different sizes to synchronously extract temporal features from the standardized multidimensional temporal data. The expression is as follows: Where H_i is the temporal feature, X is the input temporal data, k_i is the size of the i-th convolutional kernel, W_i is the weight matrix of the i-th convolutional kernel, and b_i is the bias vector of the i-th convolutional kernel. The attention mechanism module weights the temporal features output by the convolutional layer based on their importance; the calculation expression is as follows: Where Q is the query matrix, K is the key matrix, V is the value matrix, and d_k is the dimension of the key vector; The output layer generates anomaly probabilities using a fully connected network and a sigmoid activation function, expressed as: Where P_anomaly is the anomaly probability, H_attention is the attention-weighted feature representation, W_o is the weight matrix of the output layer, b_o is the bias vector of the output layer, and σ is the sigmoid function.

[0016] Compared with the prior art, the beneficial effects of the present invention are: (1) The method and system of the present invention establish a technical path for collaborative perception of multi-dimensional state parameters of seabed-based equipment. By collecting parameters related to power supply, attitude, internal environment and sensor health in stages, and generating comprehensive evaluation indices such as Power Health Index (PHI), Attitude Stability Index (ASI) and Sensor Performance Degradation Index (SDI), the health status of the equipment is fully and in real time, overcoming the technical defects of traditional monitoring methods that are single in dimension and cannot fully reflect the health status of the equipment.

[0017] (2) The method and system innovation of this invention proposes a hybrid evaluation mechanism that integrates physical mechanism model and machine learning model. By establishing a quantitative mapping relationship between equipment physical state and data error, and combining the ability of MSTCN-Attention model to identify unknown anomalies, it realizes accurate evaluation and hierarchical identification of data quality, overcomes the limitations of relying solely on physical model or data model, and the accuracy of data quality identification can reach more than 98%.

[0018] (3) The method of the present invention constructs a complete technical process of “state perception-quality assessment-decision execution”, breaks the technical barrier of the independent functional modules in the traditional system, realizes the technical leap of seabed-based equipment from “passive monitoring” to “active identification and management based on quality assessment”, can automatically switch working modes and generate quality labels according to the data quality index, and forms a complete technical solution.

[0019] (4) The method and system of the present invention work together to achieve a delay of less than 10 seconds in identifying and responding to typical faults such as abnormal power supply and sudden attitude changes in terms of status monitoring, and the completeness of status parameter acquisition reaches more than 95%. Even under extreme conditions of 30% sensor performance degradation, the data quality level can still be accurately identified. Finally, standardized data products with clear quality labels are output, which significantly improves the reliability and application value of long-term seabed observation data and promotes the transformation of marine observation from "data acquisition" to "quality controllable data service". Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for self-monitoring of the status of seabed-based equipment and dynamic evaluation of data quality according to an embodiment of the present invention.

[0021] Figure 2 This is a flowchart of step S2 in a method for self-monitoring the status of seabed-based equipment and dynamically evaluating data quality according to an embodiment of the present invention.

[0022] Figure 3 This is a flowchart of step S3 in a method for self-monitoring the status of seabed-based equipment and dynamically evaluating data quality according to an embodiment of the present invention.

[0023] Figure 4 This is a flowchart of step S23 in a method for self-monitoring the status of seabed-based equipment and dynamically evaluating data quality according to an embodiment of the present invention.

[0024] Figure 5 This is a schematic diagram of the structure of a seabed-based equipment status self-monitoring and data quality dynamic evaluation system according to an embodiment of the present invention.

[0025] Figure 6This is a schematic diagram of the structure of the self-monitoring module, the dynamic data quality assessment module, and the intelligent decision-making and control module in a seabed-based equipment status self-monitoring and data quality dynamic assessment system according to an embodiment of the present invention.

[0026] Figure 7 This is a schematic diagram of the working logic of a seabed-based equipment status self-monitoring and data quality dynamic evaluation system according to an embodiment of the present invention.

[0027] In the attached diagram: 1. State self-monitoring module; 11. Power supply monitoring unit; 12. Physical attitude monitoring unit; 13. Internal environment monitoring unit; 14. Sensor health monitoring unit; 2. Data quality dynamic evaluation module; 21. State parameter generation layer; 22. State-error correlation model; 23. MSTCN-Attention machine learning model; 24. Quality index fusion layer; 3. Intelligent decision-making and control module; 31. Execution unit; 32. Output unit. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] It should be noted that when a component is described as "fixed to" another component, it can be directly on the other component or may have a component in between. When a component is considered "connected to" another component, it can be directly connected to the other component or may have a component in between. When a component is considered "set on" another component, it can be directly set on the other component or may have a component in between. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0031] Please see Figure 1 A preferred embodiment of the present invention provides a method for self-monitoring of the status of seabed-based equipment and dynamic evaluation of data quality, comprising: S1 collects key status parameters of seabed-based equipment, including system power supply parameters, equipment attitude angle data, cabin temperature and humidity, cabin leakage, biofouling on equipment surface, and key operating parameters of scientific sensors.

