A multi-scale autonomous ship seaworthiness risk propagation path identification and control method

Abstract

This invention discloses a multi-scale method for identifying and controlling the propagation paths of airworthiness risks for autonomous vessels. The method includes: hierarchically modeling multi-source airworthiness risk factors based on ontology to construct a structured knowledge graph; establishing logical relationships between risk factors using structural equation modeling and converting them into a Bayesian network; dynamically evaluating input vessel operation data and identifying multi-scale airworthiness risk propagation paths and key risk nodes using graph theory topology analysis; and finally, generating a hierarchical control strategy based on the identification results. This invention enables dynamic and probabilistic assessment of the initial and ongoing airworthiness of autonomous vessels, effectively identifying risk propagation mechanisms and key links, and providing support for the safe design, operation, and supervision of autonomous vessels.

Description

Technical Field

[0001] This invention relates to the field of ship safety risk assessment technology, and in particular to a multi-scale method for identifying and controlling the propagation path of seaworthiness risks of autonomous ships. Background Technology

[0002] With the deepening application of intelligent and digital technologies in the shipping industry, autonomous ships are gradually moving from the proof-of-concept stage to the actual operation stage. Compared with traditional ships, autonomous ships introduce autonomous decision-making systems, remote control centers, and complex information and communication networks during operation, and their system structure exhibits highly coupled, multi-level, and cross-domain collaborative characteristics.

[0003] Seaworthiness and the ability to maintain continuous seaworthiness are prerequisites for the practical operation of autonomous vessels. Compared to conventional vessels, the disruptive changes in the interaction methods of autonomous vessel systems, the dynamic switching of operating modes, and the emergence of new risk factors all significantly interfere with and damage their seaworthiness, leading to a decline in their ability to withstand foreseeable risks and navigate safely. However, vessel autonomy and remote operation remain immature technological fields. Combining physical autonomy with remote human supervision presents unique safety challenges, particularly in maintaining and verifying seaworthiness under dynamic and distributed control conditions. Maintaining seaworthiness cannot rely solely on periodic certification; instead, it must address the systems used on board, the remote operation center, and the necessary connectivity to ensure that autonomous vessel operations offer the same or even higher safety levels than conventional vessels. Therefore, it is essential to ensure that autonomous vessels can operate efficiently and safely under different operating modes while maintaining seaworthiness.

[0004] Existing research primarily addresses the seaworthiness of autonomous vessels from a legal and regulatory perspective, while technical investigations focus on navigation risks. However, seaworthiness risk and navigation risk are fundamentally different. In navigation risk, seaworthiness status reflects the operational scope for maintaining structural and functional integrity throughout the entire lifecycle. Traditional methods based on accident chains or single reliability models struggle to characterize the complex interactions between these factors and fail to identify the propagation paths and amplification effects of risks within the system, thus limiting the effective assessment and risk prevention of the ongoing seaworthiness status of autonomous vessels. Therefore, a comprehensive assessment method is urgently needed that can provide a structured representation of seaworthiness risks, dynamic propagation analysis, and key node identification to ensure the avoidance of risks arising from the impact of operational functions on the safety of personnel, the environment, and the vessel itself. Summary of the Invention

[0005] This invention provides a multi-scale method for identifying and controlling the propagation path of seaworthiness risks of autonomous ships, in order to overcome the above-mentioned technical problems.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: A multi-scale method for identifying and controlling the propagation path of seaworthiness risks for autonomous vessels, comprising the following steps: S1: Based on ontological theory, multi-source autonomous ship seaworthiness risk factors are hierarchically modeled, their data types, accuracy and verifiable forms are analyzed, and a standardized and structured seaworthiness risk knowledge graph is formed, namely, autonomous ship seaworthiness risk ontology model. S2: Based on the autonomous ship seaworthiness risk ontology model, a structural equation model is established. The structural equation model characterizes the logical relationship between multi-source autonomous ship seaworthiness risk factors through latent variables and observed variables. S3: The structural equation model is converted into a Bayesian network form using a set model transformation strategy, thereby constructing a Bayesian model for the propagation of autonomous ship seaworthiness risk to characterize the initial and ongoing seaworthiness risk propagation of autonomous ships. S4: Obtain operational data of the autonomous vessel at different operational stages and input it into the autonomous vessel airworthiness risk propagation Bayesian model for dynamic evaluation, thereby identifying multi-scale airworthiness risk assessment results. The multi-scale airworthiness risk assessment results include airworthiness risk propagation paths and key risk nodes in the airworthiness risk propagation paths; S5: Generate a hierarchical control strategy for the seaworthiness risks of autonomous vessels based on the airworthiness risk propagation path and key risk nodes. This includes multi-level risk control measures for the risk propagation path, such as source prevention, propagation interruption, and consequence mitigation, as well as multi-dimensional control measures for key risk nodes, such as engineering, management, and technology, thereby achieving risk control.

