Method, device and medium for identifying key chain of polar maritime traffic disturbance propagation disaster

Abstract

The application discloses a polar maritime traffic disturbance propagation disaster-causing key chain identification method, equipment and medium, relates to the field of maritime traffic safety and artificial intelligence technology, and comprises the following steps: dividing the disturbance propagation disaster-causing process into seven levels, defining the connection set between the levels, forming the stage propulsion constraint, constructing the disturbance propagation disaster-causing chain network architecture; expanding the disaster-causing elements layer by layer to obtain a candidate scenario set; generating a structured representation, combining a rule base to generate a prompt word, inputting a large language model to perform multiple inferences; calculating the reliability score based on the judgment label and the confidence, and generating the verification factor; based on the reliability score and the verification factor, the reliable scenario set is obtained by screening the candidate scenarios; based on the weight of each disaster-causing element, the cumulative weight of each candidate disaster-causing chain is calculated, and finally the key disaster-causing chain is determined. The application improves the accuracy, efficiency and reliability of the key disaster-causing chain identification under the complex polar shipping scenario.

Description

Technical Field

[0001] This application relates to the fields of maritime traffic safety and artificial intelligence technology, and in particular to a method, device and medium for identifying the critical chain of disaster caused by the propagation of polar maritime traffic disturbances. Background Technology

[0002] The commercial value of polar shipping routes is increasingly evident, leading to a continuous increase in the number of vessels transiting polar waters. However, polar maritime transport systems operate in harsh environments, facing complex disturbances such as low temperatures, strong winds and waves, poor visibility, sea ice, navigational obstructions, and limited support at high latitudes. These disturbances are often not isolated disasters but rather follow a chain propagation mechanism: "disturbance exposure—gradual failure of resilience barriers—emergence of danger—accident occurrence—serious consequences," exhibiting typical multi-stage, nonlinear, and highly coupled characteristics. Therefore, accurately identifying the key chains of polar maritime transport disturbance propagation leading to disasters is of great significance for navigation risk prevention and emergency decision-making.

[0003] Relevant technologies for analyzing maritime accident chains mainly include: causal inference based on expert experience, probabilistic risk assessment based on fault trees or Bayesian networks, and statistical analysis based on historical data. However, in the specific scenario of polar shipping, these technologies have significant shortcomings: First, polar accident data samples are scarce and mostly unstructured text, making it difficult to support parameter calibration for traditional probabilistic models, and purely data-driven methods struggle to capture complex, phased evolutionary mechanisms; second, expert experience methods cannot exhaustively cover complex combinations of multi-level and multi-factor scenarios, and manually constructing chains is inefficient and highly subjective; third, related automatic generation or correlation analysis methods often lack strict constraints on the "phased progression" of the disaster-causing process, easily generating invalid chains such as backpropagation and cross-level jumps that do not conform to physical logic; finally, faced with massive amounts of generated candidate scenarios, there is a lack of an automated mechanism that can integrate multi-source evidence such as accident databases and technical specifications and use intelligent means for efficient and reliable screening, making it difficult to accurately identify the true critical disaster-causing path from the complex combinations.