[0032] Key status parameters are the raw monitoring data of various monitoring units, including system power supply parameters, equipment attitude angle data, cabin temperature and humidity, cabin leakage and biofouling on equipment surface, and key operating parameters of scientific sensors.

[0033] The power monitoring unit collects key parameters related to system power supply, including actual monitored voltage (V_actual), average operating current (I_avg), and remaining battery charge percentage (SOC). The power monitoring unit consists of a high-precision voltage sensor, a current sensor, and a smart fuel gauge chip. The output of this unit is used to determine the stability of the system power supply; abnormal voltage drops or abnormal current fluctuations will directly trigger the "unstable power supply" flag in the data quality label.

[0034] The physical attitude monitoring unit collects equipment attitude angle data, including roll and pitch (heading angle is not considered a core parameter due to its relatively minor impact on data accuracy). The core of the physical attitude monitoring unit is an integrated nine-axis inertial measurement unit (IMU), comprising a three-axis MEMS accelerometer, a three-axis MEMS gyroscope, and a three-axis magnetometer. The attitude stability index (ASI) is calculated based on the variance and frequency of abrupt changes in the attitude angle data. The monitoring data from this unit is directly input into a state-error correlation model to quantify the impact of equipment tilt on the accuracy of vector sensor data such as the ADCP (Acoustic Doppler Current Profiler).

[0035] The internal environmental monitoring unit collects data on cabin temperature and humidity, cabin leakage, and biofouling on equipment surfaces. This unit consists of digital temperature and humidity sensors, a resistive leak detection strip, and a biofouling monitoring device. The temperature and humidity sensors are deployed inside the main control cabin to monitor ambient temperature and relative humidity; the data is used to assess the operating environment of electronic components and sensor temperature drift. The leak detection strip is laid at the bottom of the cabin and detects leaks by monitoring changes in resistance. The biofouling monitoring device, which can employ an underwater camera or acoustic surface detector, periodically images or scans the equipment casing and acoustic sensor surfaces. Image recognition or echo intensity analysis is used to assess the coverage and thickness of biofouling. These monitoring results are then used to correlate and evaluate the sensitivity decay and acoustic emission power reduction of acoustic sensors (such as ADCPs and noise meters).

[0036] The sensor health monitoring unit collects key operating parameters of scientific sensors and analyzes their long-term trends and instantaneous anomalies. This unit is a dedicated monitoring interface for seabed-based scientific sensors. For ADCP (Acoustic Doppler Current Profiler), it monitors the transducer operating current, average echo intensity, and internal diagnostic flags; for pressure-type tide gauges, it monitors the pressure sensor's operating current and reading stability; for noise measuring instruments, it monitors the preamplifier gain and background noise level; and for acoustic communication equipment, it monitors the modem status, signal transmission power, and received signal-to-noise ratio. Furthermore, for optional equipment such as water quality meters (e.g., CTD, turbidimeter), key operating parameters are also monitored. By analyzing the long-term trends and instantaneous anomalies of these parameters, the unit calculates the Sensor Performance Degradation Index (SDI). Its output directly indicates the reliability of specific sensor data and triggers data quality degradation when performance degradation exceeds limits.

[0037] S2 employs a hybrid evaluation mechanism that combines a physical mechanism-based state-error correlation model with a data-driven MSTCN-Attention machine learning model to quantify the credibility of key state parameters and generate a data quality index.

[0038] Specifically, such as Figure 2 As shown, S2 includes: S21, standardize the key state parameters, and generate the power health index, attitude stability index and sensor performance degradation index based on the standardized key state parameters.

[0039] Key state parameters are standardized, including data calibration (inherent sensor errors, such as zero-point offset of voltage sensors), denoising (using moving average filtering to remove random noise), and normalization (mapping parameters of different dimensions to a unified range, such as voltage and current → [0,1], attitude angle → [-1,1], to avoid affecting model calculations due to differences in magnitude), so that monitoring data of different dimensions and magnitudes are comparable.

[0040] System power supply related parameters include actual monitored voltage value, average operating current, and remaining battery charge percentage. The formula for calculating the power health index is: Wherein, PHI is the power health index, α, β, and γ are experimentally calibrated weighting coefficients that satisfy α+β+γ=1, V_actual is the actual monitored voltage value, V_nominal is the system rated operating voltage, I_avg is the average operating current, I_max is the maximum allowable operating current, and SOC is the remaining battery charge percentage.

[0041] The variances of roll and pitch angles are calculated based on a set time window (usually 1 hour) to generate the Attitude Stability Index (ASI). The formula for calculating the Attitude Stability Index is as follows: Wherein, ASI stands for Attitude Stability Index. The variance of the roll angle within a set time window. This represents the variance of the pitch angle within a set time window.

[0042] The sensor performance degradation index is calculated independently for each sensor.

[0043] The performance degradation evaluation model for ADCP (Acoustic Doppler Current Profiler) is as follows: Here, SDI_ADCP is the sensor performance degradation index of the acoustic Doppler current profiler; I_actual is the current operating current; I_initial is the initial operating current; SNR_actual is the current signal-to-noise ratio; SNR_initial is the initial signal-to-noise ratio; BE_actual is the current echo intensity; and BE_initial is the initial echo intensity. Similar performance degradation assessment models exist for other acoustic communication equipment. This index can identify the gradual degradation of sensor performance at an early stage.

