Method and system for early warning of risk of failure of station keeping capability of a vessel dynamic positioning system

Abstract

The ship power positioning system station-keeping ability failure risk early warning method and system belong to the technical field of ship and ocean engineering equipment safety monitoring, risk early warning and intelligent operation and maintenance. The present application can unify the modeling of basic faults, subsystem faults, thrust holding ability state, control closed loop ability state and station-keeping ability failure risk state, form a hierarchical imprecise probability graph model, generate imprecise probability parameters according to the data availability of different nodes, and automatically infer and update the imprecise probability graph model according to the current observable state data, output risk early warning results based on interval probability, and identify the key causes leading to the increase of station-keeping ability failure risk.

Description

Technical Field

[0001] This invention belongs to the field of safety monitoring, risk warning, and intelligent operation and maintenance technology for ships and marine engineering equipment, specifically relating to a method and system for early warning of the failure of the stationing capability of a ship's dynamic positioning system. More specifically, this invention relates to the fusion processing of historical fault data, causal fault co-occurrence data, operation exposure duration, online monitoring data, and alarm logs during the operation of a ship's dynamic positioning system, constructing a hierarchical imprecise probabilistic graphical model to characterize the correlation between basic faults, subsystem faults, capability degradation, and stationing capability failure, and realizing a method and system for early warning of stationing capability failure risk and identification of key causes based on interval probability inference results. Background Technology

[0002] Dynamic positioning systems are used to automatically maintain a vessel's position and heading stability under conditions of no anchor or weak mooring. They typically include a propulsion subsystem, a power supply and distribution electronics system, a control subsystem, and a position reference subsystem. Dynamic positioning systems are widely used in deepwater drilling, wellhead intervention, subsea facility installation and maintenance, floating production facility berthing, and offshore wind power operation and maintenance. For these operations, a decrease in the dynamic positioning system's ability to maintain a predetermined position and heading (i.e., a decrease in its stationary capability) directly affects the safety of subsea equipment, the stress state of the umbilical cable, operational continuity, and the safety of personnel and equipment.

[0003] Ship dynamic positioning systems are characterized by complex equipment, coupled subsystems, uncertain operating environments, and severe consequences of failure. Factors such as propulsion failure, power supply anomalies, control failures, interference with position reference signals, communication link anomalies, control power outages, sensor drift, and sudden changes in environmental loads can all lead to a decline in the dynamic positioning system's holding capability. When thrust holding capability or control closed-loop capability continues to degrade, the ship may enter a state of insufficient holding capability, critical instability, degraded operation, manual takeover, or even loss of positioning. Therefore, for ship dynamic positioning systems, simply conducting accident analysis after a loss of positioning event is insufficient to meet the requirements of safe operation and maintenance; it is also necessary to implement risk warning and cause identification during the stage of holding capability degradation.

[0004] Existing technologies similar to this invention primarily transform equipment failures, alarm information, subsystem states, or expert experience of a ship's dynamic positioning system into probabilistic or graphical models for system failure risk assessment. These methods can typically establish correlations between subsystems such as propulsion, power supply and distribution, control, and position reference, and top-level failure events; however, they still fall short in providing early warning of positioning capability failure risks.

[0005] First, existing technologies often directly link local equipment failures to top-level failure events, lacking a description of intermediate states during the degradation of positioning capability. For ship dynamic positioning systems, propulsion, power supply, control, or position reference failures do not necessarily lead to immediate disorientation; their impact typically manifests as a decrease in thrust holding capability or a reduction in control closed-loop capability. If the model lacks these capability state nodes, it becomes difficult to promptly identify the risk of positioning capability evolving from a normal state to a critical failure state.

[0006] Secondly, existing technologies typically use precise point-value probabilities to represent fault probabilities and conditional probabilities. However, dynamic positioning system faults are characterized by low frequency, high consequences, and scarce samples, and the exposure duration varies across different tasks. When fault counts are low or some fault combinations are unobserved, point-value probabilities are prone to over-precision, failing to reflect the uncertainties caused by small samples, differences in exposure duration, and missing data.

[0007] Furthermore, existing probabilistic graphical models typically require a complete conditional probability table when handling relationships between multiple parent nodes. However, dynamic positioning systems suffer from a wide variety of basic fault types and complex subsystem coupling relationships, making it difficult to obtain historical samples covering all combinations of parent node states in the engineering field. Relying entirely on manual assignment results in a highly subjective model; relying solely on historical samples is limited by sample scarcity and missing combinations. Therefore, existing technologies struggle to flexibly construct probabilistic parameters based on differences in node hierarchy and data availability.

[0008] Furthermore, existing technologies typically output single probability values ​​or single risk levels, making it difficult to simultaneously reflect both the level of risk and the degree of uncertainty. They also struggle to identify critical subsystem failures, abnormal capability status, or fundamental malfunctions that lead to increased risk when the risk of resident capability failure rises. Therefore, existing technologies are insufficient to provide adequate evidence for troubleshooting and maintenance.

[0009] Therefore, it is necessary to propose a method and system for early warning of the risk of stationing capability failure in ship dynamic positioning systems. Specifically, it should be able to uniformly model the basic faults, subsystem faults, thrust holding capability status, control closed-loop capability status, and stationing capability failure risk status, forming a hierarchical inaccurate probability graphical model; it should be able to generate inaccurate probability (i.e., interval probability) parameters according to the data availability of different nodes under conditions of small samples and incomplete data; it should be able to automatically infer and update the inaccurate probability graphical model based on the current observable state data; and it should be able to output risk warning results (including risk level and degree of uncertainty) based on interval probabilities, while identifying the key causes leading to increased stationing capability failure risk. Summary of the Invention

[0010] To address the aforementioned technical problems, this invention provides a method and system for early warning of the risk of failure of the stationing capability of a ship's dynamic positioning system.

[0011] The technical solution of this invention: Early warning methods for the failure of the stationing capability of a ship's dynamic positioning system, such as Figure 1 As shown, it includes the following steps: Step S1: Construction of a hierarchical inaccurate probabilistic graphical model.

[0012] A hierarchical inaccurate probabilistic graphical model for ship dynamic positioning systems is constructed, including a basic fault layer, a subsystem fault layer, a capability state layer, and a top event layer.

[0013] Furthermore, the hierarchical imprecise probabilistic graphical model is a Credal network. A Credal network is a graphical model that represents variable dependencies using directed acyclic graphs and uses interval probabilities or probability sets to represent prior and conditional probabilities. Compared to Bayesian networks, Credal networks can preserve and propagate uncertainties caused by small samples, missing combinations, and statistical errors during inference.

