Time sequence causal chain diagnosis method and system for submersible fault

Abstract

The invention relates to the field of equipment diagnosis, and discloses a submersible fault time sequence causal chain diagnosis method and system, and the method comprises the steps: obtaining multi-dimensional operation monitoring data in the operation process of a submersible; performing exception association analysis on the multi-dimensional operation monitoring data to obtain a cross-dimensional collaborative exception coefficient of an association unit group under each monitoring dimension; performing threshold comparison on all the cross-dimension collaborative anomaly coefficients, and determining each anomaly association unit group; and determining a diagnosis dimension trigger signal of the submersible according to the number of the abnormal association unit groups and the evolution characteristics of the cross-dimension collaborative abnormal coefficients of all the abnormal association unit groups. And a diagnosis dimension trigger signal is dynamically generated through the change and evolution characteristics of the cross-dimension collaborative abnormal coefficient, so that fault diagnosis scheduling of the submersible is triggered, and the response mechanism ensures the flexibility and adaptability in the diagnosis process.

Description

Technical Field

[0001] This application relates to the field of equipment diagnostics, and in particular to a method and system for diagnosing time-series causal chain faults in submersibles. Background Technology

[0002] Currently, traditional methods typically focus on fault diagnosis using data from a single monitoring dimension. This approach neglects the interrelationships and synergies between various monitoring dimensions, which may prevent timely detection of fault linkages between different monitoring dimensions, resulting in insufficient sensitivity and accuracy in fault detection. Furthermore, these methods are often based on pre-set static models and lack the ability to dynamically adjust according to real-time data changes. This means that when the submersible's working environment or state changes, the diagnostic system's response may not be flexible enough, leading to delayed or inaccurate fault detection.

[0003] Furthermore, traditional methods often focus only on the initial stage of a fault or a specific monitoring unit, while ignoring the propagation path and process of the fault. This limitation makes it difficult for traditional methods to comprehensively assess the scope of the fault's impact, especially in complex systems, which may lead to large-scale systemic faults going undetected in a timely manner. Summary of the Invention

[0004] This application provides a method for diagnosing the temporal causal chain of submersible malfunctions, in order to at least partially solve the above-mentioned technical problems.

[0005] To achieve the above objectives, according to a first aspect of this application, a method for diagnosing the temporal causal chain of submersible malfunctions is provided, comprising: Acquire multi-dimensional operational monitoring data during the operation of the submersible; perform anomaly correlation analysis on the multi-dimensional operational monitoring data to obtain the cross-dimensional collaborative anomaly coefficient of the associated unit group under each monitoring dimension; Threshold comparison is performed on all the cross-dimensional collaborative anomaly coefficients to determine each anomaly association unit group; based on the number of the anomaly association unit groups and the evolution characteristics of the cross-dimensional collaborative anomaly coefficients of all the anomaly association unit groups, the diagnostic dimension trigger signal of the submersible is determined. In response to the diagnostic dimension trigger signal, a target diagnostic scheduling signal is determined based on the causal dependency characteristics of the historical operating data of all monitoring units of the submersible and the cross-dimensional collaborative anomaly coefficient of the abnormal associated unit group; the fault diagnosis path of the submersible is adapted and adjusted using the target diagnostic scheduling signal to obtain an adapted diagnostic path; Real-time sensing data of all sensing nodes of the adaptive diagnostic path is acquired. Based on the temporal change characteristics and compartment association characteristics of the real-time sensing data, a fault transmission feature vector is determined. Based on the multi-dimensional similarity comparison results of the fault transmission feature vector, the target accuracy adjustment parameters of the adaptive diagnostic path are determined. The real-time sensing data is subjected to interference suppression processing using the target accuracy adjustment parameters. Based on the processed real-time sensing data, fault temporal causal chain inference is performed to obtain the complete temporal causal chain of the submersible fault.

[0006] According to a second aspect of this application, a time-series causal chain diagnostic system for submersible malfunctions is provided, comprising: An anomaly correlation module is used to acquire multi-dimensional operational monitoring data during the operation of the submersible; perform anomaly correlation analysis on the multi-dimensional operational monitoring data to obtain the cross-dimensional collaborative anomaly coefficient of the correlation unit group under each monitoring dimension; The diagnostic trigger module is used to perform threshold comparison on all the cross-dimensional collaborative anomaly coefficients to determine each anomaly association unit group; and to determine the diagnostic dimension trigger signal of the submersible based on the number of the anomaly association unit groups and the evolution characteristics of the cross-dimensional collaborative anomaly coefficients of all the anomaly association unit groups. The path adjustment module is used to respond to the diagnostic dimension trigger signal, determine the target diagnostic scheduling signal based on the causal dependency characteristics of the historical operating data of all monitoring units of the submersible and the cross-dimensional collaborative anomaly coefficient of the abnormal associated unit group; and adapt and adjust the fault diagnosis path of the submersible with the target diagnostic scheduling signal to obtain an adapted diagnostic path. The precision adjustment module is used to acquire real-time sensing data of all sensing nodes of the adaptive diagnostic path, determine the fault transmission feature vector based on the temporal change characteristics and compartment association characteristics of the real-time sensing data, and determine the target precision adjustment parameters of the adaptive diagnostic path based on the multi-dimensional similarity comparison results of the fault transmission feature vector. The causal reasoning module is used to perform interference suppression processing on the real-time sensing data by adjusting the parameters according to the target accuracy, and to perform fault temporal causal chain reasoning based on the processed real-time sensing data to obtain the complete temporal causal chain of the submersible fault.

[0007] In summary, the embodiments of this application, through multi-dimensional operational monitoring data analysis, can acquire data from various monitoring units of the submersible in real time, thereby providing early warnings of various potential faults of the submersible. By using a spatiotemporal causal analysis network, it can process and analyze the complex correlations between data, thereby more accurately identifying the root causes and potential risks of faults. Moreover, by calculating cross-dimensional collaborative anomaly coefficients, it can effectively indicate the degree of fault correlation between different monitoring dimensions. This method helps to discover the mutual influence of multiple monitoring dimensions, enhances the reliability and accuracy of fault diagnosis, and is more sensitive than traditional methods.

[0008] In this embodiment, by dynamically generating diagnostic dimension trigger signals through the changes and evolution characteristics of cross-dimensional collaborative anomaly coefficients, fault diagnosis scheduling of the submersible is triggered. This response mechanism ensures flexibility and adaptability in the diagnostic process, enabling real-time adjustments based on real-time data, thereby improving the efficiency and accuracy of fault diagnosis. Moreover, by calculating temporal and spatial propagation feature vectors, the propagation process of the fault can be captured in detail. This multi-dimensional similarity comparison method improves the accuracy of fault prediction and also helps to better understand how the fault propagates from one monitoring unit to other units, effectively avoiding large-scale failures of the submersible due to a single fault.