[0004] Therefore, it is necessary to provide a method for identifying the critical chain of disaster-causing transmission of polar maritime traffic disturbances in order to solve the above problems. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, and medium for identifying critical chains of disaster-causing disturbance propagation in polar maritime traffic, thereby improving the accuracy, efficiency, and reliability of identifying critical disaster-causing chains in complex polar shipping scenarios.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for identifying the critical chain of disaster-causing propagation of polar maritime traffic disturbances, the method comprising: The disturbance propagation and disaster-causing process is divided into seven levels, and a set of inter-level connections that are only allowed to point from the current level to the next level or the level after that is defined to form a stage-progression constraint. This results in a disturbance propagation and disaster-causing chain network architecture with disaster-causing elements as nodes and allowed connections as directed edges. The seven levels are: disturbance source, disturbance absorption failure, system alarm, disturbance adaptation failure, danger and serious accident, disturbance recovery failure, and very serious accident. Under the constraints of the network architecture, the disaster-causing elements contained in each level are expanded layer by layer, and when expanding each level, paths that do not meet the stage advancement constraints are discarded to obtain a set of candidate scenarios; one candidate scenario corresponds to a complete disturbance propagation disaster-causing chain from the first level to the seventh level. Take any candidate scenario from the candidate scenario set as the current scenario, and for the current scenario: Generate a structured representation of the current scenario; the structured representation includes: a set of scenario elements, a set of directed propagation relationships, a set of evidence, and a set of target consequences; Based on preset prompt word templates, structured representations, and a preset reliability verification rule base, prompt words are generated and input into a large language model to perform multiple inferences to obtain corresponding inference results. The preset reliability verification rule base includes: phased advancement structure rules, polar water operation manual mapping consistency rules, and accident classification consistency rules. The inference results include at least a judgment label and confidence level. Based on the judgment label and confidence level, the reliability score of the current scenario is calculated, and the current scenario is matched with the reliability verification rule base to generate verification factors; Based on reliability scores and verification factors, all candidate scenarios are screened to obtain a set of reliable scenarios; Based on a set of reliable scenarios, the weight of each disaster-causing factor is calculated; the weight of any disaster-causing factor is determined based on the frequency of occurrence of the disaster-causing factor in the set of reliable scenarios, the reliability score of the scenario in which it is located, and whether there is supporting evidence. Based on the weights of each disaster-causing factor, the cumulative weight of each candidate disaster-causing chain is calculated, and the cumulative weights are sorted in descending order. The top N candidate disaster-causing chains are output as the key disaster-causing chains; N is a positive integer.

[0007] In one implementation, the disturbance propagation and disaster-causing process is divided into seven levels, and a set of inter-level connections that are only allowed to point from the current level to the next level or the level after that is defined, forming a phase advancement constraint. This constructs a disturbance propagation and disaster-causing chain network architecture with disaster-causing elements as nodes and allowed connections as directed edges, specifically including: Define a feature hierarchy mapping function and map each consistent feature to a unique level among the seven levels; Define the set of allowed connections between levels; the set of allowed connections contains only level pairs that point from the current level to the next level or the level after that. Based on the mapped disaster-causing elements and the set of allowed connections, a directed graph representation of the disturbance propagation disaster-causing chain network architecture is constructed.

[0008] In one embodiment, under the constraints of the network architecture, the disaster-causing elements contained in each layer are expanded layer by layer, and when expanding each layer, paths that do not satisfy the stage advancement constraints are discarded to obtain a set of candidate scenarios, specifically including: Starting from the first level, select one disaster-causing element in each level in turn, and form a directed connection with the disaster-causing element selected in the previous level in the currently expanded path; Verify whether the directed connection belongs to the set of allowed connections; If so, retain the current path and continue expanding to the next level; If not, discard the current path and select other disaster-causing elements at the current level to continue expanding until the seventh level is reached; Each complete path from level 1 to level 7 is treated as a candidate scenario, thus obtaining a set of candidate scenarios.

[0009] In one implementation, generating a structured representation of the current scenario specifically includes: Extract all disaster-causing factors contained in the current scenario to form a scenario element set; Extract the directed propagation relationships between the various disaster-causing factors from the current scenario to form a set of directed propagation relationships; Acquire evidence records that are relevant to the current situation to form an evidence set; the evidence records include: the element identifier corresponding to the evidence, the type of evidence source, and the confidence level of the evidence. Determine the consequences corresponding to the most serious accident level that the current scenario will ultimately lead to, and form a set of target consequences.

[0010] In one embodiment, prompt words are generated based on a preset prompt word template, a structured representation, and a preset reliability verification rule base, specifically including: Retrieve preset prompt word templates; the preset prompt word templates include: constraint condition field, output format field, rejection rule field, and reference basis field; Among them, the constraint field restricts the large language model to make judgments only based on the structured representation and the reliability verification rule base; the output format field is configured to output the judgment label, confidence level, citation basis, and conflict item; the rejection rule field is configured to output unreliable or undecidable when the information is contradictory or the evidence is insufficient; the citation basis field is configured to allow the citation basis to be mapped to the directed propagation relation set or the entry of the reliability verification rule base. Generate corresponding prompt words based on the preset prompt word template.