[0014] Furthermore, the nodes of the basic fault layer include basic fault nodes X1 to X15, namely: X1 Thruster unit and drive fault, X2 Thruster control system fault, X3 Thruster feedback signal fault, X4 Thruster hydraulic and auxiliary support fault, X5 Power station management or control equipment fault, X6 Power distribution and protection fault, X7 Engine component fault, X8 Fuel, air, cooling or lubrication auxiliary system fault, X9 Dynamic positioning control software fault, X10 Dynamic positioning control hardware or human-machine interface fault, X11 Uninterruptible power supply or control power supply fault, X12 Position reference interference fault, X1 The faults are categorized as follows: 3. Position reference software or configuration failure; 4. Position reference mechanical sensing or deployment failure; 5. Position reference communication or electrical hardware failure. The subsystem fault layer includes nodes M2 to M5, where M2 represents a propulsion subsystem failure, M3 represents a power supply electronics system failure, M4 represents a control subsystem failure, and M5 represents a position reference subsystem failure. The capability status layer includes nodes M6 and M7, where M6 represents a degraded thrust holding capability and M7 represents a degraded control closed-loop capability. The top event layer node is M1, indicating a failure of the dynamic positioning system's stationary capability.

[0015] Furthermore, the connection relationships between the nodes include at least the following: X1, X2, X3, and X4 are connected to M2; X5, X6, X7, and X8 are connected to M3; X9, X10, and X11 are connected to M4; X12, X13, X14, and X15 are connected to M5; M2 and M3 are connected to M6; M4 and M5 are connected to M7; and M6 and M7 are connected to M1, forming a hierarchical risk evolution structure composed of basic faults, subsystem faults, capability status, and dynamic positioning system stationary capability failure, i.e., a hierarchical imprecise probabilistic graphical model, such as... Figure 3 As shown.

[0016] Furthermore, in the hierarchical imprecise probabilistic graphical model, a node equal to 1 indicates that the corresponding event has occurred, and a node equal to 0 indicates that the corresponding event has not occurred.

[0017] Step S2, Data Collection and Processing.

[0018] Collect and organize historical fault event data and current observable status data of the ship's dynamic positioning system, perform time alignment, merge duplicate records, remove abnormal records, and merge fault categories to form a dataset that can be used for inference calculations.

[0019] Furthermore, the dataset includes four categories: The first category is basic fault node parameter data, corresponding to basic fault nodes X1 to X15, including the fault count of each basic fault event within the statistical period and the corresponding job exposure duration, which is used to construct the prior probability interval of the basic fault nodes in step S3.

[0020] The second category is subsystem fault node parameter data, corresponding to subsystem fault nodes M2 to M5, including the link probability interval and leakage probability interval from each parent node to the subsystem fault node, which is used to determine the conditional probability interval of the subsystem fault node in step S4.

[0021] The third category is the co-occurrence sample data of parent and child nodes, corresponding to capability state nodes M6, M7 and top event node M1. It includes the parent node state combination within the statistical period, the co-occurrence count between the target node and the parent node under the parent node state combination, and the total number of related samples, which are used to construct the conditional probability interval of capability state nodes and top event nodes in step S5.

[0022] The fourth category is current status record data, including alarm records, event logs, device availability status, fault codes, and records and corresponding nodes in operating parameters within the current time window. This data is used to construct evidence conditions in step S6. This step only records and preprocesses this type of data and does not perform evidence judgment.

[0023] Step S3: Generate the prior probability interval of the basic fault node.

[0024] For basic fault nodes X1 to X15, based on the basic fault node parameter data obtained in step S2, and according to the fault count and operation exposure duration of the basic fault events corresponding to the basic fault nodes within the statistical period, a fault frequency range is obtained through small-sample precise interval estimation. This fault frequency range is then mapped to a priori probability range within a unit time window to avoid bias caused by using point-value probabilities. Specifically: Step S3.1: Describe the fault arrival process of a node using a homogeneous Poisson procedure: Let the... iEach basic fault node is represented as: X i , representing the corresponding basic fault events, assuming each X i The number of faults within the statistical period is N i and the corresponding job exposure duration is T ,set up l i express X i The frequency of failures determines the duration of operation exposure. T Internal fault count , m i Indicates the duration of exposure during the operation. T Down X i The expected number of occurrences.

[0025] Step S3.2: Use Garwood precise interval construction. m i Confidence interval: given confidence level 1- v ( v (Default value is 0.05), obtained based on the Garwood precise interval of the chi-square quantile. m i The confidence interval, i.e. ,when N i When equal to 0, m i The lower bound is set to 0, and the upper bound is still set to 0. Calculation, where For a degree of freedom of 2 N i The cumulative probability is v / 2 of the chi-square distribution quantiles For a degree of freedom of 2 ( N i +1), the cumulative probability is 1- v / 2 of the chi-square distribution quantiles.

[0026] Step S3.3: Obtain the fault frequency range through small sample precise interval estimation: ,in, , .

[0027] Step S3.4: Map the fault frequency range to the prior probability range within a unit time window: Let the unit time window be Δ t The prior probability interval for a basic fault node to experience at least one failure within a unit time window is: ,in, , .

[0028] Step S4: The conditional probability interval of the subsystem fault node is generated.

[0029] For subsystem fault nodes M2 to M5 that lack co-occurrence sample data of parent and child nodes, conditional probability intervals are constructed based on the subsystem fault node parameter data through link probability intervals and leakage probability intervals. This avoids filling in the conditional probability table item by item and reduces the parameter requirements for multi-parent node conditional probability modeling. Specifically: Step 4.1: For a faulty node in a certain subsystem M i Set its parent node X j The link probability range to the faulty node of this subsystem is: , Indicates the parent node X j Triggers M when used alone i The lower bound of the probability, M is triggered when the parent node acts alone. i The upper bound of the probability.

[0030] Step 4.2: Set the leakage probability interval to... This is used to characterize the triggering effect of latent factors that are not explicitly modeled on the fault nodes of the subsystem, wherein M is triggered by latent factors that are not explicitly modeled. i The lower bound of the probability, M is triggered by latent factors that are not explicitly modeled. i The upper bound of the probability.

[0031] Step 4.3: Let a certain combination of parent node states be denoted as... A Under this parent node state combination, the subsystem fault node M i The upper and lower bounds of the conditional probability interval for occurrence are calculated as follows: and .in, M i =1 means M i occur, M i =0 means M i It did not happen. J ( A ) represents the combination of parent node states. A The set of parent node numbers whose state is 1.

[0032] Step 4.4, Subsystem Fault Node Mi The upper and lower bounds of the conditional probability interval for the event not occurring are calculated as follows: and .

[0033] Step S5: The conditional probability intervals for capability state nodes and top event nodes are generated.

[0034] For capability state nodes M6 and M7 and top event node M1 that have co-occurrence sample data of parent and child nodes, construct the corresponding conditional probability intervals based on the co-occurrence sample data of parent and child nodes. Specifically: Furthermore, for the node to be estimated M i’ Its parent node state combination is π In this parent node state combination π Below, the co-occurrence count between the node to be estimated and its parent node is: n 1|π The total relevant sample size of the node to be estimated is N π Then the conditional probability interval of the node to be estimated is ,in s This is a smoothing parameter, set to 1 by default, used to conservatively correct the conditional probability interval when the sample size is limited.

[0035] Step S6: Constructing evidence conditions for real-time observation.