[0009] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart of the steps of the time-series causal chain diagnosis method for submersible faults provided in the exemplary embodiments of this application; Figure 2 This is a system schematic diagram of the time-series causal chain diagnosis system for submersible faults provided in an exemplary embodiment of this application; Explanation of the attached diagram labels: 1. Anomaly Association Module; 2. Diagnosis Trigger Module; 3. Path Adjustment Module; 4. Precision Adjustment Module; 5. Causal Reasoning Module. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of this application.

[0012] This application provides a method for diagnosing the temporal causal chain of submersible failures. Please refer to [link to relevant documentation]. Figure 1 The time-series causal chain diagnosis method for submersible faults provided in this application includes the following steps: S1. Acquire multi-dimensional operational monitoring data during the operation of the submersible; use a pre-set spatiotemporal causal analysis network to perform anomaly correlation analysis on the multi-dimensional operational monitoring data to obtain the cross-dimensional collaborative anomaly coefficient of the associated unit group under each monitoring dimension.

[0013] It should be noted that the submersible generates a large amount of multi-dimensional data during operation, such as temperature, pressure, current, voltage, and vibration. This data directly reflects the operating status of the submersible's various systems and components. In this step, it is necessary to acquire this multi-dimensional monitoring data and analyze it using a pre-set spatiotemporal causal analysis network. This network helps to identify whether there are any abnormal correlations between the data. For example, it may be found that when the temperature rises abnormally, it is accompanied by a sudden change in current and increased vibration. The analysis results will provide a "cross-dimensional collaborative anomaly coefficient," which can indicate the strength of the abnormal relationship between different monitoring dimensions. A specific example: Suppose that when a submersible is working in a deep-sea environment, the temperature sensor shows a sharp rise in temperature, while the current sensor shows large fluctuations in current, and the vibration sensor also shows abnormalities; through spatiotemporal causal analysis network analysis, it is found that there is a strong correlation between these abnormal data, generating a group of related units with a co-anomaly coefficient of 0.8.

[0014] S2. Perform threshold comparison on all cross-dimensional collaborative anomaly coefficients to determine each anomaly association unit group; determine the diagnostic dimension trigger signal of the submersible based on the number of anomaly association unit groups and the evolution characteristics of the cross-dimensional collaborative anomaly coefficients of all anomaly association unit groups. It should be noted that threshold comparisons are performed on all calculated co-anomaly coefficients to determine which data combinations belong to the anomalous correlation unit groups. Based on the number of these anomalous correlation unit groups and their evolutionary characteristics, a "diagnostic dimension trigger signal" is ultimately generated. This signal can indicate which part of the submersible system may have a fault and requires further diagnosis. A specific example: By comparing the abnormal correlation coefficients, frequent abnormal fluctuations were found in the three monitoring dimensions of temperature, current and vibration, which met the threshold requirements and generated a diagnostic signal, indicating that there may be a problem with the submersible's power system.

[0015] S3. In response to the diagnostic dimension trigger signal, based on the causal dependency characteristics of the historical operating data of all monitoring units of the submersible and the cross-dimensional collaborative anomaly coefficient of the abnormal associated unit group, determine the target diagnostic scheduling signal; use the target diagnostic scheduling signal to adapt and adjust the fault diagnosis path of the submersible to obtain the adapted diagnostic path. It should be noted that, in response to the diagnostic dimension trigger signal, the system will determine a target diagnostic scheduling signal based on the causal dependency characteristics of the submersible's historical operational data and known abnormal correlation unit groups. The role of this signal is to adjust the diagnostic path to help pinpoint the source of the fault more accurately. By adjusting the diagnostic path, priority can be given to potential fault points related to the trigger signal. Example: The system determines the possible location of the fault based on the triggered diagnostic signal, and selects to prioritize the troubleshooting path to the submersible's battery pack and power system for more detailed inspection.

[0016] S4. Obtain real-time sensing data of all sensing nodes in the adaptive diagnostic path. Based on the temporal change characteristics and compartment association characteristics of the real-time sensing data, determine the fault transmission feature vector. Based on the multi-dimensional similarity comparison results of the fault transmission feature vector, determine the target accuracy adjustment parameters of the adaptive diagnostic path. It should be noted that the submersible's sensing system acquires monitoring data from different components in real time. By analyzing the temporal variation characteristics and compartment correlation characteristics of this real-time data, the system determines the "fault propagation feature vector". These feature vectors can reveal the fault propagation path between various systems of the submersible. For example, a fault may start from the battery pack, affect the current system, and then propagate to the temperature control system. Example: Suppose the battery pack temperature rises rapidly and the current fluctuates abnormally. By analyzing this data, the system discovers that battery overheating may cause unstable current supply, which in turn affects the stability of other systems, and generates a fault propagation feature vector.

[0017] S5. Adjust the parameters to target accuracy and perform interference suppression processing on the real-time sensing data. Based on the processed real-time sensing data, perform fault temporal causal chain reasoning to obtain the complete temporal causal chain of the submersible fault.

[0018] It should be noted that temporal causal chain reasoning is performed based on the processed real-time sensing data. The purpose of this step is to deduce the complete temporal causal chain of the submersible failure by analyzing the changing trends and causal relationships of the temporal data. Through this causal chain, the system can identify the source, development process, and possible scope of impact of the failure. Example: Through temporal causal chain reasoning, the system deduces that the overheating of the submersible's battery pack is caused by an internal short circuit in the battery, which leads to abnormal current, thereby affecting the entire power system and ultimately causing insufficient power output from the submersible, making it unable to maintain normal operation.

[0019] In some embodiments, the cross-dimensional collaborative anomaly coefficient is used to indicate the degree of fault correlation between units in different monitoring dimensions; A pre-defined spatiotemporal causal analysis network is used to perform anomaly correlation analysis on multi-dimensional operational monitoring data, obtaining cross-dimensional collaborative anomaly coefficients for correlated unit groups under each monitoring dimension, including: The historical data of each monitoring dimension in the multi-dimensional operational monitoring data is subjected to periodic feature quantization processing to obtain the dimensional feature vector of each monitoring dimension; wherein, the periodic feature quantization value includes at least one of waveform factor, kurtosis factor and skewness factor; Based on the structural correlation map of the submersible, the monitoring unit is divided into correlation unit groups; By using the cross-dimensional correlation layer in the spatiotemporal causal analysis network, the cross-correlation distance between the dimensional feature vectors of different monitoring dimensions within each correlation unit group is calculated; the mean of the cross-dimensional collaborative anomaly coefficient of the corresponding correlation unit group is used.