[0044] The performance degradation evaluation model for pressure-type tide gauges is as follows: Where I_actual is the current operating current, I_initial is the initial operating current, R_actual is the current pressure reading repeatability, R_initial is the initial pressure reading repeatability, S_actual is the current temperature drift stability coefficient, and S_initial is the initial temperature drift stability coefficient.

[0045] The performance degradation evaluation model for the CTD water quality analyzer is as follows: Where R_actual is the current electrode response speed, R_initial is the initial electrode response speed, A_actual is the current optical signal attenuation rate, A_initial is the initial optical signal attenuation rate, T_actual is the current temperature drift compensation coefficient stability, and T_initial is the initial temperature drift compensation coefficient stability.

[0046] S22, the standardized key state parameters are input into the state-error correlation model, and the expected error introduced by the physical state is calculated based on the physical mechanism through the state-error correlation model.

[0047] The state-error correlation model establishes a quantitative relationship between state parameters and data errors based on physical mechanisms, serving as the theoretical foundation for data quality assessment. The state-error correlation model comprises several specialized sub-models, each targeting different types of error sources: attitude-flow velocity error model, temperature-sensor drift model, and bio-attachment-acoustic performance model.

[0048] The attitude-velocity error model quantifies the impact of equipment tilting on ADCP measurement accuracy. This model establishes a specific mapping relationship between attitude angle and velocity measurement error, tailored to the unique tilting conditions of the seabed. Wherein, δV is the velocity measurement error vector, which is specifically used to characterize the degree of data quality attenuation under seabed tilt conditions, V_mesured is the ADCP measured velocity value, θ is the pitch angle, and V_offset is the error offset calibrated based on the specific installation location of the equipment and the local flow field characteristics.

[0049] The temperature-sensor drift model describes the impact of environmental temperature changes on sensor readings during long-term deployment on the seabed. This model establishes a temperature drift compensation mechanism suitable for long-term monitoring, taking into account the slow but persistent nature of seabed temperature changes. Where δC is the sensor reading drift, T is the current temperature, T_ref is the sensor calibration reference temperature, and k1 and k2 are temperature drift coefficients obtained through laboratory calibration.

[0050] A bioattachment-acoustic performance model was used to evaluate the impact of bioattachment on the performance of acoustic sensors. As deployment time increases, bioattachment on the sensor surface attenuates the acoustic signal; the model for this impact is as follows: Where A_loss is the total attenuation of the acoustic signal (dB), d is the thickness of the bio-attachment layer (mm), η is the attenuation coefficient per unit thickness, and A_base is the fundamental propagation loss of the acoustic system. This model provides a quantitative basis for assessing the degradation of acoustic data quality.

[0051] S23 uses the MSTCN-Attention machine learning model to intelligently identify abnormal patterns from historical and real-time data and generate anomaly probabilities.

[0052] The MSTCN-Attention machine learning model employs a multi-scale temporal convolutional network and attention mechanism (MSTCN-Attention model) to learn complex anomaly patterns from massive historical data, thus compensating for the shortcomings of physical models in identifying unknown anomalies.

[0053] Specifically, such as Figure 4As shown, the working process of the MSTCN-Attention machine learning model includes: S231, the input layer receives multidimensional time-series data of length L.

[0054] The MSTCN-Attention machine learning model receives multi-dimensional time-series data of length L, including all state parameters and sensor readings, as input. After standardization, the data is then used for feature extraction.

[0055] S232, the convolutional layer adopts a parallel convolutional structure, using three different kernel sizes (3, 5, and 7) to synchronously extract temporal features from the standardized multidimensional temporal data. The expression is:

[0056] Where H_i represents the temporal feature, X represents the input temporal data, k_i represents the size of the i-th convolutional kernel, W_i represents the weight matrix of the i-th convolutional kernel, and b_i represents the bias vector of the i-th convolutional kernel. This multi-scale design enables the model to simultaneously capture short-term mutations and long-term trend features.

[0057] S233, the attention mechanism module weights the temporal features output by the convolutional layer based on their importance, and the calculation expression is as follows:

[0058] Where Q is the query matrix, K is the key matrix, V is the value matrix, and d_k is the dimension of the key vector. The attention mechanism module weights the features output by the convolutional layer based on their importance, allowing the model to focus on the features and time points most relevant to the anomaly. Used to scale the dot product result and prevent gradient vanishing.

[0059] S234, the output layer generates anomaly probabilities using a fully connected network and a sigmoid activation function, expressed as follows:

[0060] Where P_anomaly is the anomaly probability, H_attention is the attention-weighted feature representation, W_o is the weight matrix of the output layer, b_o is the bias vector of the output layer, and σ is the sigmoid function. This output reflects the confidence level that there is an anomaly in the current data segment.