[0011] In one implementation, based on the judgment label and confidence level, a reliability score for the current scenario is calculated, and the current scenario is matched with a reliability verification rule base to generate a verification factor, specifically including: The consistency of labels is determined by statistical analysis of multiple inference results, and a consistency index is calculated. Calculate the average confidence level among multiple inference results, and use it as the mean confidence level; The reliability score of the current scenario is obtained by weighting and summing the consistency index with the mean confidence level. The set of directed propagation relationships and the set of evidence in the current scenario are matched with all the rules in the reliability verification rule base to obtain the matching results; If the matching result satisfies all the rules, a valid validation factor is generated; otherwise, an invalid validation factor is generated.

[0012] In one implementation, based on reliability scores and verification factors, all candidate scenarios are screened to obtain a set of reliable scenarios, specifically including: For each candidate scenario, determine whether the corresponding reliability score is higher than the preset reliability threshold and whether the corresponding verification factor is valid; If so, the corresponding candidate scenario is included in the reliable scenario set, and the reliable scenario set is finally obtained.

[0013] In one embodiment, the formula for calculating the weight of any disaster-causing element is: ; in, Disaster-causing factors The weights; Candidate scenarios The following disaster-causing factors The presence of an indicator variable, when a disaster-causing factor... Candidate scenario The set of scenario elements hour, ,otherwise ; Candidate scenarios The reliability score; Candidate scenarios The following disaster-causing factors Evidence factors; Candidate scenarios The structured representation; This is a set of reliable scenarios.

[0014] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation.

[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation.

[0016] According to the specific embodiments provided in this application, this application has the following technical effects: This application discloses a method, device, and medium for identifying critical chains of disaster-causing propagation in polar maritime traffic disturbances. First, by constructing a network architecture with phased advancement constraints, physical logic errors such as back propagation, same-layer propagation, and propagation across two or more layers are eliminated from the topological structure, significantly improving the compliance of the chain's physical mechanism and enhancing identification accuracy. Second, by adopting a layer-by-layer expansion method that discards paths that do not meet the phased advancement constraints at each expansion level, the invalid scenario space is significantly reduced, avoiding the computational explosion of full combination and subsequent screening. Simultaneously, automated batch reasoning using a large language model replaces manual review, improving screening efficiency. A dual screening mechanism of multiple reasoning and rule base cross-validation effectively suppresses the randomness and illusions of the large language model, enhancing the reliability of the screening results. Finally, when calculating the weights of disaster-causing elements, frequency of occurrence, reliability scores, and evidence support are integrated, ensuring that the identified critical disaster-causing chains possess both statistical significance and decision interpretability. This application improves the accuracy, efficiency, and reliability of critical disaster-causing chain identification through architectural constraints, pruning expansion, multiple reasoning, rule verification, and evidence fusion. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic flowchart of a method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation provided in an embodiment of this application; Figure 2 Another schematic diagram of the method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Guided by resilience theory, this application proposes a method for identifying critical chains of disaster-causing disturbance propagation in polar maritime traffic. The aim is to solve the problems of difficulty and low reliability in identifying critical disaster-causing chains under combined explosion scenarios in polar shipping by constructing a phased constrained network architecture and utilizing the semantic reasoning capabilities of large models combined with rule verification.

[0021] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] In one exemplary embodiment, such as Figure 1 and Figure 2 As shown, a method for identifying the critical chain of disaster-causing propagation of polar maritime traffic disturbances is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is described using a server as an example, and includes the following steps: Figure 1 As shown, where: Step S1: Divide the disturbance propagation disaster-causing process into seven levels, and define a set of inter-level connections that only allow connections from the current level to the next level or the level after that, forming a stage advancement constraint, and constructing a disturbance propagation disaster-causing chain network architecture with disaster-causing elements as nodes and allowed connections as directed edges; the seven levels are as follows: disturbance source, disturbance absorption failure, system alarm, disturbance adaptation failure, danger and serious accident, disturbance recovery failure, and very serious accident.

[0023] As an optional implementation method, step S1 specifically includes: Step S11: Define the feature hierarchy mapping function and map each consistent feature to a unique level among the seven levels.