[0020] It should be noted that the historical data of each monitoring dimension of the submersible are quantified according to their periodic characteristics. The purpose of periodic characteristic quantification is to extract representative information from the data, which can reflect the fluctuation patterns and anomaly patterns of the data. Specifically, the quantified periodic characteristic values ​​include at least one of the following three statistical features: waveform factor: describes the shape characteristics of the signal waveform, which can reflect the periodic and non-periodic fluctuations of the signal; kurtosis factor: used to measure the sharpness of the signal, indicating whether there are abrupt changes in the signal, usually related to instantaneous anomalies; skewness factor: indicates the degree of skewness of the data distribution, which can reveal whether the data deviates from the normal state, especially long-term deviations. Through these feature quantification processes, a dimensional feature vector for each monitoring dimension is generated. For example, assuming that the historical data of the submersible's temperature sensor in the deep sea is collected for periodic analysis, the extracted feature values ​​may be a waveform factor of 1.5, a kurtosis factor of 3.2, and a skewness factor of -0.5. These values ​​are combined to form the dimensional feature vector of the temperature sensor. Based on the structural correlation diagram of the submersible, the monitoring unit is systematically divided into different correlation unit groups. These groups are mainly based on the physical structure of the submersible, functional modules, and the inherent relationships between various sensors. For example, temperature, vibration, and current sensors in the power system are usually grouped into the same correlation unit group because they are often affected by the same operating factors, and the monitoring data usually show a strong correlation when a fault occurs. Specifically, a submersible's power system may include components such as battery packs, engines, and thrusters. When performing condition analysis, temperature sensors, current sensors, and vibration sensors installed on these components can be classified into a group of related units. They collectively reflect the operating status of the power system, and the data are closely related both physically and functionally. Through the cross-dimensional correlation layer in the spatiotemporal causal analysis network, the system calculates the cross-correlation distance between the feature vectors corresponding to different monitoring dimensions within each correlation unit group. This step aims to assess whether different monitoring data exhibit similar abnormal patterns. The cross-correlation distance is used to measure the consistency of the fluctuation trend or abnormal behavior of two data sequences. The smaller the distance, the higher their synchronicity or correlation. The average value of the correlation distance between all monitoring dimension feature vectors within the same associated unit group is calculated. This average value is defined as the cross-dimensional collaborative anomaly coefficient of the group. This coefficient reflects the overall degree of collaborative change of multiple monitoring indicators within the group when anomalies occur.

[0021] For example, suppose a dynamic system's associated unit group includes three monitoring dimensions: temperature, current, and vibration. After preliminary quantization, their corresponding dimensional feature vectors are [1.5, 3.2, -0.5], [2.0, 2.8, -0.3], and [1.7, 3.1, -0.4], respectively. The system calculates the cross-correlation distances between these vectors; for example, the distance between temperature and current is 0.6, the distance between temperature and vibration is 0.4, and the distance between current and vibration is 0.5. Then, the average of these three distances (0.6 + 0.4 + 0.5) / 3 = 0.5 is taken as the cross-dimensional cooperative anomaly coefficient of the dynamic system's associated unit group.

[0022] In some embodiments, the diagnostic dimension trigger signal includes a first diagnostic trigger signal and a second diagnostic trigger signal; Based on the number of anomalous correlation unit groups and the evolution characteristics of the cross-dimensional collaborative anomaly coefficients of all anomalous correlation unit groups, the diagnostic dimension trigger signals for the submersible are determined, including: The average value of the cross-dimensional collaborative anomaly coefficients of all abnormally correlated unit groups is normalized to obtain the collaborative imbalance coefficient. Record the duration for which the coordination imbalance coefficient exceeds the first threshold; If the number of abnormal associated unit groups exceeds the second threshold and the duration exceeds the third threshold, a second diagnostic trigger signal matching the submersible is generated; otherwise, a first diagnostic trigger signal matching each abnormal associated unit group is generated.

[0023] It should be noted that all anomaly correlation unit groups will have a cross-dimensional collaborative anomaly coefficient. This coefficient is calculated beforehand and is used to measure the strength of the anomaly correlation between different monitoring dimensions. The system needs to normalize these collaborative anomaly coefficients to obtain an index called the collaborative imbalance coefficient. The purpose of normalization is to eliminate scale differences between different unit groups, making the data more comparable. Normalization usually involves adjusting each anomaly coefficient to a uniform scale according to certain rules. For example, suppose there are three anomaly correlation unit groups with the following cross-dimensional collaborative anomaly coefficients: [2.0, 3.5, 1.0]; after normalization... After normalization, these values ​​may be adjusted to [0.4, 0.7, 0.2]. These normalized values ​​are the coordination imbalance coefficients. If the value of the normalized coordination imbalance coefficient exceeds a preset first threshold for a certain period, the system will record the duration of this state. This identifies whether the submersible has a persistent abnormal state, i.e., whether it exhibits abnormal imbalance phenomena that are interconnected between multiple monitoring dimensions over a relatively long period of time. Specifically, the system tracks the changes in the cross-dimensional coordination anomaly coefficient of each associated unit group over time. For example, suppose the coordination imbalance coefficient of a certain unit continuously takes values ​​of [0.3, 0.4, 0.7, 0.2] over a period of time. [0.5, 0.6, 0.9, 0.7], while the preset first-level threshold is 0.6; the system will identify time periods where the coefficient exceeds 0.6 and record the duration of these abnormal states; if the coefficient continues to exceed the threshold for a preset length (e.g., 3 minutes), it indicates that the unit group has entered a stable collaborative abnormal state, thereby triggering the subsequent diagnostic process; in this process, the system mainly relies on two key factors to comprehensively judge and decide which diagnostic signal to trigger: one is the number of associated unit groups with abnormalities, that is, the total number of groups whose collaborative abnormal coefficients exceed the threshold among all unit groups; the other is the duration of these abnormal states; the system will determine the diagnostic signal based on the abnormality The severity of the overall imbalance is assessed by combining the affected area (number of unit groups) with the duration of the imbalance, and different levels of diagnostic and analysis procedures are initiated accordingly. If the number of anomalously associated unit groups exceeds a second threshold and the duration of the state exceeds a third threshold, the system generates a more severe diagnostic signal, namely the second diagnostic trigger signal. If these conditions are not met, the system generates a more basic first diagnostic trigger signal for each anomalously associated unit group. Example: Suppose that the monitoring system of a submersible detects four anomalously associated unit groups at a certain moment, of which the co-abnormal coefficient of two unit groups exceeds 0.6. If the number of abnormal unit groups exceeds 3 minutes, and the second threshold is 3 unit groups and the third threshold is 2 minutes, then the system will generate a second diagnostic trigger signal because the number of abnormal unit groups exceeds the second threshold and the duration exceeds the third threshold. If the number of abnormal unit groups does not exceed the second threshold, or the duration does not exceed the third threshold, then the system will generate a first diagnostic trigger signal for each abnormal unit group. For example, if two unit groups have a co-abnormality coefficient exceeding 0.6, and the duration of each abnormal unit group is 1 minute, the system will generate a first diagnostic trigger signal for each of these two unit groups.