[0036] The current state records of the ship's dynamic positioning system are transformed into evidential conditions for a hierarchical inaccurate probabilistic graphical model. These evidential conditions include fault event evidence and capability status event evidence. The fault event evidence and capability status event evidence are combined to form predictive evidential conditions for the current time window; when a stationing capability failure event occurs, diagnostic evidential conditions are formed.

[0037] Furthermore, the evidence of the fault event is discrete state data that can directly confirm whether a basic fault node or a subsystem fault node has occurred or not. It originates from alarm records, equipment availability status, fault codes, and event logs in the current status record data, and includes the following four categories: Evidence of propulsion link failure events includes thruster unavailability, thruster drive failure, loss of thruster feedback signal, or thruster control failure. When the specific failure type can be determined, the propulsion link failure event evidence is mapped to evidence of the occurrence of the corresponding basic failure node in X1 to X4. When the specific failure type cannot be distinguished but the propulsion subsystem anomaly can be determined, the propulsion link failure event evidence is mapped to M2=1.

[0038] Evidence of power supply and distribution link failure events includes generator unavailability, bus fault, power distribution fault, power management system fault, or power supply protection action; when the specific fault type can be determined, the evidence of power supply and distribution link failure events is mapped to evidence of the occurrence of the corresponding basic fault node in X5 to X8; when the specific fault type cannot be distinguished but the abnormality of the power supply and distribution electronic system can be determined, the evidence of power supply and distribution link failure events is mapped to M3=1.

[0039] Evidence of control link failure events includes failures of dynamic positioning controllers, control software, human-machine interfaces, uninterruptible power supplies, or control power supply. When the specific failure type can be determined, the evidence of control link failure events is mapped to evidence of the occurrence of the corresponding basic failure node in X9 to X11. When the specific failure type cannot be distinguished but the abnormality of the control subsystem can be determined, the evidence of control link failure events is mapped to M4=1.

[0040] Location reference link failure event evidence includes location reference system failure, location reference signal loss, location reference signal rejection, location reference communication failure, or location reference sensor failure. When the specific failure type can be determined, the location reference link failure event evidence is mapped to evidence of the occurrence of the corresponding basic failure node in X12 to X15. When the specific failure type cannot be distinguished but the location reference subsystem is determined to be abnormal, the location reference link failure event evidence is mapped to M5=1.

[0041] Furthermore, if a certain basic fault node X i If the evidence of the failure event can confirm its occurrence, then X i =1 Add evidence conditions; if it cannot be confirmed, no evidence is set for the node, and the node remains unobserved.

[0042] Furthermore, the capability status event evidence is discrete event evidence triggered by operating parameters or event logs in the current status record data, used to characterize whether the judgment conditions for thrust holding capability anomaly, control closed-loop capability anomaly, or stationary capability failure have been met. Operating parameters are converted into event evidence for the corresponding node when preset criteria are met. This includes the following three categories: For abnormal events in thrust holding capability, the time window Δ t Internal thrust utilization rate U T The preset thrust utilization threshold is c T The duration threshold is t T ,when U T ≥ c T And the duration is not less than t T If an abnormal event in thrust maintenance capability is determined to be true, then M6=1 is added as an evidentiary condition; if the event log shows thrust allocation saturation, insufficient available thrust, or insufficient thrust redundancy capability, then M6=1 is also added as an evidentiary condition.

[0043] For abnormal events in the control closed-loop capability, the time window Δ t The number of effective internal location reference sources is N R The preset minimum number of valid position reference sources is N Rmin The duration threshold is t R ,when N R < N Rmin And the duration is not less than t R If an abnormal event is determined to be a control closed-loop capability event, then M7=1 is added as an evidentiary condition; if the event log shows that all position reference sources are lost, position reference sources are continuously rejected, or the number of position reference sources is insufficient, then M7=1 is also added as an evidentiary condition.

[0044] For a failure of the stationary capability event, the position deviation is: D p The allowable positional deviation is D pmax The heading deviation is D h Allowable heading deviation is D hmax The duration threshold is t D ,when D p ≥ D pmax or D h ≥ D hmax And the duration is not less than t D If a positional capability failure event is detected, it is considered a valid event. Similarly, if the event log shows an excessive positional deviation, excessive heading deviation, or a position loss alarm, the positional capability failure event is also considered valid. Positional capability failure events are not used as input evidence in the risk prediction phase; however, in the cause identification phase, M1=1 is used as diagnostic evidence input to the hierarchical imprecise probability graphical model.

[0045] Furthermore, evidence of failure events and evidence of capability status events are combined to form predictive evidence conditions for the current time window. E PWhen a failure of the stationary capability event is established, diagnostic evidence is formed. E D = E P ∪{M1=1}.

[0046] Step S7: Prediction of failure risk of stationary capability and determination of early warning status.

[0047] The prior probability interval obtained in step S3, the conditional probability intervals obtained in steps S4 and S5, and the predictive evidence conditions obtained in step S6 are input into the hierarchical imprecise probabilistic graphical model to perform predictive inference on the top event node M1 of the ship's dynamic positioning system, thereby achieving early warning status discrimination. Specifically: Step 7.1: Using the prior probability interval, conditional probability interval, and prediction evidence conditions in the hierarchical imprecise probabilistic graphical model as constraints, perform prediction inference within the entire set of feasible probability parameters, and calculate the time window Δ for node M1. t Minimum of the posterior probability of the following occurrence and maximum value The posterior probability interval is obtained. This is the predicted risk range, where This represents the conditions for predicting evidence. The predictive inference process is implemented using interval constraint propagation, linear programming, or hierarchical approximate updates.

[0048] Step 7.2: Calculate the center value of the predicted risk interval for the failure of the ship's dynamic positioning system's holding capability. and interval width ,in, C t Used to characterize the risk level of failure of the stationary capability. W t Used to characterize the degree of uncertainty in risk assessment under current evidence conditions.

[0049] Step 7.3, according to C t , W t , and C t Based on the changing trend within a continuous time window, the warning status is determined: (1) When C t Below the preset risk threshold and W t When the value is below the preset uncertainty threshold, it is determined to be in a normal state; (2) when C t Below the preset risk threshold but W t When the value exceeds the preset uncertainty threshold, it is determined to be an uncertainty concern state; (3) whenC t Higher than the preset risk threshold, or When the risk level exceeds the preset conservative risk threshold, it is determined to be a warning state of failure of the stationary capability; (4) when C t When the price continues to rise over multiple consecutive time windows and the increase exceeds a preset growth threshold, it is considered a trend warning state.

[0050] Step 7.4: If the evidence conditions already include M6=1 or M7=1, then output a prompt indicating abnormal thrust holding capability or abnormal control closed-loop capability simultaneously; if M6 or M7 is not observed, then calculate its posterior probability interval to help determine the risk of abnormal capability status.

[0051] Step S8: Key Cause Identification and Result Output. When the risk of stationary capability failure reaches the warning state as determined in step S7, or when the evidence condition for a stationary capability failure event is formed in step S6 (i.e., M1=1), the diagnostic evidence condition is input into the hierarchical imprecise probabilistic graphical model to perform diagnostic reasoning on candidate cause nodes, achieving reverse posterior update. The candidate cause nodes include basic fault nodes X1 to X15, subsystem fault nodes M2 to M5, and capability status nodes M6 and M7.