[0007] Furthermore, the autonomous ship seaworthiness risk ontology model includes: The category includes several primary indicators: hull seaworthiness, operator suitability, and cargo suitability. The primary indicators include several secondary indicators, among which: Seaworthiness of a vessel includes: the configuration of relevant certificates or documents, structural integrity, vessel stability, supplies, nautical publications, functional integrity of equipment, ROC environmental conditions, availability of redundant systems, and cybersecurity. Operator suitability includes: qualification compliance, personnel obedience compliance, and compliance with the Maritime Labour Convention. Cargo suitability includes: the configuration of relevant freight business documents, cargo status, and the completeness of the cargo status monitoring and alarm system functions; Secondary indicators include several tertiary indicators, among which: Structural integrity includes: hull structural integrity and fire-resistant structural integrity; The functional integrity of the equipment includes: the functional integrity of the onboard emergency system, the functional integrity of the onboard emergency system, the functional integrity of the onboard navigation equipment, the functional integrity of the onboard navigation equipment, the functional integrity of the onboard anti-pollution system, and the functional integrity of the onboard anti-pollution system. Redundancy system availability includes: data redundancy system availability, equipment redundancy system availability, power redundancy system availability, and redundancy necessity verification system availability; Compliance with qualifications includes: the configuration of certificates for crew members on board and the configuration of certificates for remote operators; Personnel compliance includes: the onboard crew's watchkeeping and lookout status, the remote operator's watchkeeping and lookout status, the onboard crew's training and drills, and the remote operator's training and drills; Cargo condition includes: cargo hold tightness, cargo preparation status, cargo stowage or securing status, and suitability for loading dangerous goods. Object attributes, which are used to represent the semantic relationship between several secondary indicators without corresponding tertiary indicators and the risk factors contained in the tertiary indicators; Data attributes are used to characterize the quantifiable, verifiable, and recordable external manifestations of risk factors contained in secondary and tertiary indicators that do not have corresponding tertiary indicators. These attributes include text-type and logical data.

[0008] Furthermore, in S2, the specific steps for establishing a structural equation model based on the autonomous ship seaworthiness risk ontology model include: Based on the operating modes of different types of ships, several secondary indicators are extracted from the classes of the autonomous ship seaworthiness risk ontology model and used as potential variables in the structural equation model. Based on the secondary indicators without corresponding tertiary indicators and their corresponding data attributes, and the tertiary indicators and their corresponding data attributes, observation variables that match the latent variables are formed. The structural equation model extraction results include latent variables extracted based on the autonomous ship seaworthiness risk ontology model and their corresponding observed variables, where: The latent variable X1 is the ship-related certificates and documents, and the corresponding observed variable s0 is the ship-related certificates and documents carried on board or held in the ship. Latent variable X2 represents the structural condition of the ship and the remote control center, corresponding to the observed variables: s1 represents the integrity of the ship's hull structure; S2 represents the fire-resistant structural integrity. S3 represents ship stability; s4 represents the environmental conditions of the remote control center; Latent variable X3 represents the functional integrity of equipment on board and in the remote control center, corresponding to the observed variables: S5 represents the configuration and performance status of the remote control center and shipborne navigation equipment; S6 represents the configuration and performance status of the remote control center and shipboard emergency facilities; S7 represents the configuration and performance status of the remote control center and shipborne pollution prevention facilities; The latent variable X4 is the qualification certification of remote operators, and the corresponding observed variable h1 is the holding of remote operator qualification certificates and documents; Latent variable X5 represents the compliance of personnel in the remote control center, corresponding to the observed variable: h2 shows the staffing and duty status of the remote control center; h3 is for training and exercises for remote operators; The latent variable X6 is cargo-related documents, and the corresponding observed variable c1 is cargo-carrying or cargo-holding documents. Latent variable X7 represents the condition of the goods, corresponding to the observed variable: c2 refers to cargo hold airtightness; c3 refers to cargo stowage and securing; c4 indicates that the cabin is ready for use. C5 is the condition for the suitability of dangerous goods for loading; The latent variable X8 represents the functional integrity of the cargo monitoring and alarm system, while the corresponding observed variable c6 represents the configuration and performance status of the cargo status monitoring and alarm system. Latent variable X9 represents the availability of the redundant system, corresponding to the observed variable: r1 represents equipment redundancy; r2 represents power redundancy; r3 represents data storage redundancy; r4 is a verification of the redundancy necessity; The latent variable X10 represents network security resilience, and the corresponding observed variable r5 represents the anti-interference capability of the communication connection.

[0009] Furthermore, in S3, the specific steps for converting the structural equation model into a Bayesian network form using a predetermined model transformation strategy include: The latent and observed variables in the structural equation model are transformed into nodes of a Bayesian network, and the nodes of the Bayesian network are defined, including: The latent variables in the structural equation model are defined as target child nodes of a Bayesian network and named airworthiness state SW. The state of the target child node is discretized into three mutually exclusive states, including: SYES: Airworthiness; Par: Partially airworthy; UnS: Unairworthy; In the structural equation model, the observed variables are defined as parent nodes, and the parent nodes are discretized into three states, including: NO: No defect detected; UN: Non-retention defect detected; YES: Defect detected; Preserve the correlation between latent variables and observed variables in the structural equation model and map it as a directed connection between Bayesian nodes; Frequency statistics and probability transformations are performed on latent and observed variables to establish a conditional probability table, quantifying the impact of parent node state combinations on child node states. The path coefficients in the structural equation model are replaced with the conditional probability table, which also includes deterministic logical mapping rules from parent node state combinations to child node states. Highest priority: If any parent node shows "YES", then SW must be "UnS"; Second priority: If there is no "YES" but the parent node has "UN", then SW cannot be "SYES", it can only be "Par" or "UnS"; Normal state: If all parent nodes are "NO", then SW is "SYES"; Based on the expert panel's evaluation, the structure and path relationships of the Bayesian network were modified and optimized, thereby achieving the modeling transformation from structural equation modeling to Bayesian network modeling.

[0010] Furthermore, in S4, the steps of acquiring operational data of the autonomous vessel at different operational stages and inputting it into the autonomous vessel's airworthiness risk propagation Bayesian model for dynamic evaluation, thereby identifying the airworthiness risk propagation path in the multi-scale airworthiness risk assessment results, include: Preprocessing of operational data from autonomous vessels at different operational stages; Based on the processed dataset, the EM algorithm is executed to learn parameters and obtain the prior probabilities of some nodes. For the seaworthiness risk factors of autonomous ships, the prior probabilities are determined by combining the suggested values ​​of similar event failure probabilities extracted from accident reports with the experience of domain experts. Based on the prior probability and the Bayesian model of airworthiness risk propagation of autonomous ships, the posterior probability of each node is calculated, and the airworthiness risk propagation path is identified according to the difference between the posterior probability and the prior probability of each node.