[0024] In some embodiments, the target diagnostic scheduling signal includes a fault isolation scheduling signal and a path calibration scheduling signal; Based on the causal dependency characteristics of historical operational data from all monitoring units of the submersible and the cross-dimensional collaborative anomaly coefficients of anomalous associated unit groups, target diagnostic scheduling signals are determined, including: Based on the cross-dimensional collaborative anomaly coefficient of each anomaly-related unit group, the fault isolation scheduling signal of the corresponding anomaly-related unit group is determined. The mutual information causal analysis method was used to analyze the dependency relationship of the historical operation data of the remaining monitoring units, and a causal dependency matrix was constructed based on the analysis results. A weighted diagnostic model is adopted, which combines the current diagnostic path structure and causal dependency matrix of the submersible to generate multiple candidate diagnostic scheduling signals; For each candidate diagnostic scheduling signal, a fault scenario is simulated, and the target diagnostic scheduling signal is determined based on the diagnostic accuracy of the simulation results.

[0025] It should be noted that after analyzing the cross-dimensional collaborative anomaly coefficient of each abnormally related unit group, the system will determine the corresponding fault isolation scheduling signal for each unit group. The core purpose of this step is to identify those parts that are most directly related to potential faults of the submersible from all abnormal unit groups, and generate targeted scheduling instructions based on the intensity of their abnormal performance and the collaborative characteristics between dimensions. For example, suppose the system detects that the collaborative anomaly coefficient of a certain abnormally related unit group is as high as 0.75, which clearly shows that there is a strong abnormal correlation between multiple monitoring dimensions within the group. Based on this analysis result, the system will generate a clearly targeted fault isolation scheduling signal. This signal will indicate that the subsequent fault location process should focus on this unit group first, thereby helping to quickly locate and isolate the root cause of the fault. A thorough dependency analysis is conducted on the historical operational data of the remaining normal monitoring units on the submersible. The method employed in this stage is mutual information causal analysis. This method effectively reveals the implicit correlations and causal dependencies between different monitoring data sequences by calculating and analyzing the mutual information, thereby depicting the internal influence network between each monitoring unit. This method aims to help the system deeply understand the complex mutual influence relationships between different monitoring units on the submersible. For example, in actual operation, there may be significant causal links between the readings of temperature and pressure sensors; when the temperature inside the cabin or equipment rises, the pressure sensor values ​​may also change regularly, reflecting the state changes inside and outside the submersible. Mutual information causal analysis can effectively identify and quantify such potential causal associations, enabling the system to use these dependencies as important reasoning bases in subsequent diagnostic analysis. Based on the results of causal analysis, the system constructs a causal dependency matrix. This matrix systematically records the causal relationships between each monitoring unit and the strength of their interdependence, forming a structured map describing the internal influence relationships of the monitoring network. This matrix will become one of the key bases for generating diagnostic scheduling signals in the future. The system employs a weighted diagnostic model. This model combines the submersible's current real-time diagnostic path structure with the aforementioned causal dependency matrix to generate multiple possible candidate diagnostic scheduling signals. The weighted diagnostic model comprehensively considers the dependency strength between different monitoring units, the current abnormal state, and the overall system operating context, thereby deduceing multiple reasonable diagnostic paths and their corresponding scheduling instructions. For example, based on the causal dependency matrix and the current state, the system may generate the following candidate scheduling signals: Candidate Signal 1: When the temperature sensor detects an anomaly, prioritize scheduling the diagnostic program for the pressure sensor; Candidate Signal 2: If the vibration sensor malfunctions, simultaneously schedule the diagnostics for both the temperature and pressure sensors; Candidate Signal 3: Based on the anomalies detected by the vibration and pressure sensors... The system considers different fault scenarios and proposes possible diagnostic paths. After generating candidate diagnostic scheduling signals, the system performs fault scenario simulations. By simulating the impact of different fault situations, the system can evaluate the effectiveness of each candidate diagnostic scheduling signal in actual operation. The simulation results help the system select the scheduling signal that can most accurately diagnose the fault, i.e., the target diagnostic scheduling signal. For example, suppose that in a simulated fault scenario, candidate 1 can accurately locate the submersible fault and solve the problem, while candidates 2 and 3 have lower accuracy. In this case, the system will select candidate 1 as the final target diagnostic scheduling signal.

[0026] In some embodiments, a fault propagation feature vector is determined based on the temporal change characteristics and compartment association characteristics of real-time sensing data, including: A rolling window is used to perform time-series analysis on the real-time sensing data of each sensing node to determine the signal time-series transition vector of each sensing node; feature aggregation is performed on all signal time-series transition vectors that fit the diagnostic path to obtain the time-series transmission feature vector. The real-time sensing data of all sensing nodes of the adaptive diagnostic path at the same time node are analyzed for differences to obtain the spatial transmission feature vector; the spatial transmission feature vector is used to indicate the differences in the distribution of real-time sensing data of sensing nodes in different compartments. The temporal and spatial transmission feature vectors are recombined to obtain fault transmission feature vectors that are adapted to the diagnostic path.

[0027] It should be noted that the temporal transition feature vector is obtained by performing temporal analysis on the real-time sensing data of each sensing node. For this purpose, the system employs a rolling window method. A rolling window means that the system divides the real-time data into multiple small intervals, for example, every few seconds or minutes, and analyzes the data change trend within each interval. These data change trends will form the temporal transition vector of that node. For example, suppose one sensing node of a submersible is a temperature sensor. Using the rolling window method, the system collects temperature data and analyzes its temporal changes. Suppose that in the past 30 seconds, the temperature sensor data changes as follows: Second 1: 22°C; Second 2: 22.1°C; Second 3: 22.3°C; ...; Second 30: 23°C. This temperature change process will generate a temporal transition vector representing the trend of temperature change. Possible features include the temperature fluctuation range, acceleration or deceleration, etc. Each sensing node will generate a similar temporal transition vector. Feature aggregation is performed on the signal time-series transition vectors along all adapted diagnostic paths. By aggregating these time-series transition vectors, a more comprehensive time-series transmission feature vector can be obtained, which can describe the time-series relationships and interactions of different sensing nodes of the submersible. For example, if the submersible has multiple monitoring units, such as temperature sensors, pressure sensors, and vibration sensors, the system will aggregate their time-series transition vectors. Suppose the time-series vector of the temperature sensor shows that the temperature is gradually increasing, the pressure sensor shows pressure fluctuations, and the vibration sensor shows no significant change. By aggregating this information, the system may derive a comprehensive feature that shows the changes in the submersible's state and provides a basis for fault diagnosis. The system performs differential analysis on the real-time sensing data of all sensing nodes in the adapted diagnostic path at the same time point; that is, by comparing the data of different sensing nodes at the same moment, a spatial transmission feature vector is obtained. This vector is mainly used to describe the distribution differences of sensing node data in different compartments or areas. For example, suppose the submersible has two compartments, one is a temperature compartment and the other is a pressure compartment. At the same time point, the sensor in the temperature compartment shows a temperature of 20°C, while the sensor in the pressure compartment shows a pressure of 10 bar. The system analyzes these data differences to obtain a spatial transmission feature vector, indicating the data differences between different compartments. This vector reflects whether the state changes between compartments are consistent, which can help diagnose whether there are certain faults or abnormal phenomena in the submersible. The temporal and spatial transmission feature vectors are recombined to obtain a new comprehensive feature vector—the fault transmission feature vector. This feature vector integrates temporal and spatial data, more accurately reflecting the overall state of the submersible and providing more basis for fault diagnosis. For example, suppose the temporal transmission feature vector shows that temperature and pressure are changing, while the spatial transmission feature vector shows that there are significant differences in data between the two compartments. After feature recombination, the system may obtain a comprehensive fault transmission feature vector indicating that temperature changes may have caused pressure fluctuations, and this phenomenon is manifested in different compartments. This vector will guide the subsequent fault diagnosis path and help the system locate the possible types of faults in the submersible.