[0061] S24. Based on the expected error and the probability of anomalies, a data quality index is generated through weighted fusion. During the fusion process, the weight coefficients of the state-error association model and the MSTCN-Attention machine learning model are dynamically adjusted according to the application scenario and data characteristics.

[0062] This is accomplished through a quality index fusion layer, which integrates the outputs of the physical model and the machine learning model, generating the final Data Quality Index (DQI) through weighted fusion. During the fusion process, the system dynamically adjusts the weight coefficients of the physical model and the machine learning model according to different application scenarios and data characteristics. When a known type of equipment status anomaly is detected, the weight of the physical model is appropriately increased; when a new anomaly pattern appears, the weight of the machine learning model is increased. This adaptive fusion mechanism ensures the accuracy and robustness of the DQI assessment, providing a reliable basis for subsequent intelligent decision-making.

[0063] S3 executes preset control strategies based on the data quality index, realizes the switching of working modes and the automatic generation of data quality labels, and outputs standardized data product packages.

[0064] Its core lies in using a predefined working mode-quality threshold mapping table to achieve automatic switching of working modes and dynamic generation of data quality labels. Based on DQI, it achieves adaptive switching of working modes and automatic generation of data quality labels, forming a closed loop of "evaluation-decision-output," ensuring the reliability and traceability of output data. Specifically, such as... Figure 3 As shown, S3 includes: S31, executes multi-threshold judgment logic based on data quality index and key status parameters.

[0065] S32 encapsulates all data, status reports, and quality labels into a standardized data product package for output.

[0066] The multi-threshold judgment logic includes: when the data quality index is greater than the excellent threshold, maintain the current mode; when the data quality index is less than or equal to the degradation threshold or a specific abnormal state is detected, switch to a lower power consumption or safer working mode (such as switching from "full power" to "standard" or "energy saving" mode), and attach a structured quality label to the output data (this label clearly contains key metadata such as quality level, main influencing factors and timestamp); when the data quality index is less than or equal to the danger threshold or a fatal fault signal is received (such as cabin leakage), trigger the highest priority emergency response, switch to the safe mode and send a high-level warning simultaneously.

[0067] Ultimately, all data, status reports, and quality labels are packaged into a standardized data product package for output, thereby forming a complete closed-loop control from status perception to decision execution, ensuring that the system can output data with clear credibility under different operating conditions.

[0068] Another embodiment of the present invention also provides a system for self-monitoring of the status of seabed-based equipment and dynamic evaluation of data quality, such as... Figure 5 As shown, it includes a status self-monitoring module 1, a data quality dynamic evaluation module 2, and an intelligent decision-making and control module 3 connected in sequence.

[0069] The status self-monitoring module 1 is used to collect key status parameters of the seabed-based equipment. These key status parameters include system power supply parameters, equipment attitude angle data, cabin temperature and humidity, cabin leakage, biological adhesion status on the equipment surface, and key operating parameters of scientific sensors.

[0070] The data quality dynamic assessment module 2 is used to quantify the credibility of key state parameters and generate a data quality index by adopting a hybrid assessment mechanism that combines a physical mechanism-based state-error correlation model 22 with a data-driven MSTCN-Attention machine learning model 23.

[0071] The intelligent decision-making and control module 3 is used to execute preset control strategies based on the data quality index, realize the switching of working modes and the automatic generation of data quality labels, and output standardized data product packages.

[0072] like Figure 6 As shown, the self-monitoring module 1 includes: a power monitoring unit 11, a physical attitude monitoring unit 12, an internal environment monitoring unit 13, a sensor health monitoring unit 14, and a biological attachment monitoring device.

[0073] The power monitoring unit 11 consists of a voltage sensor, a current sensor, and a smart fuel meter chip, and is used to collect system power supply related parameters.

[0074] The physical attitude monitoring unit 12 adopts an integrated nine-axis inertial measurement unit, including a three-axis MEMS accelerometer, a three-axis MEMS gyroscope and a three-axis magnetometer, for collecting equipment attitude angle data.

[0075] The internal environment monitoring unit 13 includes a digital temperature and humidity sensor, a resistive leak detection strip, and a bio-attachment monitoring device, which are used to collect data on the temperature and humidity inside the cabin, the cabin leakage situation, and the bio-attachment status on the equipment surface.

[0076] The sensor health monitoring unit 14 is used to collect key operating parameters of the scientific sensor.

[0077] This unit is a dedicated monitoring interface for scientific sensors mounted on the seabed. For ADCP (Acoustic Doppler Current Profiler), it monitors the transducer operating current, average echo intensity, and internal diagnostic flags; for pressure-type tide gauges, it monitors the operating current and reading stability of the pressure sensor; for noise measuring instruments, it monitors the preamplifier gain and background noise level; and for acoustic communication equipment, it monitors the modem status, signal transmit power, and receive signal-to-noise ratio. Furthermore, for optional equipment such as water quality meters (e.g., CTD, turbidimeter), its key operating parameters are also monitored. By analyzing the long-term trends and instantaneous anomalies of these parameters, this unit calculates the Sensor Performance Degradation Index (SDI). Its output is directly used to identify the reliability of specific sensor data and triggers data quality degradation when performance degradation exceeds limits.