[0011] Furthermore, in S4, operational data of the autonomous vessel at different operational stages is acquired and input into the autonomous vessel's seaworthiness risk propagation Bayesian model for dynamic evaluation, thereby identifying key risk nodes in the multi-scale seaworthiness risk assessment results, including: A topological analysis of the Bayesian model of seaworthiness risk propagation for autonomous vessels is performed using graph theory analysis methods. The degree centrality, betweenness centrality, and compact centrality are calculated to quantify the importance of nodes in the Bayesian network. Nodes are ranked based on degree centrality, betweenness centrality, and tight centrality, and the top-ranked nodes are defined as key risk nodes in the airworthiness risk propagation path.

[0012] Beneficial effects: This invention establishes an ontology model of autonomous vessel seaworthiness risk, enabling a structured expression of autonomous vessel seaworthiness risk and improving the consistency and reusability of risk information. By establishing a structural equation model and employing a set model transformation strategy to convert the structural equation model into a Bayesian network form, it can dynamically characterize the propagation and amplification process of seaworthiness risk within the system, realize dynamic assessment of the initial and continuous seaworthiness of autonomous vessels, identify key risk propagation paths and control nodes, and provide technical support for the design optimization, operational decision-making, and supervision of autonomous vessels. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a multi-scale autonomous ship airworthiness risk propagation path identification and control method in this invention; Figure 2 This is a schematic diagram of a typical SEM in an embodiment of the present invention; Figure 3 This is a structured representation of the seaworthiness risks of autonomous vessels in this embodiment of the invention; Figure 4 This is a structural model diagram of the propagation of seaworthiness risks of autonomous vessels in an embodiment of the present invention; Figure 5 This is a schematic diagram of the Bayesian model structure for the propagation of airworthiness risks of autonomous vessels in an embodiment of the present invention; Figure 6 This is a constraint diagram in an embodiment of the present invention; Figure 7 This is the Bayesian result of the initial airworthiness risk propagation network of autonomous ships after parameter learning using the EM algorithm in this embodiment of the invention; Figure 8 This is the Bayesian result of the autonomous ship continuous airworthiness risk propagation network after parameter learning using the EM algorithm in this embodiment of the invention; Figure 9 This is a schematic diagram illustrating the evidence propagation results of initial airworthiness in an embodiment of the present invention; Figure 10 This is a comparison diagram of the probabilities of nodes before and after initial airworthiness in an embodiment of the present invention; Figure 11 This is a schematic diagram illustrating the evidence propagation results of continued airworthiness in an embodiment of the present invention; Figure 12 This is a comparison diagram of the probabilities of nodes in the continuous airworthiness of this invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] This embodiment provides a multi-scale method for identifying and controlling the propagation path of airworthiness risks for autonomous vessels, such as... Figure 1 As shown, the specific steps include: S1: Based on ontological theory, multi-source autonomous ship seaworthiness risk factors are hierarchically modeled, their data types, accuracy and verifiable forms are analyzed, and a standardized and structured seaworthiness risk knowledge graph is formed, namely, autonomous ship seaworthiness risk ontology model. S2: Based on the aforementioned autonomous vessel seaworthiness risk ontology model, a structural equation model is established. The structural equation model characterizes the logical relationships between multi-source autonomous vessel seaworthiness risk factors through latent variables (LV) and observed variables (OV). S3: The structural equation model is converted into a Bayesian network form using a set model transformation strategy, thereby constructing a Bayesian model for the propagation of autonomous ship seaworthiness risk to characterize the initial and ongoing seaworthiness risk propagation of autonomous ships. S4: Obtain operational data of the autonomous vessel at different operational stages and input it into the autonomous vessel airworthiness risk propagation Bayesian model for dynamic evaluation, thereby identifying multi-scale airworthiness risk assessment results. The multi-scale airworthiness risk assessment results include airworthiness risk propagation paths and key risk nodes in the airworthiness risk propagation paths; S5: Generate a hierarchical control strategy for the seaworthiness risks of autonomous vessels based on the airworthiness risk propagation path and key risk nodes. This includes multi-level risk control measures for the risk propagation path, such as "source prevention - propagation interruption - consequence mitigation", and multi-dimensional control measures for key risk nodes, such as engineering, management and technology, thereby achieving risk control.

[0017] Specifically, this embodiment includes ontology modeling, probabilistic reasoning, and graph theory analysis to identify the propagation paths and key nodes of seaworthiness risks for autonomous vessels and to establish targeted control measures. It is particularly suitable for assessing the initial and ongoing seaworthiness of vessels in remote control or autonomous operation modes throughout their entire voyage lifecycle. This embodiment develops a novel multi-scale analysis framework, which, by developing and validating a framework for analyzing complex, dynamic, and data-scarce autonomous vessel systems, contributes to enhancing system safety research. These results also provide actionable insights for regulatory agencies and shipping operators, highlighting how cost-effective interventions at key nodes can significantly improve system reliability. Therefore, this embodiment establishes a theoretical foundation and decision support tools for the safety-oriented design, regulatory oversight, and management of autonomous vessels.