[0028] In some embodiments, a scrolling window is used to perform time-series analysis on the real-time sensing data of each sensing node to determine the signal time-series transition vector of each sensing node, including: The first scrolling window is used to extract transient features from the real-time sensing data to obtain the transient transition vector of each sensing node; A second scrolling window is used to extract trend features from the real-time sensing data to obtain the trend change vector of each sensing node; The duration of the first scrolling window is greater than the duration of the second scrolling window; The transient transition vector and trend transition vector of each sensing node are integrated to obtain the signal time-series transition vector of each sensing node.

[0029] It's important to note that the rolling window is a progressively sliding analysis method that divides real-time sensing data into several small segments for processing. In this step, the system uses a first rolling window, which is relatively short, to extract transient features of the data. Transient features focus on short-term fluctuations or rapid changes in the data. For example, suppose the pressure sensor on a submersible collects the following data over a certain period: Second 1: 100 bar; Second 2: 100.5 bar; Second 3: 101 bar; Second 4: 100.8 bar; Second 5: 101.2 bar. Using the first rolling window, for example, with a window size of 3 seconds, the system extracts the pressure data fluctuations within this time period. For instance, if there are significant pressure fluctuations between the first and third seconds, the system can calculate the transient features of this data segment, such as the maximum fluctuation amplitude and fluctuation rate. Through this rolling window, the system obtains a transient transition vector, which contains the rapid changes in pressure data during this period, reflecting whether there are short-term disturbances or anomalies in the sensor. A second scrolling window, longer than the first, is used to extract trend features from the data. Trend features focus on the gradual change in data over a longer time period, such as rising, falling, or stabilizing data. For example, continuing with the pressure sensor example, over a longer time period, such as the past 30 seconds, the system collected the following data: Second 1: 100 bar; Second 2: 100.1 bar; Second 3: 100.3 bar; ...; Second 30: 101 bar. Using a second scrolling window, for example, with a window size of 10 seconds, the system analyzes the trend of the data over a longer time period. For example, within the window from second 1 to second 10, the pressure data steadily increases. The system calculates trend features, such as the rate of increase and the magnitude of increase. This trend transition vector reflects the long-term change pattern of the sensor data, helping to identify whether the submersible is experiencing a gradual performance decline or other long-term changes. The transient and trend transition vectors of each sensing node are integrated. The resulting vector is a signal time-series transition vector that encompasses both short-term fluctuations and long-term trends, comprehensively reflecting the state changes of the sensing nodes. For example, suppose the transient transition vector of a pressure sensor contains the amplitude and rate of pressure fluctuations over the past few seconds, while the trend transition vector describes the gradual upward trend of pressure over the past 30 seconds. After merging these two vectors, the resulting signal time-series transition vector might contain the following information: Transient characteristics: the range and frequency of rapid pressure fluctuations, e.g., ±0.5 bar fluctuations; Trend characteristics: the average rate of pressure rise over 30 seconds, e.g., 0.1 bar / second. Through this integrated vector, the system can not only understand the long-term trend of pressure changes but also analyze the existence of rapid short-term fluctuations, which is crucial for fault diagnosis.

[0030] In some embodiments, the target accuracy adjustment parameters for the adapted diagnostic path are determined based on the multi-dimensional similarity comparison results of the fault propagation feature vectors, including: The fault propagation feature vector is compared with a pre-set fault propagation pattern library in multiple dimensions to determine the matching fault propagation pattern. The fault propagation mode library includes steady-state propagation modes, cascaded fault propagation modes, and latent fault propagation modes. Based on the accuracy adjustment rules corresponding to the matched fault propagation mode, determine the target accuracy adjustment parameters for the adaptive diagnostic path.

[0031] It should be noted that the real-time acquired fault propagation feature vector is compared with a pre-defined fault propagation mode library. The fault propagation feature vector is characteristic data describing how a fault propagates through various components or sensors when a fault occurs in the system. It includes information such as the time, magnitude, and propagation path of the fault. The fault propagation mode library, on the other hand, is predefined and contains different types of fault modes, each representing a specific fault propagation method. The fault propagation mode library typically includes the following types of modes: Steady-state propagation mode: The fault gradually and steadily propagates in the system, and the propagation process is relatively smooth. It is usually a long-term, stable fault. Cascaded fault propagation mode: The failure of one component triggers the failure of another component, leading to a series of chain reactions. The propagation path is more complex, and the impact of the fault gradually amplifies. Latent fault propagation mode: Fault propagation does not show obvious signs in the short term, only suddenly appearing when a certain critical point is reached, and is usually difficult to detect in the early stages. Through multi-dimensional similarity comparison, the system calculates the similarity of each fault mode according to each dimension, and then selects the fault mode that best matches the real-time fault propagation feature vector. Example: Suppose the system detects a fluctuation in the output of a sensor, and this fluctuation lasts for several hours. The system finds through comparison that this fluctuation mode best matches the steady-state propagation mode because the fluctuation is stable and has no obvious cascading effect; therefore, the system will match the steady-state propagation mode. Based on the results of multi-dimensional similarity comparison, the system determines the best-matching fault propagation mode. For example, if the propagation mode of a real-time fault is characterized by a failure in one component triggering failures in other components, and this fault has a gradually amplifying trend, then the system may match a cascading fault propagation mode. Example: Suppose that the motor in the system first experiences a current overload fault, which then triggers a battery overheating fault, eventually leading to the shutdown of the entire power system. Through multi-dimensional comparison, the system will match a cascading fault propagation mode. Each fault propagation mode corresponds to a different accuracy adjustment rule. These rules are used to adjust the target accuracy of the diagnostic path to ensure that the system can provide the best fault diagnosis results according to the matched mode. The adjustment of the accuracy adjustment parameters depends on the characteristics of the fault mode. For example, some fault modes may require higher diagnostic accuracy, while other modes can appropriately reduce accuracy to improve diagnostic efficiency. For example, if the system matches a steady-state propagation mode, since the fault propagation is relatively smooth and stable, the system can choose a lower accuracy adjustment parameter to save computing resources. If the system matches a cascading fault propagation mode, the system needs higher accuracy to identify the interrelationships and propagation paths of each fault, so the accuracy adjustment parameter will be higher. If the system matches a latent fault propagation mode, since the fault manifestation is delayed and difficult to detect, the system may need more sensitive accuracy adjustment in order to detect latent faults as early as possible.