[0052] Furthermore, for any candidate causative node Y Calculate the posterior probability interval of its diagnostic inference: That is, the diagnostic risk range, in which Indicates the conditions for diagnostic evidence. This represents the lower bound of the probability interval for diagnosing risk. This represents the upper bound of the probability range for diagnosis risk.

[0053] Furthermore, the central value of the diagnostic risk interval is calculated. Candidate causal nodes Y The posterior increment is .

[0054] Furthermore, based on the central value of the diagnostic risk interval posterior increment And the upper bound of the probability of the diagnostic risk interval. The candidate causative nodes are sorted. The sorting value is... I Y Calculated as , α 1. α 2. α 3 represents the weighting coefficient, which can be set to default values ​​of 0.4, 0.3, and 0.3 respectively. According to... I YA sequence of key causative nodes is generated from high to low, with the nodes ranked higher being the focus of investigation for those that increase the risk of failure of stationary capability.

[0055] Furthermore, the final output is as follows: (1) Early warning results of station capability failure, including the predicted risk range of M1, the center value of the predicted risk range, the range width and the early warning status; (2) Key cause node identification results, including the diagnostic risk range, posterior increment, ranking value and key cause node sequence of candidate cause nodes; (3) Operation and maintenance handling suggestions, including one or more of the following: promotion link inspection, power supply and distribution link inspection, control link inspection, location reference link inspection, redundancy capability verification, manual review, system degradation, work suspension and restoration of normal monitoring.

[0056] Furthermore, the method can be updated on a rolling basis based on newly added observation data.

[0057] Furthermore, the present invention also provides a system for early warning and cause identification of the failure risk of a ship's dynamic positioning system's stationing capability, such as... Figure 2 As shown, it includes a data access module 100, a model configuration module 200, an interval parameter generation module 300, an evidence condition construction module 400, a risk prediction and early warning module 500, a cause identification module 600, and a result output module 700.

[0058] The data access module 100 is used for data collection and processing; the model configuration module 200 is used to construct a hierarchical imprecise probability graph model; the interval parameter generation module 300 is used to generate the prior probability interval of the basic fault node, the conditional probability interval of the subsystem fault node, and the conditional probability interval of the capability status node and the top event node; the evidence condition construction module 400 is used to construct the evidence conditions for real-time observation; the risk prediction and early warning module 500 is used to perform risk prediction and early warning status discrimination for stationary capability failure; the cause identification module 600 is used to identify key causes; and the result output module 700 is used to output the results.

[0059] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention sets up a subsystem fault layer and a capability status layer between the basic fault layer and the top event layer, and in particular sets up a thrust holding capability degradation node M6 and a control closed loop capability degradation node M7, so that the model can characterize the process of local equipment faults developing into stationary capability degradation through subsystem faults, thereby shifting the risk identification object from the dislocation result to the stationary capability degradation stage.

[0060] (2) The present invention constructs conditional probability intervals for M2 to M5 using link probability intervals and leakage probability intervals, reducing the parameter requirements of a complete conditional probability table. In an embodiment containing M2, M3, M4 and M5, a total of 56 sets of conditional probability parameters are required if a complete conditional probability table is used, while only 19 sets of link probability and leakage probability parameters are required using the present invention, reducing the number of parameters constructed by about 66.1%, and reducing the difficulty of model construction under small sample and combined missing conditions.

[0061] (3) The present invention adopts a small sample precise interval estimation method based on the job exposure time for basic fault nodes, which can preserve the upper bound of risk when the fault count is small or even zero. For example, under a job exposure time of 2400 hours, the basic fault node with a fault count of 0 can still obtain a prior probability interval of [0, 0.001536] per unit time window, avoiding misjudging unobserved faults as zero risk.

[0062] (4) The present invention uses the predicted risk interval to represent the risk of failure of the stationing capability, and uses the interval center value to represent the risk level, the interval width to represent the degree of uncertainty, and the interval upper limit to represent the conservative risk boundary, so that the early warning result can simultaneously reflect the risk size and the credibility of the risk judgment, avoiding the shortcomings of a single probability value that cannot distinguish between high risk and high uncertainty.

[0063] (5) This invention distinguishes between predictive evidence conditions and diagnostic evidence conditions: in the risk prediction stage, the failure event of the stationary capability is not used as input evidence, but M1 is used as the target node to be predicted; in the cause identification stage, M1=1 is used as diagnostic evidence for reverse posterior update. Thus, it can achieve risk warning before failure and identify key cause nodes after risk increases or failure events occur.

[0064] (6) The present invention generates a ranking value based on the central value of the diagnostic risk interval, the posterior increment and the upper bound of the diagnostic risk interval of the candidate causative node, which can identify the key risk propagation node that leads to the increased risk of failure of the stationary capability, rather than being limited to outputting the basic fault node that has been alarmed, thereby improving the pertinence of fault investigation and operation and maintenance. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0066] Figure 1 This is a flowchart of the method for early warning of failure of the stationing capability of a ship dynamic positioning system in this invention; Figure 2 This is a schematic diagram of the ship dynamic positioning system's stationing capability failure risk early warning system in this invention; Figure 3This is a schematic diagram of the hierarchical imprecise probabilistic graphical model structure in this invention. Detailed Implementation

[0067] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the ship type, statistical period, operation exposure duration, fault count, threshold, probability interval, and weight parameters in this embodiment are only used to illustrate the implementation process of the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0068] Example 1: Early warning method for the failure of the stationing capability of a ship's dynamic positioning system: This embodiment focuses on an offshore engineering vessel employing dynamic positioning operations. The vessel's dynamic positioning system includes a propulsion subsystem, a power supply and distribution electronics system, a control subsystem, and a position reference subsystem. The unit time window is 1 hour, and the vessel's cumulative dynamic positioning operation exposure time within the historical statistical period is 2400 hours, with a confidence level of 95%.

[0069] S1, Construction of a hierarchical imprecise probabilistic graphical model: This embodiment constructs a hierarchical inaccurate probabilistic graphical model including a basic fault layer, a subsystem fault layer, a capability state layer, and a top event layer, such as... Figure 3 As shown, this forms a hierarchical risk evolution structure of basic failure, subsystem failure, capability status degradation, and stationary capability failure.

[0070] S2, Data Collection and Processing: Historical fault event data and current observable status data of the ship's dynamic positioning system were collected and organized. After time alignment, merging of duplicate records, removal of abnormal records, and merging of fault categories, basic fault node parameter data, subsystem fault node parameter data, parent-child node co-occurrence sample data, and current status record data were generated, as shown in Tables 1, 2, 3, and 4, respectively.