[0078] Bioattachment monitoring devices use underwater cameras or acoustic surface detectors to assess the coverage and thickness of bioattachments through image recognition or echo intensity analysis.

[0079] Scientific sensors include at least acoustic Doppler current profilers, pressure-type tide gauges, and CTD water quality meters. They may also include noise measuring instruments and acoustic communication equipment.

[0080] like Figure 6 As shown, the data quality dynamic evaluation module 2 includes a state parameter generation layer 21, a state-error correlation model 22, an MSTCN-Attention machine learning model 23, and a quality index fusion layer 24.

[0081] The state parameter generation layer 21 is used to standardize key state parameters and generate power health index, attitude stability index and sensor performance degradation index based on the standardized key state parameters.

[0082] The state-error correlation model 22 is used to calculate the expected error introduced by the physical state based on the physical mechanism.

[0083] MSTCN-Attention machine learning model 23 is used to intelligently identify abnormal patterns from historical and real-time data and generate anomaly probabilities.

[0084] The quality index fusion layer 24 is used to generate a data quality index by weighted fusion based on the expected error and the probability of anomalies. During the fusion process, the weight coefficients of the state-error association model 22 and the MSTCN-Attention machine learning model 23 are dynamically adjusted according to the application scenario and data characteristics.

[0085] like Figure 6 As shown, the intelligent decision-making and control module 3 includes an execution unit 31 and an output unit 32.

[0086] Execution unit 31 is used to execute multi-threshold judgment logic based on data quality index and key status parameters.

[0087] Output unit 32 is used to encapsulate all data, status reports and quality labels into a standardized data product package for output.

[0088] The multi-threshold judgment logic includes: when the data quality index is greater than the good threshold, maintain the current mode; when the data quality index is less than or equal to the degradation threshold or a specific abnormal state is detected, switch to a lower power consumption or safer working mode, and attach a structured quality label to the output data; when the data quality index is less than or equal to the danger threshold or a fatal fault signal is received, trigger the highest priority emergency response, switch to the safe mode and send a high-level warning simultaneously.

[0089] System power supply related parameters include actual monitored voltage value, average operating current, and remaining battery charge percentage. The formula for calculating the power health index is:

[0090] Wherein, PHI is the power health index, α, β, and γ are experimentally calibrated weighting coefficients that satisfy α+β+γ=1, V_actual is the actual monitored voltage value, V_nominal is the system rated operating voltage, I_avg is the average operating current, I_max is the maximum allowable operating current, and SOC is the remaining battery charge percentage.

[0091] The equipment attitude angle data includes the roll and pitch angles within a set time window. The formula for calculating the attitude stability index is:

[0092] Wherein, ASI stands for Attitude Stability Index. The variance of the roll angle within a set time window. This represents the variance of the pitch angle within a set time window.

[0093] The sensor performance degradation index is calculated independently for each sensor.

[0094] The performance degradation evaluation model for the acoustic Doppler current profiler is as follows:

[0095] Among them, SDI_ADCP is the sensor performance degradation index of the acoustic Doppler current profiler, I_actual is the current operating current, I_initial is the initial operating current; SNR_actual is the current signal-to-noise ratio, SNR_initial is the initial signal-to-noise ratio; BE_actual is the current echo intensity, BE_initial is the initial echo intensity.

[0096] The performance degradation evaluation model for pressure-type tide gauges is as follows:

[0097] Where I_actual is the current operating current, I_initial is the initial operating current, R_actual is the current pressure reading repeatability, R_initial is the initial pressure reading repeatability, S_actual is the current temperature drift stability coefficient, and S_initial is the initial temperature drift stability coefficient.

[0098] The performance degradation evaluation model for the CTD water quality analyzer is as follows:

[0099] Where R_actual is the current electrode response speed, R_initial is the initial electrode response speed, A_actual is the current optical signal attenuation rate, A_initial is the initial optical signal attenuation rate, T_actual is the current temperature drift compensation coefficient stability, and T_initial is the initial temperature drift compensation coefficient stability.

[0100] The state-error correlation model 22 includes: attitude-flow velocity error model, temperature-sensor drift model and bio-attachment-acoustic performance model.

[0101] The mathematical expression for the attitude-flow error model is:

[0102] Where δV is the velocity measurement error vector, V_measured is the ADCP measured velocity value, θ is the pitch angle, and V_offset is the error offset calibrated based on the equipment installation location and local flow field characteristics.

[0103] The mathematical expression for the temperature-sensor drift model is:

[0104] Where δC is the sensor reading drift, T is the current temperature, T_ref is the sensor calibration reference temperature, and k1 and k2 are temperature drift coefficients obtained through laboratory calibration.

[0105] The mathematical expression for the bioattachment-acoustic performance model is:

[0106] Where A_loss is the total attenuation of the acoustic signal, d is the thickness of the biological attachment layer, η is the attenuation coefficient per unit thickness, and A_base is the basic propagation loss of the acoustic system.

[0107] The working process of MSTCN-Attention machine learning model 23 includes: The input layer receives multidimensional time-series data of length L and performs standardization processing on the multidimensional time-series data.