[0032] In some embodiments, the training method for a spatiotemporal causal analysis network includes: The initial causal analysis network was trained using a submersible failure simulation dataset. The parameters of the cross-dimensional correlation layer and feature quantization layer in the network were iteratively optimized to obtain the spatiotemporal causal analysis network. The submersible fault simulation dataset includes propulsion system faults, buoyancy regulation faults, sealing system faults, and navigation system faults.

[0033] It should be noted that the training is based on an initial causal reasoning network; this network may be a preliminary model that already possesses some basic causal reasoning capabilities, but further optimization is still needed; the training data comes from a submersible failure simulation dataset, which includes various types of failure simulations, specifically propulsion system failures, buoyancy regulation failures, sealing system failures, and navigation system failures; each failure includes a series of spatiotemporal data describing the changes in various components and functions of the submersible system when a failure occurs; this data provides the network with useful information about the occurrence, propagation, and impact of failures; Propulsion system failure: Describes the changes in various parameters when the submersible's propulsion system fails, including a decrease in thrust and changes in speed; Buoyancy adjustment failure: Records the changes in the submersible's ascent and descent speeds when buoyancy adjustment fails, as well as the failure of the buoyancy system; Sealing system failure: Records the process of water leakage in the sealing system and water ingress into the submersible, causing the system to be unable to maintain underwater stability; Navigation system failure: Records the failure of the submersible's navigation control, leading to abnormal positioning and path tracking; These data will be used as training samples for the model. The initial causal analysis network will learn how to extract the causal relationships between various systems from these failure simulation data and determine the correlations between different failures; Causal analysis networks are typically constructed using a multi-layered structure, with the cross-dimensional correlation layer and the feature quantization layer being the two core components. The cross-dimensional correlation layer is responsible for identifying and learning the complex relationships between different data dimensions. In the context of submersible fault diagnosis, these dimensions encompass the readings of various sensors and their evolution trends over time. The main purpose of this layer is to uncover the causal relationships between different fault types and system components at different temporal and spatial scales. For example, a failure in the propulsion system may trigger a chain reaction in the buoyancy control system; anomalies in the navigation system may also occur simultaneously with conditions such as seal failure. Through this correlation analysis, the network can gradually understand how various faults in the system influence and propagate each other. The feature quantization layer focuses on preprocessing the input data. It effectively compresses and quantizes the original high-dimensional data, transforming it into a low-dimensional feature representation that is easier for the network to process and reason about. This process not only reduces computational complexity but also allows the model to focus on the most critical diagnostic information. For example, after processing the massive amount of time-series fault data of a submersible, it may be compressed into several feature vectors that characterize the severity or pattern of the fault, thus providing efficient and clear input for subsequent causal analysis. The entire network undergoes multiple rounds of iterative training to continuously optimize the internal parameters of these two layers. The core objective of the training is to enable the network to understand and infer the causal relationships between various failure modes more and more accurately, thereby continuously improving its prediction accuracy.

[0034] After thorough training and parameter optimization, the resulting spatiotemporal causal analysis network possesses powerful causal reasoning capabilities. It can not only process multi-source fault data from submersibles but also integrate temporal and spatial information for in-depth analysis. This network can effectively reveal the intricate causal relationships between various subsystems of the submersible and provide reliable support for accurate prediction and maintenance decisions when faults occur.

[0035] In some embodiments, fault temporal causal chain inference is performed based on processed real-time sensing data to obtain the complete temporal causal chain of the submersible fault, including: Acquire fault response event data corresponding to the processed real-time sensing data; Key fault data is selected by matching and filtering real-time sensing data and fault response event data, and an undirected sensing association graph is constructed based on the key fault data. In the undirected perception association graph, the nodes are key fault data, and the edges are the source unit information and target unit information in the corresponding fault response event data. The source unit information includes the perception unit number and the installation section, and the target unit information includes the associated unit number and the connecting pipeline. The nodes in the undirected perceptual association graph are aggregated to obtain the aggregated feature results. Density clustering is then performed on the aggregated feature results to obtain the core fault data in the critical fault data. A directed fault transmission graph is generated based on the fault response event data corresponding to the core fault data. The nodes in the directed fault transmission graph are the sensing unit identifiers of the fault response event data, and the edges in the directed fault transmission graph are directed edges from the source unit to the target unit. The attributes of the edges include the corresponding fault type and the occurrence timestamp. Traverse all nodes in the directed fault propagation graph, select all nodes with an in-degree of 0 as the starting point of the initial causal chain, and generate all temporal causal chains based on the starting point and the directed edges.