[0071] Table 1: Basic Fault Node Parameter Data

[0072] Table 2: Subsystem Fault Node Parameter Data

[0073] Table 3: Sample data of co-occurrence of parent and child nodes

[0074] Table 4: Current Status Record Data

[0075] Through the above processing, Table 1 is used to generate the prior probability interval of the basic fault node in step S3, Table 2 is used to generate the conditional probability interval of the subsystem fault node in step S4, Table 3 is used to generate the conditional probability interval of the capability status node and the top event node in step S5, and Table 4 is used to construct the predictive evidence condition and the diagnostic evidence condition in step S6.

[0076] S3, Generation of prior probability intervals for basic fault nodes: Based on Table 1, the prior probability intervals from X1 to X15 are constructed as shown in Table 5.

[0077] Table 5: Prior Probability Intervals of Basic Failure Nodes

[0078] Taking X2 as an example, X2 occurred twice within the statistical period, corresponding to an exposure time of 2400 hours. Using the small sample precise interval estimation method, the prior probability interval of X2 within a 1-hour time window can be obtained as [0.000101, 0.003006]. Other basic fault nodes are calculated in the same way.

[0079] S4, Subsystem Fault Node Conditional Probability Interval Generation: For M2 to M5, this embodiment uses link probability intervals and leakage probability intervals to construct conditional probability intervals, as shown in Table 2.

[0080] In this embodiment, the leakage probability range for M2 to M5 is [0.005, 0.020]. In other embodiments, the leakage probability range can be adjusted according to the ship type, system redundancy level, and historical operation and maintenance data.

[0081] When only X2 occurs in the parent node state combination, the conditional probability interval for M2 to occur is determined by the link probability interval from X2 to M2 and the leakage probability interval of M2. Specifically, the link probability interval from X2 to M2 is [0.50, 0.75], and the leakage probability interval of M2 is [0.005, 0.020], then the conditional probability interval for M2 to occur is [0.503, 0.755].

[0082] Similarly, when only X12 occurs in the parent node state combination, the link probability interval from X12 to M5 is [0.45, 0.70], the leakage probability interval of M5 is [0.005, 0.020], and the conditional probability interval of M5 is [0.453, 0.706].

[0083] When multiple parent nodes occur simultaneously, the link probability interval for each occurring parent node is taken, and together with the leakage probability interval, the conditional probability interval for the faulty node in the target subsystem is calculated. This allows us to obtain the conditional probability intervals for M2 to M5 under different combinations of parent node states without manually filling in the complete conditional probability table item by item.

[0084] S5, Generation of conditional probability intervals for capability state nodes and top event nodes: For M6, M7 and M1, conditional probability intervals are constructed based on Table 3, resulting in Table 6.

[0085] Table 6: Conditional Probability Intervals for Capability State Nodes and Top Event Nodes

[0086] Taking the state combination M6=1 and M7=1 in Table 6 as an example, if the total relevant sample size is 10, and the failure of the stationary ability occurs 6 times (M1=1), that is, the co-occurrence count is 6, then the corresponding conditional probability interval [0.60, 0.85] can be obtained.

[0087] S6, Construction of evidence conditions for real-time observation: Assume the current time window is from 13:00 to 14:00. Based on Table 4, the following status records are obtained: (1) An alarm was triggered in the thruster control system, and it was confirmed that the corresponding thruster control system was faulty, thus forming evidence of a fault event X2=1; (2) A location reference interference alarm occurs, and the corresponding location reference interference fault is confirmed, thus forming evidence of a fault event X12=1; (3) The thrust utilization rate did not meet the thrust maintenance capability degradation event criterion, therefore no evidence of the capability status event of M6=1 was formed; (4) The number of valid location reference sources is not lower than the minimum number threshold, therefore no evidence of a capability state event with M7=1 is formed; (5) No position deviation exceeding the limit, heading deviation exceeding the limit, or displacement alarm occurred, therefore no diagnostic evidence of M1=1 was formed.

[0088] Therefore, the current predictive evidence conditions are: X2=1, X12=1.

[0089] S7, Prediction and early warning of failure risk in stationary capability: Inputting the prior probability interval obtained from S3, the conditional probability intervals obtained from S4 and S5, and the predictive evidence condition obtained from S6 into the hierarchical imprecise probabilistic graphical model, the predicted risk interval of M1 is calculated, and the predicted risk interval for the failure of M1's stationary capability within the current time window is [0.109, 0.432]. Therefore, the center value of the predicted risk interval is 0.271, and the interval width is 0.323.

[0090] In this embodiment, the risk threshold is set to 0.25, the uncertainty threshold is set to 0.25, and the conservative risk threshold is set to 0.40. Since the center value of the predicted risk interval (0.271) is higher than the risk threshold (0.25), the upper bound of the predicted risk interval (0.432) is higher than the conservative risk threshold (0.40), and the interval width (0.323) is higher than the uncertainty threshold (0.25), the system outputs a "stationing capability failure risk warning status" and marks it as a "high uncertainty warning".

[0091] The results indicate that while no berthing capability failures have yet occurred, propulsion link and position reference link failures have significantly increased the risk of berthing capability failures, warranting close monitoring and manual review.

[0092] S8, Key Cause Identification and Result Output: Since S7 has triggered a risk warning of failure in its stationary capability, the system uses M1=1 as one of the diagnostic evidence conditions, which, together with the current predictive evidence conditions, form the diagnostic evidence conditions. That is, the diagnostic evidence conditions include: X2=1, X12=1, and M1=1.

[0093] The diagnostic evidence conditions are input into a hierarchical imprecise probabilistic graphical model, and the candidate causative nodes are updated using a reverse posterior approach. The candidate causative nodes include X1 to X15 and M2 to M7.

[0094] In this embodiment, the diagnostic calculation results of some candidate nodes with larger ranking values ​​are shown in Table 7.

[0095] Table 7: Diagnostic Results of Some Candidate Causative Nodes

[0096] In this embodiment, the ranking value is calculated by weighting the central value of the diagnostic risk interval, the posterior increment, and the upper bound of the diagnostic risk interval; wherein the weight of the central value of the diagnosis is 0.4, the weight of the posterior increment is 0.3, and the weight of the upper bound of the diagnostic risk interval is 0.3.

[0097] Taking M6 as an example, its diagnostic center value is 0.700, its posterior increment is 0.343, and its upper limit of the diagnostic risk interval is 0.820. Therefore, the ranking value of M6 is 0.629.

[0098] As shown in Table 7, M2, M6, M5, and M7 are ranked highest. Therefore, it can be concluded that the increased risk of stationary capability failure is mainly related to propulsion subsystem failure, thrust holding capability degradation, position reference subsystem failure, and control closed-loop capability degradation.

[0099] Finally, the system outputs the following results: First, the results of the early warning of the risk of failure of the stationary capability: the predicted risk range of M1 is [0.109, 0.432], the center value of the range is 0.271, the range width is 0.323, the early warning status is the risk of failure of the stationary capability, and it is marked as a high uncertainty warning.

[0100] Second, the results of the identification of key causal nodes: the faulty nodes have been confirmed as X2 and X12; the key nodes to be investigated are M2, M6, M5 and M7 in order.