[0108] The convolutional layers employ a parallel convolutional structure, using convolutional kernels of 3, 5, and 7 different sizes to synchronously extract temporal features from the standardized multidimensional temporal data. The expression is as follows:

[0109] Where H_i is the temporal feature, X is the input temporal data, k_i is the size of the i-th convolutional kernel, W_i is the weight matrix of the i-th convolutional kernel, and b_i is the bias vector of the i-th convolutional kernel.

[0110] The attention mechanism module weights the temporal features output by the convolutional layer based on their importance; the calculation expression is as follows:

[0111] Where Q is the query matrix, K is the key matrix, V is the value matrix, and d_k is the dimension of the key vector.

[0112] The output layer generates anomaly probabilities using a fully connected network and a sigmoid activation function, expressed as:

[0113] Where P_anomaly is the anomaly probability, H_attention is the attention-weighted feature representation, W_o is the weight matrix of the output layer, b_o is the bias vector of the output layer, and σ is the sigmoid function.

[0114] The working logic of the seabed-based equipment status self-monitoring and data quality dynamic evaluation system is as follows: Figure 7 As shown.

[0115] The above description is a detailed description of the preferred embodiments of the present invention. However, the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modifications made under the technical spirit of the present invention should fall within the patent scope covered by the present invention.

Claims

1. A method for self-monitoring the condition of seabed-based equipment and dynamically evaluating data quality, characterized in that, include: S1. Collect key status parameters of the seabed-based equipment. The key status parameters include system power supply parameters, equipment attitude angle data, cabin temperature and humidity, cabin leakage, biological attachment status on the equipment surface, and key operating parameters of scientific sensors. S2 employs a hybrid evaluation mechanism that combines a physical mechanism-based state-error correlation model with a data-driven MSTCN-Attention machine learning model to quantify the credibility of key state parameters and generate a data quality index. S3 executes preset control strategies based on the data quality index, realizes the switching of working modes and the automatic generation of data quality labels, and outputs standardized data product packages; S2 includes: S21, standardize the key state parameters, and generate the power health index, attitude stability index and sensor performance degradation index based on the standardized key state parameters. S22, input the standardized key state parameters into the state-error correlation model, and calculate the expected error introduced by the physical state based on the physical mechanism through the state-error correlation model; S23, input the standardized key state parameters into the MSTCN-Attention machine learning model, and use the MSTCN-Attention machine learning model to intelligently identify abnormal patterns from historical and real-time data and generate abnormal probabilities. S24. Based on the expected error and the probability of anomalies, a data quality index is generated through weighted fusion. During the fusion process, the weight coefficients of the state-error association model and the MSTCN-Attention machine learning model are dynamically adjusted according to the application scenario and data characteristics.

2. The method for self-monitoring of the status of seabed-based equipment and dynamic evaluation of data quality according to claim 1, characterized in that, S3 includes: S31, execute multi-threshold judgment logic based on data quality index and key status parameters; S32 encapsulates all data, status reports, and quality labels into a standardized data product package for output. The multi-threshold judgment logic includes: maintaining the current mode when the data quality index is greater than the excellent threshold; switching to a lower power consumption or safer working mode when the data quality index is less than or equal to the degradation threshold or a specific abnormal state is detected, while attaching a structured quality label to the output data; and triggering the highest priority emergency response when the data quality index is less than or equal to the danger threshold or a fatal fault signal is received, switching to the safe mode and sending a high-level warning simultaneously.

3. The method for self-monitoring of the status of seabed-based equipment and dynamic evaluation of data quality according to claim 2, characterized in that, The process for collecting data on the bio-attachment status of the equipment surface is as follows: The coverage and thickness of the attachments are assessed by periodically imaging the equipment shell and acoustic sensor surface using an underwater camera or periodically scanning the equipment shell and acoustic sensor surface using an acoustic surface detector, and by image recognition or echo intensity analysis.

4. The method for self-monitoring of the status and dynamic evaluation of data quality of seabed-based equipment according to claim 3, characterized in that, The system power supply related parameters include the actual monitored voltage value, average operating current, and remaining battery power percentage. The formula for calculating the power health index is as follows: Wherein, PHI is the power health index, α, β, γ are experimentally calibrated weighting coefficients that satisfy α+β+γ=1, V_actual is the actual monitored voltage value, V_nominal is the system rated operating voltage, I_avg is the average operating current, I_max is the maximum allowable operating current, and SOC is the remaining battery charge percentage. The equipment attitude angle data includes the roll angle and pitch angle within a set time window, and the formula for calculating the attitude stability index is: Wherein, ASI stands for Attitude Stability Index. The variance of the roll angle within a set time window. The variance of the pitch angle within the set time window; The sensor performance degradation index is calculated independently for each sensor. The performance degradation evaluation model for the acoustic Doppler current profiler is as follows: Wherein, SDI_ADCP is the sensor performance degradation index of the acoustic Doppler current profiler, I_actual is the current operating current, I_initial is the initial operating current; SNR_actual is the current signal-to-noise ratio, SNR_initial is the initial signal-to-noise ratio; BE_actual is the current echo intensity, BE_initial is the initial echo intensity; The performance degradation evaluation model for pressure-type tide gauges is as follows: Where I_actual is the current operating current, I_initial is the initial operating current, R_actual is the current pressure reading repeatability, R_initial is the initial pressure reading repeatability, S_actual is the current temperature drift stability coefficient, and S_initial is the initial temperature drift stability coefficient. The performance degradation evaluation model for the CTD water quality analyzer is as follows: Where R_actual is the current electrode response speed, R_initial is the initial electrode response speed, A_actual is the current optical signal attenuation rate, A_initial is the initial optical signal attenuation rate, T_actual is the current temperature drift compensation coefficient stability, and T_initial is the initial temperature drift compensation coefficient stability.