[0018] In this embodiment, taking an autonomous cargo transport vessel in remote control mode as an example, a seaworthiness risk ontology model of an autonomous vessel was designed using the Protégé ontology editing tool within the scope of seaworthiness risk of autonomous vessels. The seaworthiness risk ontology model of an autonomous vessel includes: (1) Class, which includes several primary indicators: hull seaworthiness, operator suitability and cargo suitability; The primary indicators include several secondary indicators, among which: Seaworthiness of a vessel includes: the configuration of relevant certificates or documents, structural integrity, vessel stability, supplies, nautical publications, functional integrity of equipment, environmental conditions of the Remote Operations Centre (ROC), availability of redundant systems, and cybersecurity. Specifically, the supplies include lubricating oil, heavy oil, etc. Operator suitability includes: qualification compliance, personnel obedience compliance, and compliance with the Maritime Labour Convention. Cargo suitability includes: the configuration of relevant freight business documents, cargo status, and the completeness of the cargo status monitoring and alarm system functions; Secondary indicators include several tertiary indicators, among which: Structural integrity includes: hull structural integrity and fire-resistant structural integrity. Fire-resistant structural integrity includes fire doors, fire compartments, etc. The functional integrity of the equipment includes: the functional integrity of the onboard emergency system, the functional integrity of the onboard emergency system, the functional integrity of the onboard navigation equipment, the functional integrity of the onboard navigation equipment, the functional integrity of the onboard anti-pollution system, and the functional integrity of the onboard anti-pollution system. Redundancy system availability includes: data redundancy system availability, equipment redundancy system availability, power redundancy system availability, and redundancy necessity verification system availability; Compliance with qualifications includes: the configuration of certificates for crew members on board and the configuration of certificates for remote operators; Personnel compliance includes: the onboard crew's watchkeeping and lookout status, the remote operator's watchkeeping and lookout status, the onboard crew's training and drills, and the remote operator's training and drills; Cargo condition includes: cargo hold tightness, cargo preparation status, cargo stowage or securing status, and suitability for loading dangerous goods.

[0019] Specifically, in this embodiment, secondary indicators without corresponding tertiary indicators are directly used as underlying indicators, and their specific connotations (components / parts) are illustrated through examples. For example, the secondary indicator "ship certificates and related documents" is not further subdivided, but can be explained by listing specific certificate names as examples.

[0020] (2) Object attributes, which are used to represent the semantic relationship between several secondary indicators without corresponding tertiary indicators and the risk factors contained in the tertiary indicators; Specifically, object attributes are derived from the analysis of historical accident reports, and semantic relationships are given in numerical form, such as "has_effect_0", "has_effect_1", "has_effect_2", and "has_effect_3". A value of 0 indicates that no two factors among the risk factors of each indicator will occur simultaneously, and no interaction has been observed. A value of 1 indicates a one-way triggering relationship, meaning one risk may unilaterally trigger the occurrence of another risk. A value of 2 indicates that the two risk factors are parallel; they remain independent but may simultaneously affect the seaworthiness of autonomous vessels under certain conditions. A value of 3 indicates a two-way interactive relationship, where the two factors interact and jointly produce a synergistic negative impact on the seaworthiness of autonomous vessels, with an impact greater than the simple additive result (i.e., 1+1>2). In the case of asymmetric two-way influence, i.e., when the impact of risk factor A on risk factor B is greater than (or less than) the impact of risk factor B on risk factor A, the value is also 3.

[0021] (3) Data attributes, used to characterize the quantifiable, inspectable and recordable external manifestations of risk factors contained in several secondary and tertiary indicators that do not have corresponding tertiary indicators, including text type and logical data; Data attributes are the recording methods for observations and results corresponding to on-site inspections, tests, and judgments of risk status. They are either text-based (e.g., xsd:string) or logical data types (e.g., xsd:Boolean), representing the format of the results given after the inspection. Text types are suitable for describing inspection results. For example, when inspecting the structural integrity of a ship (a level 3 indicator), it's necessary to check its corrosion and deformation levels. Corrosion and deformation are data attributes, and the inspection results need to be output as text, such as specifying the corrosion level and deformation area of ​​the deck (an instance corresponding to hull structural integrity). Boolean logic is suitable for risk factor data attributes that can be judged by binary values ​​such as yes / no, qualified / unqualified, present / absent. For example, the data attributes corresponding to certificate-type risk factors include possession and carrying; the output after inspection is Boolean logic data, such as yes or no.

[0022] In this embodiment, the autonomous vessel seaworthiness risk ontology model supports the integration of heterogeneous data sources. By refining the seaworthiness risk factors of autonomous vessels, it achieves standardized representation and structured storage of autonomous vessel seaworthiness risk information, such as... Figure 3 As shown, the autonomous vessel seaworthiness risk ontology model adopts a hierarchical structure, comprising "classes," "object attributes," and "data attributes." Classes constitute the core conceptual system, organized into multiple levels of seaworthiness risk indicators. Object attributes represent the semantic relationships between risk factors. Data attributes are introduced as a supplement to object attributes, capturing the attributes of each instance in a precise and standardized manner.

[0023] In a specific embodiment, S2, the specific steps for establishing a structural equation model based on the autonomous ship seaworthiness risk ontology model include: Based on the operating modes of different types of ships, several secondary indicators are extracted from the classes of the autonomous ship seaworthiness risk ontology model and used as potential variables in the structural equation model. Based on the secondary indicators and their corresponding data attributes that do not have corresponding tertiary indicators, and the tertiary indicators and their corresponding data attributes, observation variables that match the latent variables are formed; wherein, the observation variables are selected and defined from the perspective of inspectable form, so that they can directly characterize the observable and detectable features of the risk in the corresponding latent variables.

[0024] The results of the structural equation model extraction are shown in Table 1, including the latent variables extracted based on the autonomous ship seaworthiness risk ontology model and their corresponding observed variables. Table 1. Extraction results of structural equation model

[0025] Specifically, structural equation modeling reveals the underlying mechanisms and structural relationships among airworthiness risk factors, thus becoming an important tool for theory-driven research. Figure 2 A typical SEM diagram is shown. Latent variables are represented by ellipses, observed variables by rectangles, and the direction of risk propagation is represented by each arrow. Structural equation modeling can reveal the transmission of airworthiness risks.