[0101] Third, maintenance and handling recommendations: Inspect the thruster control system and its communication links; verify whether the thruster control commands and feedback are consistent; check the interference sources and reference source rejection records of the position reference system; verify the number and quality of valid position reference sources; manually review the thrust maintenance capability and control closed-loop capability; if necessary, enter degraded operation or suspend high-risk operations.

[0102] As can be seen from this embodiment, the present invention can obtain the predicted risk range of the stationary capability failure based on the current fault event evidence, even before actual misalignment or position deviation exceeds the limit, and further identify the key causal nodes that lead to increased risk.

[0103] Example 2: System Example: This embodiment also provides a risk warning system for the failure of the stationing capability of a ship's dynamic positioning system. The system is deployed in a shipborne monitoring terminal or a shore-based maintenance platform and includes a data access module 100, a model configuration module 200, an interval parameter generation module 300, an evidence condition construction module 400, a risk prediction and early warning module 500, a cause identification module 600, and a result output module 700.

[0104] The data access module 100 accesses historical fault event data and current observable status data of the ship's dynamic positioning system, and generates the datasets shown in Tables 1 to 4 for other modules to use.

[0105] The model configuration module 200 stores the node structure and connection relationships of X1 to X15, M2 to M7 and M1.

[0106] The interval parameter generation module 300 generates the prior probability interval and conditional probability interval for each node.

[0107] The evidence condition construction module 400 generates predictive evidence conditions based on the current state record data, namely X2=1 and X12=1.

[0108] The risk prediction and early warning module 500 calculates the predicted risk range of M1 as [0.109, 0.432] and outputs the risk warning status of the stationary capability failure.

[0109] After the risk warning state is triggered, the cause identification module 600 adds M1=1 to the diagnostic evidence conditions and generates the key cause node ranking results shown in Table 7.

[0110] The output module 700 outputs the warning results of the failure risk of the stationing capability, the identification results of the key causal nodes, and the operation and maintenance disposal suggestions to the shipboard display terminal or the shore-based operation and maintenance platform.

[0111] The data processing procedure of this system is as follows: The data access module 100 sends structured data to the model configuration module 200, the interval parameter generation module 300, and the evidence condition construction module 400 respectively; the model configuration module 200 provides node structure and connection relationships to the interval parameter generation module 300, the risk prediction and early warning module 500, and the cause identification module 600; the interval parameter generation module 300 sends the prior probability interval and the conditional probability interval to the risk prediction and early warning module 500 and the cause identification module 600; the evidence condition construction module 400 sends the predicted evidence conditions to the risk prediction and early warning module 500, and provides diagnostic evidence conditions to the cause identification module 600 when the stationary capability failure event is established; the risk prediction and early warning module 500 starts the cause identification module 600 after triggering the early warning; the result output module 700 receives the calculation results of the risk prediction and early warning module 500 and the cause identification module 600 and displays the final result.

[0112] Example 3: Electronic devices and storage media: This embodiment also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory. When executed by the processor, the computer program implements the ship dynamic positioning system stationary capability failure risk warning method described in Embodiment 1.

[0113] The electronic device can be a shipborne industrial control computer, a shipborne monitoring terminal, a shore-based server, an edge computing device, or a cloud server.

[0114] This embodiment also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the method described in Embodiment 1.

[0115] The fault count, link probability interval, co-occurrence samples, threshold, and weight coefficient in the above embodiments are all exemplary values.

[0116] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, adjustments can still be made according to different ship types, dynamic positioning levels, operational scenarios, historical data scales, and maintenance requirements without departing from the spirit and scope of the invention. These adjustments do not alter the basic technical concept of the present invention. Therefore, the scope of the present invention should be understood and recognized as covering all such other variations or modifications.

Claims

1. A method for early warning of the risk of failure of the stationing capability of a ship's dynamic positioning system, characterized in that, Includes the following steps: Step S1: Construction of a hierarchical inaccurate probabilistic graphical model; A hierarchical inaccurate probabilistic graphical model for ship dynamic positioning systems is constructed as the Credal network, which includes a basic fault layer, a subsystem fault layer, a capability state layer, and a top event layer. Step S2, Data Collection and Processing; Collect and organize historical fault event data and current observable status data of the ship's dynamic positioning system, perform time alignment, merge duplicate records, remove abnormal records and merge fault categories to form a dataset for inference calculation; Step S3: Generate the prior probability interval of the basic fault node; For basic fault nodes X1 to X15, based on the basic fault node parameter data obtained in step S2, the fault frequency range is obtained by small sample precise interval estimation according to the fault count and operation exposure duration of the basic fault events corresponding to the basic fault nodes in the statistical period, and the fault frequency range is mapped to the prior probability range within a unit time window. Step S4: The conditional probability interval of the subsystem fault node is generated; For subsystem fault nodes M2 to M5 that lack parent-child node co-occurrence sample data, conditional probability intervals are constructed based on subsystem fault node parameter data through link probability intervals and leakage probability intervals. The specific process of step S4 is as follows: Step 4.1: For a faulty node in a certain subsystem M i Set its parent node X j The link probability range to the faulty node of this subsystem is: , Indicates the parent node X j Triggers M when used alone i The lower bound of the probability, M is triggered when the parent node acts alone. i The upper bound of the probability; Step 4.2: Set the leakage probability interval to... This is used to characterize the triggering effect of latent factors that are not explicitly modeled on the fault nodes of the subsystem, wherein M is triggered by latent factors that are not explicitly modeled. i The lower bound of the probability, M is triggered by latent factors that are not explicitly modeled. i The upper bound of the probability; Step 4.3: Let a certain combination of parent node states be denoted as... A Under this parent node state combination, the subsystem fault node M i The upper and lower bounds of the conditional probability interval for occurrence are calculated as follows: and ; in, M i =1 means M i occur, M i =0 means M i It did not happen. J ( A ) represents the combination of parent node states. A The set of parent node numbers whose state is 1; Step 4.4, Subsystem Fault Node M i The upper and lower bounds of the conditional probability interval for the event not occurring are calculated as follows: and ; Step S5: The conditional probability intervals for capability state nodes and top event nodes are generated. For state nodes M6 and M7 and top event node M1 that have co-occurrence sample data of parent and child nodes, construct the corresponding conditional probability intervals based on the co-occurrence sample data of parent and child nodes. Step S6: Constructing evidence conditions for real-time observation; The current state record data of the ship's dynamic positioning system is transformed into evidence conditions for a hierarchical inaccurate probabilistic graphical model. The evidence conditions include failure event evidence and capability status event evidence. The failure event evidence and capability status event evidence are combined to form predictive evidence conditions for the current time window. When the stationing capability failure event is established, diagnostic evidence conditions are formed. Step S7: Prediction of failure risk of stationary capability and determination of early warning status; The prior probability interval obtained in step S3, the conditional probability interval obtained in steps S4 and S5, and the prediction evidence condition obtained in step S6 are input into the hierarchical imprecise probability graphical model to perform prediction and reasoning on the top event node M1 of the ship dynamic positioning system, thereby realizing the early warning state discrimination. Step S8, key cause identification and result output; when the condition for determining the risk of stationary capability failure in step S7 reaches the warning state, or when the evidence condition for forming a stationary capability failure event in step S6, i.e., M1=1, the diagnostic evidence condition is input into the hierarchical imprecise probability graph model to perform diagnostic reasoning on the candidate cause nodes and realize reverse posterior update; the candidate cause nodes include basic fault nodes X1 to X15, subsystem fault nodes M2 to M5, and capability status nodes M6 and M7.