[0036] It should be noted that during the operation of the submersible, the sensing system monitors various parameters in real time, such as temperature, pressure, and propulsion, and transmits this data to the central processing system. When the submersible malfunctions, the system records relevant fault response event data, including information such as fault type, time of occurrence, fault source, and target. For example, when the submersible's propulsion system malfunctions, the sensing system may record real-time data on the decrease in propulsion, while the fault response event data may record the reactions of other systems caused by the decrease in propulsion, such as changes in the buoyancy control system or errors in the navigation system. After acquiring real-time sensing data and fault response event data, the system determines which data is critical to fault analysis through matching and filtering. For example, in the context of a propulsion system fault, propulsion force data, buoyancy control system data, and navigation system data may all be critical fault data. Based on this critical data, the system constructs an undirected sensing association graph, where nodes represent critical fault data and edges represent the relationships between different fault response events. For example, there is a certain spatiotemporal correlation between propulsion system fault data and buoyancy control fault data, which is represented as an edge in the graph. When constructing this graph, the edge attributes include: source unit information: for example, the source unit is a sensor of the propulsion system, sensing unit numbered "P1", installed in the "propulsion section" of the submersible; target unit information: for example, the target unit is a sensor of the buoyancy control system, sensing unit numbered "F1", installed in the "buoyancy section" of the submersible, and connected to the propulsion system through a pipeline. Once the undirected perceptual relationship graph is constructed, the next step is to aggregate the nodes in the graph. This aggregation process will help extract key fault data from a larger dataset and identify those data that play a core role in fault propagation. After aggregation, some aggregation feature results will be obtained, which reflect the strength and importance of the relationship between different fault data. For example, some nodes may be key fault sources, which are strongly associated with other nodes, while other nodes may play a secondary role in the fault propagation process. Density clustering is a method of grouping data points with the aim of clustering similar data together based on their close relationships. By performing density clustering on the aggregated feature results, the system can identify core failure data that plays a central role in the temporal causal chain. For example, a propulsion system failure may be a core node in the entire failure chain, as it is strongly correlated with failures in multiple other systems and is considered core failure data. Other failures that are less correlated with the core node may be considered non-core failure data. After identifying the core fault data, the next step is to construct a directed fault propagation graph. Nodes in this graph represent data from various fault response events, such as sensing unit identifiers, while edges represent the propagation process of the fault from the source unit to the target unit. The edges of the directed graph have the following attributes: Source unit information: the identifier and related information of the fault source unit; for example, if a fault source is the propulsion system, its sensing unit number is "P1"; Target unit information: the target unit to which the fault propagates; for example, a propulsion system fault may affect the buoyancy control system, its sensing unit number is "F1"; Fault type: such as propulsion system fault, buoyancy control fault, etc.; Occurrence timestamp: the specific time the fault occurred. Through this information, the system can trace the fault propagation path and identify the temporal and causal relationships of various fault events within the submersible. Having constructed a directed fault propagation graph, the next step is to analyze the nodes in the graph. Nodes with an in-degree of 0 represent faults that have not been propagated from other nodes and are the starting point of the causal chain. For example, in a directed graph, a failure in the propulsion system may be the first failure event to occur, and failures in other systems are caused by it. Therefore, the node with the failure in the propulsion system may be the starting point with an in-degree of 0. By traversing the directed fault propagation graph, the system can generate a complete temporal causal chain. This causal chain describes the entire process from the source of the fault, through a series of fault propagations, ultimately leading to the submersible's failure. For example, assuming the propulsion system failure is the starting point of the causal chain, the temporal causal chain might be as follows: Propulsion system failure, for example: initial failure, source unit is the propulsion system, target unit is the buoyancy control system; Buoyancy control system failure, for example: triggered by the propulsion system, target unit is the navigation system; Navigation system failure, for example: triggered by the buoyancy control system, ultimately leading to the submersible losing control. In this way, the temporal sequence and causal relationship of the entire failure are clearly defined, which is helpful for subsequent fault diagnosis and handling.

[0037] Reference Figure 2 The second embodiment of the present invention provides a time-series causal chain diagnosis system for submersible faults, comprising: Anomaly correlation module 1 is used to acquire multi-dimensional operation monitoring data during the operation of the submersible; a preset spatiotemporal causal analysis network is used to perform anomaly correlation analysis on the multi-dimensional operation monitoring data to obtain the cross-dimensional collaborative anomaly coefficient of the correlation unit group under each monitoring dimension; Diagnostic trigger module 2 is used to perform threshold comparison on all cross-dimensional collaborative anomaly coefficients to determine each anomaly association unit group; and to determine the diagnostic dimension trigger signal of the submersible based on the number of anomaly association unit groups and the evolution characteristics of the cross-dimensional collaborative anomaly coefficients of all anomaly association unit groups. The path adjustment module 3 is used to respond to the diagnostic dimension trigger signal, determine the target diagnostic scheduling signal based on the causal dependency characteristics of the historical operating data of all monitoring units of the submersible and the cross-dimensional collaborative anomaly coefficient of the abnormal associated unit group, and adapt and adjust the fault diagnosis path of the submersible with the target diagnostic scheduling signal to obtain the adapted diagnostic path. The precision adjustment module 4 is used to acquire real-time sensing data of all sensing nodes of the adaptive diagnostic path, determine the fault transmission feature vector based on the temporal change characteristics and compartment association characteristics of the real-time sensing data, and determine the target precision adjustment parameters of the adaptive diagnostic path based on the multi-dimensional similarity comparison results of the fault transmission feature vector. The causal reasoning module 5 is used to perform interference suppression processing on real-time sensing data by adjusting parameters with target accuracy, and to perform fault temporal causal chain reasoning based on the processed real-time sensing data to obtain the complete temporal causal chain of the submersible fault.

[0038] It should be noted that the submersible fault temporal causal chain diagnosis system provided in this embodiment of the invention is used to execute all the process steps of the submersible fault temporal causal chain diagnosis method in the above embodiment. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0039] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0040] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0041] The embodiments, implementation methods, and related technical features of this application can be combined and substituted for each other without conflict.

[0042] The above are merely preferred embodiments of this application and are not intended to limit this application in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this application without departing from the scope of the technical solution of this application shall still fall within the scope of the technical solution of this application.

Claims

1. A method for diagnosing the temporal causal chain of submersible malfunctions, characterized in that, include: Acquire multi-dimensional operational monitoring data during the submersible's operation; Anomaly correlation analysis is performed on the multi-dimensional operation monitoring data to obtain the cross-dimensional collaborative anomaly coefficient of the associated unit group under each monitoring dimension; Threshold comparison is performed on all the cross-dimensional collaborative anomaly coefficients to determine each anomaly association unit group; based on the number of the anomaly association unit groups and the evolution characteristics of the cross-dimensional collaborative anomaly coefficients of all the anomaly association unit groups, the diagnostic dimension trigger signal of the submersible is determined. In response to the diagnostic dimension trigger signal, a target diagnostic scheduling signal is determined based on the causal dependency characteristics of the historical operating data of all monitoring units of the submersible and the cross-dimensional collaborative anomaly coefficient of the abnormal associated unit group; the fault diagnosis path of the submersible is adapted and adjusted using the target diagnostic scheduling signal to obtain an adapted diagnostic path; Real-time sensing data of all sensing nodes of the adaptive diagnostic path is acquired. Based on the temporal change characteristics and compartment association characteristics of the real-time sensing data, a fault transmission feature vector is determined. Based on the multi-dimensional similarity comparison results of the fault transmission feature vector, the target accuracy adjustment parameters of the adaptive diagnostic path are determined. The real-time sensing data is subjected to interference suppression processing using the target accuracy adjustment parameters. Based on the processed real-time sensing data, fault temporal causal chain inference is performed to obtain the complete temporal causal chain of the submersible fault.

2. The method according to claim 1, characterized in that, The cross-dimensional collaborative anomaly coefficient is used to indicate the degree of fault correlation between units in different monitoring dimensions; Anomaly correlation analysis is performed on the multi-dimensional operational monitoring data to obtain the cross-dimensional collaborative anomaly coefficient of the correlated unit group under each monitoring dimension, including: The historical data of each monitoring dimension in the multi-dimensional operational monitoring data are subjected to periodic feature quantization processing to obtain the dimensional feature vector of each monitoring dimension; wherein, the periodic feature quantization value includes at least one of waveform factor, kurtosis factor and skewness factor; The monitoring unit is divided into associated unit groups based on the structural correlation map of the submersible; Calculate the cross-correlation distance between the dimensional feature vectors of different monitoring dimensions within each associated unit group; use the mean of the cross-correlation distances as the cross-dimensional collaborative anomaly coefficient of the corresponding associated unit group.