[0024] Step S12, define the set of allowed connections between levels; the set of allowed connections only contains level pairs that point from the current level to the next level or the level below that.

[0025] Step S13: Based on the mapped disaster-causing elements and the set of allowed connections, construct a directed graph representation of the disturbance propagation disaster-causing chain network architecture.

[0026] Specifically, (1) a disaster-causing chain network architecture for the propagation of disturbances in polar maritime transportation is proposed and constructed. Based on resilience theory, starting from the disturbance source, the resilience of the polar maritime transportation system is deconstructed in three stages: disturbance absorption, disturbance adaptation, and disturbance recovery. A disaster-causing chain network architecture for the propagation of disturbances that meets the stage advancement constraints is constructed, specifically including: 1) Chain hierarchy division and element hierarchy mapping. For example... Figure 2 As shown, the process of disturbance propagation leading to disaster is divided into a set of seven progressively advancing levels. ,in, As a source of disturbance, For disturbance absorption failure, For system alarm, For perturbation adaptation failure, For dangerous situations and serious accidents, Disturbance recovery failed. This is a very serious accident.

[0027] Define feature hierarchy mapping function , so that any element All were mapped to a unique level .

[0028] 2) Definition of the allowed set of connections for phased advancement. Define the allowed set of connections between layers. , It should include at least the following hierarchical pairs: (1) And any directed connection ,satisfy and ,in express From the elements To elements The perturbation propagation relationship is established to form a phased advancement constraint, thereby preventing the perturbation propagation chain from reversing or looping.

[0029] 3) Output chain network architecture. Based on hierarchical partitioning and allowed connection sets. A directed graph representation of the network architecture for the chain of disturbance propagation leading to disaster. : (2) in, The set of elements involved in the chain of events leading to disaster through disturbance propagation. A set of directed connections that satisfy the phase advancement constraints; This serves as the foundational topological constraint for subsequent disaster-causing factor identification, candidate scenario generation, scenario reliability screening, and factor weight acquisition.

[0030] (2) Using literature review to investigate the disaster-causing factors of polar maritime traffic disturbance propagation Perform identification.

[0031] The sources of disturbance are defined as originating from the natural environment of the polar maritime transport system. Based on practical experience in polar waters and referencing the *Arctic Navigation Guide (Northeast Passage) 2014*, *Arctic Navigation Guide (Northeast Passage) 2021*, *Arctic Navigation Guide (Northwest Passage) 2015*, and the *Polar Vessel Guide*, six types of disturbance sources are identified: low temperature, strong winds and swells, poor visibility, severe ice conditions, obstructions, and high latitude effects. These are labeled as follows: .

[0032] The identification of the constituent elements of disturbance absorption failure, disturbance adaptation failure, and disturbance recovery failure involves transforming the safety measures in the Polar Water Operational Manual (PWOM) into safety measure failures, and then performing bottom-up causal reasoning to determine the functional causes of the safety measure failures and classify them.

[0033] Disturbance absorption failure includes eight elements: insufficient environmental information reception and perception, insufficient crew knowledge and experience, inadequate voyage planning, unavoidable exposure to severe weather systems, unavoidable restrictions on low-temperature operations, unavoidable restrictions on ice-covered areas, unavoidable restrictions on high-latitude operations, and unavoidable restrictions on operations in special areas. These are coded and marked as follows: .

[0034] Disturbance adaptation failures include six elements: insufficient supervision and inspection; failure to maintain equipment and system functionality; lack of caution and good seamanship; improper deployment of ice navigators and icebreaking aids; failure of emergency mitigation under adverse temperatures; failure of emergency mitigation in confined environments; and failure of emergency mitigation under adverse ice conditions. These are coded and labeled as follows: .

[0035] Disturbance recovery failures include nine elements: medical assistance limitations, buoyancy emergency response failure, corrective action and righting emergency failure, damage control emergency failure, oil and other pollutant spill emergency failure, fire emergency failure, life support and ship integrity failure, abandon ship and evacuation failure, and SAR limitations. These are coded and marked as follows: .