2. The method for early warning of failure risk of a ship's dynamic positioning system's stationing capability according to claim 1, characterized in that, In step S1: The hierarchical imprecise probabilistic graphical model is the Credal network; the Credal network is a graphical model that uses directed acyclic graphs to represent variable dependencies and uses interval probabilities or probability sets to represent prior probabilities and conditional probabilities. The basic fault layer includes basic fault nodes X1 to X15, namely: X1 Thruster unit and drive fault, X2 Thruster control system fault, X3 Thruster feedback signal fault, X4 Thruster hydraulic and auxiliary support fault, X5 Power station management or control equipment fault, X6 Power distribution and protection fault, X7 Engine component fault, X8 Fuel, air, cooling or lubrication auxiliary system fault, X9 Dynamic positioning control software fault, X10 Dynamic positioning control hardware or human-machine interface fault, X11 Uninterruptible power supply or control power supply fault, X12 Position reference interference fault, X13 Position The fault layers are categorized into three levels: reference software or configuration failure, X14 position reference mechanical sensor or deployment failure, and X15 position reference communication or electrical hardware failure. The subsystem fault layer includes nodes M2 to M5, where M2 represents a propulsion subsystem failure, M3 represents a power supply electronics system failure, M4 represents a control subsystem failure, and M5 represents a position reference subsystem failure. The capability status layer includes nodes M6 and M7, where M6 represents thrust holding capability degradation and M7 represents control closed-loop capability degradation. The top event layer is node M1, representing the failure of the dynamic positioning system's stationary capability. The connection relationships between the nodes include at least the following: X1, X2, X3, and X4 are connected to M2; X5, X6, X7, and X8 are connected to M3; X9, X10, and X11 are connected to M4; X12, X13, X14, and X15 are connected to M5; M2 and M3 are connected to M6; M4 and M5 are connected to M7; and M6 and M7 are connected to M1, forming a hierarchical risk evolution structure consisting of basic faults, subsystem faults, capability status, and dynamic positioning system stationary capability failure, i.e., a hierarchical inaccurate probability graphical model. In the hierarchical imprecise probabilistic graphical model, a node equal to 1 indicates that the corresponding event has occurred, and a node equal to 0 indicates that the corresponding event has not occurred.

3. The method for early warning of failure risk of the stationing capability of a ship dynamic positioning system according to claim 2, characterized in that, In step S2: The dataset includes four categories: The first category is basic fault node parameter data, corresponding to basic fault nodes X1 to X15, including the fault count of each basic fault event within the statistical period and the corresponding job exposure duration, which is used to construct the prior probability interval of the basic fault node in step S3. The second category is subsystem fault node parameter data, corresponding to subsystem fault nodes M2 to M5, including the link probability interval and leakage probability interval from each parent node to the subsystem fault node, which is used to determine the conditional probability interval of the subsystem fault node in step S4. The third category is the co-occurrence sample data of parent and child nodes, corresponding to capability state nodes M6, M7 and top event node M1. It includes the parent node state combination within the statistical period, the co-occurrence count between the target node and the parent node under the parent node state combination, and the total relevant sample size, which are used to construct the conditional probability interval of capability state nodes and top event nodes in step S5. The fourth category is current status record data, including alarm records, event logs, device availability status, fault codes, records in operating parameters and corresponding nodes within the current time window, which are used to construct evidence conditions in step S6.

4. The method for early warning of failure risk of the stationing capability of a ship dynamic positioning system according to claim 3, characterized in that, The specific process of step S3 is as follows: Step S3.1: Describe the fault arrival process of a node using a homogeneous Poisson procedure: Let the... i Each basic fault node is represented as: X i , representing the corresponding basic fault events, assuming each X i The number of faults within the statistical period is N i and the corresponding job exposure time is T ,set up λ i express X i The frequency of failures determines the duration of operation exposure. T Internal fault count , μ i Indicates the duration of exposure during the operation. T Down X i The expected number of occurrences; Step S3.2: Use Garwood precise interval construction. μ i Confidence interval: given confidence level 1- v Garwood exact intervals based on chi-square quantiles are obtained μ i The confidence interval, i.e. ,when N i When equal to 0, μ i The lower bound is set to 0, and the upper bound is still set to 0. Calculation, where For a degree of freedom of 2 N i The cumulative probability is v / 2 of the chi-square distribution quantiles For a degree of freedom of 2 ( N i +1), the cumulative probability is 1- v / 2 of the chi-square distribution quantiles; Step S3.3: Obtain the fault frequency range through small sample precise interval estimation: ,in, , ; Step S3.4: Map the fault frequency range to a prior probability range within a unit time window: Let the unit time window be Δ t The prior probability interval for a basic fault node to experience at least one failure within a unit time window is: ,in, , .

5. The method for early warning of failure risk of the stationing capability of a ship dynamic positioning system according to claim 1, characterized in that, In step S5: For the node to be estimated M i’ Its parent node state combination is π In this parent node state combination π Below, the co-occurrence count between the node to be estimated and its parent node is: n 1|π The total relevant sample size of the node to be estimated is N π Then the conditional probability interval of the node to be estimated is: ,in s This is a smoothing parameter used to conservatively correct the conditional probability interval when the sample size is limited.