5. The method for self-monitoring of the status of seabed-based equipment and dynamic evaluation of data quality according to claim 3, characterized in that, In S22, the state-error correlation model includes: attitude-flow velocity error model, temperature-sensor drift model and bio-attachment-acoustic performance model; The mathematical expression for the attitude-flow velocity error model is: Where δV is the velocity measurement error vector, V_measured is the ADCP measured velocity value, θ is the pitch angle, and V_offset is the error offset calibrated based on the equipment installation position and local flow field characteristics; The mathematical expression for the temperature-sensor drift model is: Where δC is the sensor reading drift, T is the current temperature, T_ref is the sensor calibration reference temperature, and k1 and k2 are the temperature drift coefficients obtained through laboratory calibration; The mathematical expression for the bioattachment-acoustic performance model is: Where A_loss is the total attenuation of the acoustic signal, d is the thickness of the biological attachment layer, η is the attenuation coefficient per unit thickness, and A_base is the basic propagation loss of the acoustic system.

6. The method for self-monitoring of the status and dynamic evaluation of data quality of seabed-based equipment according to claim 3, characterized in that, In S23, the working process of the MSTCN-Attention machine learning model includes: S231, The input layer receives multidimensional time-series data of length L; S232, the convolutional layer adopts a parallel convolutional structure, using three different kernel sizes (3, 5, and 7) to synchronously extract temporal features from the standardized multidimensional temporal data. The expression is: Where H_i is the temporal feature, X is the input temporal data, k_i is the size of the i-th convolutional kernel, W_i is the weight matrix of the i-th convolutional kernel, and b_i is the bias vector of the i-th convolutional kernel. S233, the attention mechanism module weights the temporal features output by the convolutional layer based on their importance, and the calculation expression is as follows: Where Q is the query matrix, K is the key matrix, V is the value matrix, and d_k is the dimension of the key vector; S234, the output layer generates anomaly probabilities using a fully connected network and a sigmoid activation function, expressed as follows: Where P_anomaly is the anomaly probability, H_attention is the attention-weighted feature representation, W_o is the weight matrix of the output layer, b_o is the bias vector of the output layer, and σ is the sigmoid function.

7. A self-monitoring system for the status of seabed-based equipment and a dynamic evaluation system for data quality, characterized in that: It includes a status self-monitoring module, a data quality dynamic assessment module, and an intelligent decision-making and control module connected in sequence; The self-monitoring module is used to collect key status parameters of the seabed-based equipment. These key status parameters include system power supply parameters, equipment attitude angle data, cabin temperature and humidity, cabin leakage, biological adhesion status on the equipment surface, and key operating parameters of scientific sensors. The data quality dynamic evaluation module is used to quantify the credibility of key state parameters and generate a data quality index by employing a hybrid evaluation mechanism that combines a state-error correlation model based on physical mechanisms with a data-driven MSTCN-Attention machine learning model. The intelligent decision-making and control module is used to execute preset control strategies based on the data quality index, realize the switching of working modes and the automatic generation of data quality labels, and output standardized data product packages. The status self-monitoring module includes: The power monitoring unit consists of a voltage sensor, a current sensor, and a smart fuel meter chip, and is used to collect system power supply related parameters. The physical attitude monitoring unit adopts an integrated nine-axis inertial measurement unit, including a three-axis MEMS accelerometer, a three-axis MEMS gyroscope and a three-axis magnetometer, to collect equipment attitude angle data; The internal environment monitoring unit includes a digital temperature and humidity sensor, a resistive leak detection strip, and a biofouling monitoring device, which are used to collect data on the temperature and humidity inside the cabin, the cabin leakage situation, and the biofouling status on the equipment surface. Sensor health monitoring unit, used to collect key operating parameters of scientific sensors; The bio-attachment monitoring device uses an underwater camera or an acoustic surface detector to assess the coverage and thickness of the attached material through image recognition or echo intensity analysis. The scientific sensors include at least an acoustic Doppler current profiler, a pressure-type tide gauge, and a CTD water quality meter. The data quality dynamic assessment module includes: The state parameter generation layer is used to standardize key state parameters and generate power health index, attitude stability index and sensor performance degradation index based on the standardized key state parameters. State-error correlation model, used to calculate the expected error introduced by physical state based on physical mechanism; The MSTCN-Attention machine learning model is used to intelligently identify abnormal patterns from historical and real-time data and generate anomaly probabilities. The quality index fusion layer is used to generate a data quality index by weighted fusion based on expected error and anomaly probability. During the fusion process, the weight coefficients of the state-error association model and the MSTCN-Attention machine learning model are dynamically adjusted according to the application scenario and data characteristics. The intelligent decision-making and control module includes: The execution unit is used to execute multi-threshold judgment logic based on the data quality index and key status parameters; The output unit is used to encapsulate all data, status reports, and quality labels into a standardized data product package for output. The multi-threshold judgment logic includes: maintaining the current mode when the data quality index is greater than the excellent threshold; switching to a lower power consumption or safer working mode when the data quality index is less than or equal to the degradation threshold or a specific abnormal state is detected, while attaching a structured quality label to the output data; and triggering the highest priority emergency response when the data quality index is less than or equal to the danger threshold or a fatal fault signal is received, switching to the safe mode and sending a high-level warning simultaneously.