[0026] Specifically, from the perspective of control model innovation, tasks traditionally handled by humans, such as equipment failure repair, cargo condition maintenance, and hull structural integrity and stability maintenance, may be difficult to manage carefully. Therefore, new control models and operating mechanisms are needed to achieve equivalent replacement of human labor. On the other hand, autonomous vessels need to establish a secure and reliable network connection with one or more remote control centers throughout the entire voyage. Based on existing research findings and established industry consensus, autonomous vessels need to establish an airworthiness management system covering all aspects and adjust the airworthiness period to continuous airworthiness throughout the entire voyage.

[0027] To ensure that autonomous vessels possess the ability to withstand risks throughout the entire voyage, i.e., maintain seaworthiness throughout the voyage, this embodiment defines seaworthiness as an inherent attribute, enabling autonomous vessels to withstand general or reasonably foreseeable risks during contractually agreed voyages within their operational design domain, while maintaining physical and functional integrity. Seaworthiness describes the seaworthiness capability of an autonomous vessel, which this embodiment divides into initial seaworthiness and ongoing seaworthiness. Initial seaworthiness refers to the seaworthiness capability and maintained seaworthiness state possessed by the autonomous vessel before and at the start of voyage, denoted as SW_T1; ongoing seaworthiness refers to the seaworthiness capability and maintained seaworthiness state possessed by the autonomous vessel throughout the entire voyage, denoted as SW_T2. The corresponding structural model for the transmission of seaworthiness risks is as follows: Figure 4 As shown.

[0028] In a specific embodiment, S3, the specific steps of converting the structural equation model into a Bayesian network form using a set model transformation strategy include: The latent and observed variables in the structural equation model are transformed into nodes of a Bayesian network, and the nodes of the Bayesian network are defined, including: The latent variables in the structural equation model are defined as target child nodes of a Bayesian network and named airworthiness state SW. The state of the target child node is discretized into three mutually exclusive states, including: SYES: Airworthiness; Par: Partially airworthy; UnS: Unairworthy; The observed variables in the structural equation model are defined as parent nodes, and the parent nodes are discretized into three states, which strictly correspond to the PSC defect records, including: NO: No defect detected; UN: Non-retention defect detected; YES: Defect detected; Preserve the correlation between latent variables and observed variables in the structural equation model and map it as a directed connection between Bayesian nodes; Specifically, the fused network is a hierarchical, two-layer Bayesian network. The first layer (input layer) consists entirely of parent nodes; the second layer (output layer) has only one child node, representing the airworthiness state. Connections are established by a directed arrow emanating from each parent node, pointing to the child node SW. This signifies that all identified risks collectively determine the final airworthiness state.

[0029] Frequency statistics and probability transformations are performed on latent and observed variables to establish a conditional probability table, quantifying the impact of parent node state combinations on child node states. The path coefficients in the structural equation model are replaced with the conditional probability table, which also includes deterministic logical mapping rules from parent node state combinations to child node states. Highest priority: If any parent node shows "YES" (delay defect) → SW must be "UnS" (unairworthy); Second priority: If there is no "YES", but a parent node shows "UN" (non-delay defect) → SW cannot be "SYES" (airworthiness), it can only be "Par" (partial airworthiness) or "UnS". However, according to the previous point, it will not be "UnS" at this time, so the actual result is locked as "Par"; Normal state: All parent nodes are "NO" (no defects) → SW can only be "SYES" (airworthiness) in this case; Based on the expert panel's evaluation, the structure and path relationships of the Bayesian network were modified and optimized, thereby achieving the modeling transformation from structural equation modeling to Bayesian network modeling. For example, for a child node X1 with only a single parent node, only its parent node s0 is retained as a valid node in the Bayesian network, thereby reducing the subjectivity of the conditional probability definition process and improving the model accuracy; a directed edge from node s0 to node X2 is added to the Bayesian network to improve the causal relationships between nodes.

[0030] Specifically, according to the requirements of the SEM model, each latent variable needs to include at least three observed variables. However, in practice, some models only have latent variables without observed variables, or latent variable X1 has only one observed variable s0. Therefore, the quantization process of this structural equation model will be limited. To solve this problem, this embodiment uses a model transformation strategy to convert the structural model framework in the original structural equation model into a Bayesian network, such as... Figure 5 As shown, this transformation breaks through the constraints of the original model, enhances the ability to capture nonlinear relationships, and thus ensures the effective identification of airworthiness.

[0031] Specifically, this embodiment uses the Port State Control Records (APCIS) database in the Asia-Pacific region to obtain risk variables affecting the seaworthiness status of autonomous vessels. The PSC inspection adopts a progressive, tiered approach, with an initial inspection coverage of 100%, focusing on verifying the validity of certificates and the basic condition of the vessel. This inspection mechanism provides a systematic basis for determining a vessel's seaworthiness. A vessel is considered seaworthy when it fully complies with all regulations and has no known risks. On the other hand, partially acceptable seaworthiness describes a vessel that has not been detained but still has potential risks that must be addressed. A vessel is considered unseaworthy when it is detained due to a major defect that poses a direct or immediate threat to the safety of the vessel or the lives of those on board. Therefore, seaworthiness status can be divided into three states: seaworthy, partially seaworthy, and unseaworthy. In the PSC defect inspection database, defect status can be systematically divided into three categories: "No defect detected" status, indicating that no defects were found during the inspection; "Detected non-detention defect" status, referring to situations where defects were found but their severity did not meet the detention criteria; and "Detected detention defect" status, specifically referring to situations where an inspection confirms the existence of a defect that directly leads to the vessel's detention. Therefore, airworthiness risk is discretized into three states: NO, UN, and YES. In summary, this embodiment defines three constraints, such as... Figure 6 As shown.