6. The method for early warning of failure risk of stationing capability of ship dynamic positioning system according to claim 5, characterized in that, In step S6: The evidence of the fault event is discrete state data that can directly confirm whether a basic fault node or a subsystem fault node has occurred or not. It comes from alarm records, equipment availability status, fault codes, and event logs in the current status record data, and includes the following four categories: Evidence of propulsion link failure events includes thruster unavailability, thruster drive failure, loss of thruster feedback signal, or thruster control failure. When the specific failure type can be determined, the propulsion link failure event evidence is mapped to evidence of the occurrence of the corresponding basic failure node in X1 to X4. When the specific failure type cannot be distinguished but the propulsion subsystem anomaly can be determined, the propulsion link failure event evidence is mapped to M2=1. Evidence of power supply and distribution link failure events includes generator unavailability, bus fault, power distribution fault, power management system fault, or power supply protection action; when the specific fault type can be determined, the evidence of power supply and distribution link failure events is mapped to evidence of the occurrence of the corresponding basic fault node in X5 to X8; when the specific fault type cannot be distinguished but the abnormality of the power supply and distribution electronic system can be determined, the evidence of power supply and distribution link failure events is mapped to M3=1. Evidence of control link failure events includes failures of dynamic positioning controllers, control software, human-machine interface, uninterruptible power supply, or control power supply. When the specific failure type can be determined, the evidence of control link failure events is mapped to evidence of the occurrence of the corresponding basic failure node in X9 to X11. When the specific failure type cannot be distinguished but the abnormality of the control subsystem can be determined, the evidence of control link failure events is mapped to M4=1. Location reference link failure event evidence includes location reference system failure, location reference signal loss, location reference signal rejection, location reference communication failure, or location reference sensor failure. When the specific failure type can be determined, the location reference link failure event evidence is mapped to evidence of the occurrence of the corresponding basic failure node in X12 to X15. When the specific failure type cannot be distinguished but the location reference subsystem anomaly can be determined, the location reference link failure event evidence is mapped to M5=1. If a basic fault node X i If the evidence of the failure event can confirm its occurrence, then X i =1. Add evidence conditions; If it cannot be confirmed, no evidence is set for that node, and the node remains unobserved. The capability status event evidence is discrete event evidence triggered by operating parameters or event logs in the current status record data. It is used to characterize whether the judgment conditions for thrust maintenance capability anomaly, control closed-loop capability anomaly, or stationary capability failure have been met. Operating parameters are converted into event evidence for the corresponding node when preset criteria are met. It includes the following three categories: For abnormal events in thrust holding capability, the time window Δ t Internal thrust utilization rate U T The preset thrust utilization threshold is γ T The duration threshold is τ T ,when U T ≥ γ T And the duration is not less than τ T If an abnormal event in thrust maintenance capability is determined to be true, then M6=1 is added as an evidentiary condition; if the event log shows thrust allocation saturation, insufficient available thrust, or insufficient thrust redundancy capability, then M6=1 is also added as an evidentiary condition. For abnormal events in the control closed-loop capability, the time window Δ t The number of effective internal location reference sources is N R The preset minimum number of valid position reference sources is N Rmin The duration threshold is τ R ,when N R < N Rmin And the duration is not less than τ R When an abnormal event is determined to be a control loop capability issue, M7=1 is added as an evidentiary condition. If the event log shows that all location reference sources are lost, location reference sources are continuously rejected, or the number of location reference sources is insufficient, M7=1 will also be added as an evidence condition. For a failure of the stationary capability event, the position deviation is: D p The allowable positional deviation is D pmax The heading deviation is D h Allowable heading deviation is D hmax The duration threshold is τ D ,when D p ≥ D pmax or D h ≥ D hmax And the duration is not less than τ D If the event log shows an alarm indicating excessive position deviation, excessive heading deviation, or loss of position, the event is considered to be a failure of the stationary capability. The failure of stationary capability is not used as an input evidence condition in the risk prediction stage, but in the cause identification stage, M1=1 is used as a diagnostic evidence condition input into the stratified imprecise probability graphical model. By combining evidence of failure events and evidence of capability status events, predictive evidence conditions for the current time window are formed. E P When a failure of the stationary capability event is established, diagnostic evidence is formed. E D = E P ∪{M1=1}.

7. The method for early warning of failure risk of the stationing capability of a ship dynamic positioning system according to claim 6, characterized in that, The specific process of step S7 is as follows: Step 7.1: Using the prior probability interval, conditional probability interval, and prediction evidence conditions in the hierarchical imprecise probabilistic graphical model as constraints, perform prediction inference within the entire set of feasible probability parameters, and calculate the time window Δ for node M1. t Minimum of the posterior probability of occurrence and maximum value The posterior probability interval is obtained. This is the predicted risk range, where The conditions for predicting evidence are represented; the predictive reasoning process is implemented using interval constraint propagation, linear programming, or hierarchical approximate update methods. Step 7.2: Calculate the center value of the predicted risk interval for the failure of the ship's dynamic positioning system's holding capability. and interval width ,in, C t Used to characterize the level of risk of failure of the stationary capability. W t Used to characterize the degree of uncertainty in risk assessment under current evidence conditions; Step 7.3, according to C t , W t , and C t Based on the changing trend within a continuous time window, the warning status is determined: (1) When C t Below the preset risk threshold and W t When the value is below the preset uncertainty threshold, it is determined to be in a normal state; (2) when C t Below the preset risk threshold but W t When the value exceeds the preset uncertainty threshold, it is determined to be an uncertainty concern state; (3) when C t Higher than the preset risk threshold, or When the risk level exceeds the preset conservative risk threshold, it is determined to be a warning state of failure of the stationary capability; (4) when C t When the price continues to rise over multiple consecutive time windows and the increase exceeds a preset growth threshold, it is considered a trend warning state. Step 7.4: If the evidence conditions already include M6=1 or M7=1, then output a prompt indicating abnormal thrust holding capability or abnormal control closed-loop capability simultaneously; if M6 or M7 is not observed, then calculate its posterior probability interval to help determine the risk of abnormal capability status.

8. The method for early warning of failure risk of stationing capability of ship dynamic positioning system according to claim 7, characterized in that, In step S8: For any candidate causative node Y Calculate the posterior probability interval of its diagnostic inference. That is, the diagnostic risk range, in which Indicates the conditions for diagnostic evidence. This represents the lower bound of the probability interval for diagnosing risk. This represents the upper bound of the probability range for diagnosis risk. Calculate the center value of the diagnostic risk interval Candidate causal nodes Y The posterior increment is ; Based on the central value of the diagnostic risk interval posterior increment And the upper bound of the probability of the diagnostic risk interval. Sort the candidate causative nodes. Sort value I Y Calculated as , α 1. α 2. α 3 is the weighting coefficient; according to I Y A sequence of key causative nodes is generated from high to low, and the nodes ranked at the top are the key targets for investigation that increase the risk of failure of the stationary capability. The final output is as follows: (1) Early warning results of station capability failure, including the predicted risk range of M1, the center value of the predicted risk range, the range width and the early warning status; (2) Key cause node identification results, including the diagnostic risk range of candidate cause nodes, the posterior increment, the ranking value and the sequence of key cause nodes; (3) Operation and maintenance handling suggestions, including one or more of the following: promotion link inspection, power supply and distribution link inspection, control link inspection, location reference link inspection, redundancy capability verification, manual review, system downgrade, work suspension and restoration of normal monitoring.

9. A system for early warning and cause identification of failure risk in a ship's dynamic positioning system, used to execute the method described in any one of claims 1-8, characterized in that, It includes a data access module (100), a model configuration module (200), an interval parameter generation module (300), an evidence condition construction module (400), a risk prediction and early warning module (500), a cause identification module (600), and a result output module (700). The data access module (100) is used for data collection and processing, the model configuration module (200) is used to construct a hierarchical inaccurate probability graph model, the interval parameter generation module (300) is used to generate the prior probability interval of the basic fault node, the conditional probability interval of the subsystem fault node, and the conditional probability interval of the capability status node and the top event node, the evidence condition construction module (400) is used to construct the evidence conditions for real-time observation, the risk prediction and early warning module (500) is used to perform risk prediction and early warning status discrimination of stationary capability failure, the cause identification module (600) is used to identify key causes, and the result output module (700) is used to output the results.

Images