8. The seabed-based equipment status self-monitoring and data quality dynamic evaluation system according to claim 7, characterized in that, The system power supply related parameters include the actual monitored voltage value, average operating current, and remaining battery power percentage. The formula for calculating the power health index is as follows: Wherein, PHI is the power health index, α, β, γ are experimentally calibrated weighting coefficients that satisfy α+β+γ=1, V_actual is the actual monitored voltage value, V_nominal is the system rated operating voltage, I_avg is the average operating current, I_max is the maximum allowable operating current, and SOC is the remaining battery charge percentage. The equipment attitude angle data includes the roll angle and pitch angle within a set time window, and the formula for calculating the attitude stability index is: Wherein, ASI stands for Attitude Stability Index. The variance of the roll angle within a set time window. The variance of the pitch angle within the set time window; The sensor performance degradation index is calculated independently for each sensor. The performance degradation evaluation model for the acoustic Doppler current profiler is as follows: Wherein, SDI_ADCP is the sensor performance degradation index of the acoustic Doppler current profiler, I_actual is the current operating current, I_initial is the initial operating current; SNR_actual is the current signal-to-noise ratio, SNR_initial is the initial signal-to-noise ratio; BE_actual is the current echo intensity, BE_initial is the initial echo intensity; The performance degradation evaluation model for pressure-type tide gauges is as follows: Where I_actual is the current operating current, I_initial is the initial operating current, R_actual is the current pressure reading repeatability, R_initial is the initial pressure reading repeatability, S_actual is the current temperature drift stability coefficient, and S_initial is the initial temperature drift stability coefficient. The performance degradation evaluation model for the CTD water quality analyzer is as follows: Where R_actual is the current electrode response speed, R_initial is the initial electrode response speed, A_actual is the current optical signal attenuation rate, A_initial is the initial optical signal attenuation rate, T_actual is the current temperature drift compensation coefficient stability, and T_initial is the initial temperature drift compensation coefficient stability.

9. The seabed-based equipment status self-monitoring and data quality dynamic evaluation system according to claim 7, characterized in that, The state-error correlation model includes: attitude-flow velocity error model, temperature-sensor drift model, and bioattachment-acoustic performance model; The mathematical expression for the attitude-flow velocity error model is: Where δV is the velocity measurement error vector, V_measured is the ADCP measured velocity value, θ is the pitch angle, and V_offset is the error offset calibrated based on the equipment installation position and local flow field characteristics; The mathematical expression for the temperature-sensor drift model is: Where δC is the sensor reading drift, T is the current temperature, T_ref is the sensor calibration reference temperature, and k1 and k2 are the temperature drift coefficients obtained through laboratory calibration; The mathematical expression for the bioattachment-acoustic performance model is: Where A_loss is the total attenuation of the acoustic signal, d is the thickness of the biological attachment layer, η is the attenuation coefficient per unit thickness, and A_base is the basic propagation loss of the acoustic system.

10. The seabed-based equipment status self-monitoring and data quality dynamic evaluation system according to claim 7, characterized in that, The working process of the MSTCN-Attention machine learning model includes: The input layer receives multi-dimensional time-series data of length L and performs standardization processing on the multi-dimensional time-series data. The convolutional layers employ a parallel convolutional structure, using convolutional kernels of 3, 5, and 7 different sizes to synchronously extract temporal features from the standardized multidimensional temporal data. The expression is as follows: Where H_i is the temporal feature, X is the input temporal data, k_i is the size of the i-th convolutional kernel, W_i is the weight matrix of the i-th convolutional kernel, and b_i is the bias vector of the i-th convolutional kernel. The attention mechanism module weights the temporal features output by the convolutional layer based on their importance; the calculation expression is as follows: Where Q is the query matrix, K is the key matrix, V is the value matrix, and d_k is the dimension of the key vector; The output layer generates anomaly probabilities using a fully connected network and a sigmoid activation function, expressed as: Where P_anomaly is the anomaly probability, H_attention is the attention-weighted feature representation, W_o is the weight matrix of the output layer, b_o is the bias vector of the output layer, and σ is the sigmoid function.

Images