3. The method according to claim 2, characterized in that, The diagnostic dimension trigger signal includes a first diagnostic trigger signal and a second diagnostic trigger signal; The step of determining the diagnostic dimension trigger signal for the submersible based on the number of the abnormal correlation unit groups and the evolution characteristics of the cross-dimensional collaborative anomaly coefficients of all the abnormal correlation unit groups includes: The average value of the cross-dimensional collaborative anomaly coefficients of all the aforementioned abnormal correlation unit groups is normalized to obtain the collaborative imbalance coefficient. Record the duration for which the cooperative imbalance coefficient exceeds the first threshold; If the number of abnormal associated unit groups exceeds the second threshold and the duration exceeds the third threshold, a second diagnostic trigger signal matching the submersible is generated; otherwise, a first diagnostic trigger signal matching each of the abnormal associated unit groups is generated.

4. The method according to claim 3, characterized in that, The target diagnostic scheduling signal includes a fault isolation scheduling signal and a path calibration scheduling signal; The determination of the target diagnostic scheduling signal based on the causal dependency characteristics of historical operational data of all monitoring units of the submersible and the cross-dimensional collaborative anomaly coefficient of the abnormal correlation unit group includes: Based on the cross-dimensional collaborative anomaly coefficient of each of the anomaly association unit groups, determine the fault isolation scheduling signal for the corresponding anomaly association unit group. Dependency analysis was performed on the historical operational data of the remaining monitoring units, and a causal dependency matrix was constructed based on the analysis results. Based on the current diagnostic path structure of the submersible and the causal dependency matrix, multiple candidate diagnostic scheduling signals are generated; For each candidate diagnostic scheduling signal, a fault scenario is simulated, and the target diagnostic scheduling signal is determined based on the diagnostic accuracy of the simulation results.

5. The method according to claim 4, characterized in that, The determination of the fault propagation feature vector based on the temporal change characteristics and compartment association characteristics of the real-time sensing data includes: A rolling window is used to perform time-series analysis on the real-time sensing data of each sensing node to determine the signal time-series transition vector of each sensing node; feature aggregation is performed on all signal time-series transition vectors of the adaptive diagnostic path to obtain the time-series transmission feature vector. The real-time sensing data of all sensing nodes of the adaptive diagnostic path at the same time node are analyzed for differences to obtain a spatial transmission feature vector; wherein, the spatial transmission feature vector is used to indicate the differences in the distribution of real-time sensing data of sensing nodes in different compartments. The temporal transmission feature vector and the spatial transmission feature vector are subjected to feature recombination processing to obtain the fault transmission feature vector of the adapted diagnostic path.

6. The method according to claim 5, characterized in that, The step of using a rolling window to perform time-series analysis on the real-time sensing data of each sensing node to determine the signal time-series transition vector of each sensing node includes: The first scrolling window is used to extract transient features from the real-time sensing data to obtain the transient transition vector of each sensing node; A second scrolling window is used to extract trend features from the real-time sensing data to obtain the trend change vector of each sensing node; Wherein, the duration of the first scrolling window is greater than the duration of the second scrolling window; The transient transition vector and the trend transition vector of each sensing node are processed by vector integration to obtain the signal time-series transition vector of each sensing node.

7. The method according to claim 6, characterized in that, The step of determining the target accuracy adjustment parameters of the adapted diagnostic path based on the multi-dimensional similarity comparison results of the fault propagation feature vector includes: The fault propagation feature vector is compared with a preset fault propagation pattern library in multiple dimensions to determine the matching fault propagation pattern. The fault propagation mode library includes steady-state propagation modes, cascaded fault propagation modes, and latent fault propagation modes. Based on the accuracy adjustment rules corresponding to the matched fault propagation mode, the target accuracy adjustment parameters of the adaptive diagnostic path are determined.

8. The method according to claim 7, characterized in that, The fault temporal causal chain inference is performed based on the processed real-time sensing data to obtain the complete temporal causal chain of the submersible fault, including: Acquire fault response event data corresponding to the processed real-time sensing data; Based on the matching and filtering of the real-time sensing data and the fault response event data, key fault data is selected, and an undirected sensing association graph is constructed based on the key fault data. In the undirected sensing association graph, the nodes are the key fault data, and the edges are the source unit information and target unit information corresponding to the fault response event data. The source unit information includes the sensing unit number and the installation section, and the target unit information includes the associated unit number and the connecting pipeline.

9. The method according to claim 8, characterized in that, The step of performing fault temporal causal chain reasoning based on the processed real-time sensing data to obtain the complete temporal causal chain of the submersible fault also includes: The nodes in the undirected perceptual association graph are aggregated to obtain aggregated feature results. Density clustering is then performed on the aggregated feature results to obtain the core fault data in the critical fault data. A directed fault transmission graph is generated based on the fault response event data corresponding to the core fault data; wherein, the nodes in the directed fault transmission graph are the sensing unit identifiers of the fault response event data, and the edges in the directed fault transmission graph are directed edges from the source unit to the target unit, and the attributes of the edges include the corresponding fault type and occurrence timestamp. Traverse all nodes in the directed fault propagation graph, select all nodes with an in-degree of 0 as the starting point of the initial causal chain, and generate all temporal causal chains based on the starting point and the directed edges.

10. A time-series causal chain diagnostic system for submersible malfunctions, characterized in that, include: The anomaly correlation module is used to acquire multi-dimensional operational monitoring data during the operation of the submersible; Anomaly correlation analysis is performed on the multi-dimensional operation monitoring data to obtain the cross-dimensional collaborative anomaly coefficient of the associated unit group under each monitoring dimension; The diagnostic trigger module is used to perform threshold comparison on all the cross-dimensional collaborative anomaly coefficients to determine each anomaly association unit group; and to determine the diagnostic dimension trigger signal of the submersible based on the number of the anomaly association unit groups and the evolution characteristics of the cross-dimensional collaborative anomaly coefficients of all the anomaly association unit groups. The path adjustment module is used to respond to the diagnostic dimension trigger signal, determine the target diagnostic scheduling signal based on the causal dependency characteristics of the historical operating data of all monitoring units of the submersible and the cross-dimensional collaborative anomaly coefficient of the abnormal associated unit group; and adapt and adjust the fault diagnosis path of the submersible with the target diagnostic scheduling signal to obtain an adapted diagnostic path. The precision adjustment module is used to acquire real-time sensing data of all sensing nodes of the adaptive diagnostic path, determine the fault transmission feature vector based on the temporal change characteristics and compartment association characteristics of the real-time sensing data, and determine the target precision adjustment parameters of the adaptive diagnostic path based on the multi-dimensional similarity comparison results of the fault transmission feature vector. The causal reasoning module is used to perform interference suppression processing on the real-time sensing data by adjusting the parameters according to the target accuracy, and to perform fault temporal causal chain reasoning based on the processed real-time sensing data to obtain the complete temporal causal chain of the submersible fault.

Images