[0032] In a specific embodiment, S4 involves acquiring operational data of the autonomous vessel at different operational stages and inputting it into the autonomous vessel's Bayesian model for airworthiness risk propagation for dynamic evaluation, thereby identifying the airworthiness risk propagation path in the multi-scale airworthiness risk assessment results. The specific steps include: Based on the Asia-Pacific Port State Control Records (APCIS) database, 514,306 inspection records were collected, containing detailed vessel information. This embodiment primarily focuses on vessels engaged in cargo transportation; therefore, data on the main types of cargo ships inspected in China were retained, including bulk carriers, chemical tankers, container ships, oil tankers, and roll-on / roll-off (Ro-Ro) cargo ships. After screening, a total of 62,175 data points were obtained, and 34,284 data points from 9,771 ships were retained to verify the method proposed in this embodiment. Perform data cleaning and correlation matching on the filtered data: Delete incomplete data; Since the defect codes in the APCIS database cannot be directly matched with Bayesian nodes, it is necessary to refer to the file "List of Tokyo MOU Deficiency Codes1" to retrieve the meaning of the defect codes in the database, match them with Bayesian nodes, and then change the defect codes in the database to node names. This converts the database into a form that can be read by the Bayesian network, and then inputs it into the Bayesian model for computation. The EM (Expectation-Maximization) algorithm is executed on the processed dataset for parameter learning, thereby obtaining the prior probabilities of some nodes. However, this dataset cannot cover all Bayesian network nodes. For some seaworthiness risk factors unique to autonomous vessels, the prior probabilities are determined by extracting suggested failure probability values ​​from similar events in traditional vessels from accident reports, and combining these with years of experience from domain experts, optimizing the extracted data based on actual production conditions. The final Bayesian results of the initial seaworthiness risk and the continuous seaworthiness risk propagation network after using the EM algorithm are as follows: Figure 7 and Figure 8 As shown, dynamic propagation analysis employs probabilistic inference using Bayesian networks, sequentially transmitting local evidence within the network to accurately simulate the cascading effects along risk paths. By utilizing the backward inference capability of the Bayesian model, the posterior probability of each node is determined. Based on the differences between the posterior and prior probabilities of each node, airworthiness risk propagation paths are identified. For example, the difference between the posterior and prior probabilities of each node is calculated, and nodes with a difference greater than 0 are selected. Nodes that can be connected to form a path are then chosen to obtain the airworthiness risk propagation path. The established Bayesian network model is used to infer the posterior probabilities of relevant nodes. The corresponding differences in risk propagation indicate a stronger cascading effect on airworthiness, thus providing a fundamental tool for risk system modeling. Specifically, in this embodiment, by setting the states of SW_T1 and SW_T2 to UnS, the posterior probability of relevant nodes is inferred using the autonomous ship seaworthiness risk propagation Bayesian model, and the corresponding changes in risk propagation are as follows: Figure 9 and Figure 11 As shown. Figure 10 and Figure 12 The comparison between the prior and posterior probabilities of the target node is shown; a larger difference indicates a stronger cascading effect on airworthiness. According to... Figure 11 and Figure 12 The results show that the larger the difference between the prior and posterior probabilities of a node, the greater its impact on seaworthiness, and nodes with significant changes can usually form chains. In the experimental results, the initial seaworthiness risk propagation paths are r4-X9-X3-SW_T1 and c3-X7-T1-X2-SW_T1, meaning that when node SW_T1 experiences an anomaly, its impact on the upstream nodes of that path is strongest. The continuous seaworthiness risk propagation paths are c2 / c3 / c4-X7_T1-X2-SW_T1-SW_T2 and h3-X5_T2-SW_T2. The initial and continuous seaworthiness risk propagation paths based on autonomous vessels can provide important theoretical basis for the targeted implementation of risk mitigation strategies.

[0033] In a specific embodiment, in S4, operational data of the autonomous vessel at different operational stages is acquired and input into the autonomous vessel's airworthiness risk propagation Bayesian model for dynamic evaluation, thereby identifying key risk nodes in the multi-scale airworthiness risk assessment results, including: A topological analysis of the Bayesian model of seaworthiness risk propagation for autonomous vessels is performed using graph theory analysis methods. The degree centrality, betweenness centrality, and compact centrality are calculated to quantify the importance of nodes in the Bayesian network. Nodes are ranked according to degree centrality, betweenness centrality, and tight centrality. The top-ranked nodes are the key risk nodes in the airworthiness risk propagation path. Specifically, the top 7 airworthiness risks are shown in Table 2; Table 2 Top 7 airworthiness risks

[0034] Specifically, each centrality metric reflects a different dimension of a node's influence. Degree centrality determines a node's direct influence; betweenness centrality determines a node's path control capability, its ability to control whether risk can be transmitted between other nodes; and tight centrality determines the efficiency of risk propagation between nodes. The larger the centrality metric value, the more significant the node's role in the corresponding dimension. In this embodiment, the top three nodes in each centrality metric are designated as key risk nodes.

[0035] In a specific embodiment, S5 generates a hierarchical control strategy for the seaworthiness risks of autonomous vessels based on the airworthiness risk propagation path and key risk nodes, including: The resulting critical risk propagation paths with cascading effects help accelerate and amplify the propagation of systemic risks. Centrality analysis reveals specific nodes dominating the network topology, providing information for prioritizing security barriers in risk propagation paths. Based on this information, a layered and multi-dimensional barrier strategy can be adopted to reward risk control measures at the source, propagation, and consequence levels, strategically disrupting risk paths. For example, at the source stage, attention is paid to the starting node of the risk propagation chain, and corresponding measures are formulated; during the propagation stage, highly centrality nodes are dynamically monitored to isolate or decouple interdependent risks within the system at the propagation level; at the consequence level, mitigation control measures emphasize controlling the impact as risks escalate. For critical risk nodes in the airworthiness risk propagation path, due to the different nature of the nodes, a multi-dimensional collaborative control measure system integrating engineering, management, and technology can be constructed to precisely formulate risk control measures. For example, the risk chain r4-X9-X3-SW_T1 indicates the impact of redundant systems on the functionality of equipment on board and in remote control centers, further affecting the initial seaworthiness of autonomous vessels, while mitigation measures are suitable for the equipment at the engineering level. Risk chain h3-X5_T2-SW_T2 indicates the impact of remote control center personnel compliance on the continued seaworthiness of autonomous vessels. Therefore, from a management perspective, corresponding control measures should be formulated.

[0036] In this embodiment, the Bayesian model of seaworthiness risk propagation for autonomous vessels uses conditional probability to capture nonlinear risks on small sample data. By combining probabilistic reasoning with graph theory analysis, it can dynamically characterize the propagation and amplification process of seaworthiness risks within the system. Identification of key risk propagation paths reveals the causal chain of risk events propagating within the system, thus clarifying the potential pathways of risk propagation. Based on this, key node identification determines the locations that play an amplifying or controlling role in the risk propagation process. Risk control strategies are formulated at key nodes and main causal paths to disrupt or weaken the propagation of risk events, effectively acting as protective barriers within the network. These risk control strategies effectively prevent the seaworthiness of autonomous vessels from deteriorating to an unacceptable state, reduce the probability of cascading failures, and improve the overall safety and reliability of the system.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying and controlling the propagation path of seaworthiness risks for autonomous vessels at multiple scales, characterized in that, The specific steps include: S1: Based on ontological theory, multi-source autonomous ship seaworthiness risk factors are hierarchically modeled, their data types, accuracy and verifiable forms are analyzed, and a standardized and structured seaworthiness risk knowledge graph is formed, namely, autonomous ship seaworthiness risk ontology model. S2: Based on the autonomous ship seaworthiness risk ontology model, a structural equation model is established. The structural equation model characterizes the logical relationship between multi-source autonomous ship seaworthiness risk factors through latent variables and observed variables. S3: The structural equation model is converted into a Bayesian network form using a set model transformation strategy, thereby constructing a Bayesian model for the propagation of autonomous ship seaworthiness risk to characterize the initial and ongoing seaworthiness risk propagation of autonomous ships. S4: Obtain operational data of the autonomous vessel at different operational stages and input it into the autonomous vessel airworthiness risk propagation Bayesian model for dynamic evaluation, thereby identifying multi-scale airworthiness risk assessment results. The multi-scale airworthiness risk assessment results include airworthiness risk propagation paths and key risk nodes in the airworthiness risk propagation paths; S5: Generate a hierarchical control strategy for the seaworthiness risks of autonomous vessels based on the airworthiness risk propagation path and key risk nodes. This includes multi-level risk control measures for the risk propagation path, such as source prevention, propagation interruption, and consequence mitigation, as well as multi-dimensional control measures for key risk nodes, such as engineering, management, and technology, thereby achieving risk control.

2. The method for identifying and controlling the propagation path of multi-scale autonomous ship seaworthiness risks according to claim 1, characterized in that, The autonomous ship seaworthiness risk ontology model includes: The category includes several primary indicators: hull seaworthiness, operator suitability, and cargo suitability. The primary indicators include several secondary indicators, among which: Seaworthiness of a vessel includes: the configuration of relevant certificates or documents, structural integrity, vessel stability, supplies, nautical publications, functional integrity of equipment, ROC environmental conditions, availability of redundant systems, and cybersecurity. Operator suitability includes: qualification compliance, personnel obedience compliance, and compliance with the Maritime Labour Convention. Cargo suitability includes: the configuration of relevant freight business documents, cargo status, and the completeness of the cargo status monitoring and alarm system functions; Secondary indicators include several tertiary indicators, among which: Structural integrity includes: hull structural integrity and fire-resistant structural integrity; The functional integrity of the equipment includes: the functional integrity of the onboard emergency system, the functional integrity of the onboard emergency system, the functional integrity of the onboard navigation equipment, the functional integrity of the onboard navigation equipment, the functional integrity of the onboard anti-pollution system, and the functional integrity of the onboard anti-pollution system. Redundancy system availability includes: data redundancy system availability, equipment redundancy system availability, power redundancy system availability, and redundancy necessity verification system availability; Compliance with qualifications includes: the configuration of certificates for crew members on board and the configuration of certificates for remote operators; Personnel compliance includes: the onboard crew's watchkeeping and lookout status, the remote operator's watchkeeping and lookout status, the onboard crew's training and drills, and the remote operator's training and drills; Cargo condition includes: cargo hold tightness, cargo preparation status, cargo stowage or securing status, and suitability for loading dangerous goods. Object attributes, which are used to represent the semantic relationship between several secondary indicators without corresponding tertiary indicators and the risk factors contained in the tertiary indicators; Data attributes are used to characterize the quantifiable, verifiable, and recordable external manifestations of risk factors contained in secondary and tertiary indicators that do not have corresponding tertiary indicators. These attributes include text-type and logical data.

3. The method for identifying and controlling the propagation path of multi-scale autonomous ship seaworthiness risks according to claim 2, characterized in that, In S2, the specific steps for establishing a structural equation model based on the autonomous ship seaworthiness risk ontology model include: Based on the operating modes of different types of ships, several secondary indicators are extracted from the classes of the autonomous ship seaworthiness risk ontology model and used as potential variables in the structural equation model. Based on the secondary indicators without corresponding tertiary indicators and their corresponding data attributes, and the tertiary indicators and their corresponding data attributes, observation variables that match the latent variables are formed. The structural equation model extraction results include latent variables extracted based on the autonomous ship seaworthiness risk ontology model and their corresponding observed variables, where: The latent variable X1 is the ship-related certificates and documents, and the corresponding observed variable s0 is the ship-related certificates and documents carried on board or held in the ship. Latent variable X2 represents the structural condition of the ship and the remote control center, corresponding to the observed variables: s1 represents the integrity of the ship's hull structure; S2 represents the fire-resistant structural integrity. S3 represents ship stability; s4 represents the environmental conditions of the remote control center; Latent variable X3 represents the functional integrity of equipment on board and in the remote control center, corresponding to the observed variables: S5 represents the configuration and performance status of the remote control center and shipborne navigation equipment; S6 represents the configuration and performance status of the remote control center and shipboard emergency facilities; S7 represents the configuration and performance status of the remote control center and shipborne pollution prevention facilities; The latent variable X4 is the qualification certification of remote operators, and the corresponding observed variable h1 is the holding of remote operator qualification certificates and documents; Latent variable X5 represents the compliance of personnel in the remote control center, corresponding to the observed variable: h2 shows the staffing and duty status of the remote control center; h3 is for training and exercises for remote operators; The latent variable X6 is cargo-related documents, and the corresponding observed variable c1 is cargo-carrying or cargo-holding documents. Latent variable X7 represents the condition of the goods, corresponding to the observed variable: c2 refers to cargo hold airtightness; c3 refers to cargo stowage and securing; c4 indicates that the cabin is ready for use. C5 is the condition for the suitability of dangerous goods for loading; The latent variable X8 represents the functional integrity of the cargo monitoring and alarm system, while the corresponding observed variable c6 represents the configuration and performance status of the cargo status monitoring and alarm system. Latent variable X9 represents the availability of the redundant system, corresponding to the observed variable: r1 represents equipment redundancy; r2 represents power redundancy; r3 represents data storage redundancy; r4 is a verification of the redundancy necessity; The latent variable X10 represents network security resilience, and the corresponding observed variable r5 represents the anti-interference capability of the communication connection.

4. The method for identifying and controlling the propagation path of multi-scale autonomous ship seaworthiness risks according to claim 3, characterized in that, In S3, the specific steps for converting the structural equation model into a Bayesian network form using a predefined model transformation strategy include: The latent and observed variables in the structural equation model are transformed into nodes of a Bayesian network, and the nodes of the Bayesian network are defined, including: The latent variables in the structural equation model are defined as target child nodes of a Bayesian network and named airworthiness state SW. The state of the target child node is discretized into three mutually exclusive states, including: SYES: Airworthiness; Par: Partially airworthy; UnS: Unairworthy; In the structural equation model, the observed variables are defined as parent nodes, and the parent nodes are discretized into three states, including: NO: No defect detected; UN: Non-retention defect detected; YES: Defect detected; Preserve the correlation between latent variables and observed variables in the structural equation model and map it as a directed connection between Bayesian nodes; Frequency statistics and probability transformations are performed on latent and observed variables to establish a conditional probability table, quantifying the impact of parent node state combinations on child node states. The path coefficients in the structural equation model are replaced with the conditional probability table, which also includes deterministic logical mapping rules from parent node state combinations to child node states. Highest priority: If any parent node shows "YES", then SW must be "UnS"; Second priority: If there is no "YES" but the parent node has "UN", then SW cannot be "SYES", it can only be "Par" or "UnS"; Normal state: If all parent nodes are "NO", then SW is "SYES"; Based on the expert panel's evaluation, the structure and path relationships of the Bayesian network were modified and optimized, thereby achieving the modeling transformation from structural equation modeling to Bayesian network modeling.

5. The method for identifying and controlling the propagation path of multi-scale autonomous ship seaworthiness risks according to claim 4, characterized in that, In S4, the steps of acquiring operational data of the autonomous vessel at different operational stages and inputting it into the autonomous vessel's airworthiness risk propagation Bayesian model for dynamic evaluation, thereby identifying the airworthiness risk propagation path in the multi-scale airworthiness risk assessment results, include: Preprocessing of operational data from autonomous vessels at different operational stages; Based on the processed dataset, the EM algorithm is executed to learn parameters and obtain the prior probabilities of some nodes. For the seaworthiness risk factors of autonomous ships, the prior probabilities are determined by combining the suggested values ​​of similar event failure probabilities extracted from accident reports with the experience of domain experts. Based on the prior probability and the Bayesian model of airworthiness risk propagation of autonomous ships, the posterior probability of each node is calculated, and the airworthiness risk propagation path is identified according to the difference between the posterior probability and the prior probability of each node.

6. The method for identifying and controlling the propagation path of multi-scale autonomous ship seaworthiness risks according to claim 5, characterized in that, In S4, operational data of the autonomous vessel at different operational stages is acquired and input into the autonomous vessel's seaworthiness risk propagation Bayesian model for dynamic evaluation, thereby identifying key risk nodes in the multi-scale seaworthiness risk assessment results, including: A topological analysis of the Bayesian model of seaworthiness risk propagation for autonomous vessels is performed using graph theory analysis methods. The degree centrality, betweenness centrality, and compact centrality are calculated to quantify the importance of nodes in the Bayesian network. Nodes are ranked based on degree centrality, betweenness centrality, and tight centrality, and the top-ranked nodes are defined as key risk nodes in the airworthiness risk propagation path.

Images