[0036] System alerts are a direct consequence of high-level exposure caused by absorption failure. It should be noted that the impact of early warning system and equipment reliability on alert functionality is not considered here. Assuming disturbance absorption failure, the system issues alerts regarding potential adverse effects from ice zones, low temperatures, high latitudes, severe weather, and navigational obstructions. To further refine the disturbance propagation chain, severe weather is categorized into three weather and sea states closely related to navigation safety: wind and waves, rainfall, and sea fog. Therefore, system alerts are defined as including seven elements: potential impact from wind and waves, potential impact from rainfall, potential impact from sea fog, potential impact from low temperatures, potential impact from sea ice, potential environmental limitations from high latitudes and remote locations, and potential navigational obstructions such as shoals and reefs. These are coded and labeled as follows. .

[0037] Ship incidents and serious accidents are classified according to the IHS Markit Maritime Accident Classification. Based on the accident's development and the International Maritime Organization's (IMO) recognition of loss of watertightness as a relevant event in accident investigations, loss of watertightness is added as a significant incident. This stage is ultimately subdivided into nine elements: illness / injury, grounding, capsizing, engine / ship damage, watertightness, pollution, contact, collision, and fire / explosion. .

[0038] Very serious accidents are defined according to the IMO Accident Investigation Code as having three elements: serious environmental damage, total loss of the vessel, and fatalities, and are coded as follows: .

[0039] Step S2: Under the constraints of the network architecture, the disaster-causing elements contained in each level are expanded layer by layer, and when expanding each level, paths that do not meet the stage advancement constraints are discarded to obtain a set of candidate scenarios; one candidate scenario corresponds to a complete disturbance propagation disaster-causing chain from the first level to the seventh level.

[0040] As an optional implementation method, step S2 specifically includes: Step S21: Starting from the first level, select one disaster-causing element in each level in sequence, and form a directed connection with the disaster-causing element selected in the previous level in the currently expanded path.

[0041] Step S22: Verify whether the directed connection belongs to the set of allowed connections.

[0042] Step S23: If yes, retain the current path and continue expanding to the next level.

[0043] Step S24: If not, discard the current path and select other disaster-causing elements at the current level to continue expanding until the seventh level is reached.

[0044] Step S25: Each complete path from the first level to the seventh level is taken as a candidate scenario, thus obtaining a set of candidate scenarios.

[0045] Specifically, a layer-by-layer expansion approach is used to construct the candidate scenario set. Based on the aforementioned network architecture... Phase-based advancement constraints, constructing a candidate scenario set , For the first One candidate scenario; This represents the index of candidate scenarios. Six × 8 × 7 × 6 × 9 × 9 × 3 types of candidate disturbance-causing scenarios are generated through hierarchical full-element combination, using a layer-by-layer expansion approach. to In the layer-by-layer combination generation, each time a layer is expanded, the edge direction and hierarchical order are checked according to the stage advancement constraints. Those that do not meet the requirements are immediately discarded, thereby greatly reducing the invalid scenario space at the generation source.

[0046] Step S3: Take any candidate scenario from the candidate scenario set as the current scenario, and for the current scenario: Generate a structured representation of the current scenario; the structured representation includes: a set of scenario elements, a set of directed propagation relationships, a set of evidence, and a set of target consequences.

[0047] As an optional implementation, step S3, generating a structured representation of the current scenario, specifically includes: Step S31: Extract all disaster-causing elements contained in the current scenario to form a scenario element set.

[0048] Step S32: Extract the directed propagation relationships between the disaster-causing elements from the current scenario to form a set of directed propagation relationships.

[0049] Step S33: Obtain evidence records related to the current situation to form an evidence set; the evidence records include: the element identifier corresponding to the evidence, the type of evidence source, and the confidence level of the evidence.

[0050] Step S34: Determine the consequences corresponding to the most serious accident level that the current scenario will ultimately lead to, and form a target consequence set.

[0051] Specifically, for any candidate scenario The generated structured representation is as follows: (3) in, It is the set of elements involved in the scenario (i.e., the scenario element set). In order to be in A set of directed propagation relations under constraints; For the collection of evidence; The set of target consequences, including the layer of very serious accidents. Three elements.

[0052] For each candidate scenario Constructing a set of evidence Collection of evidence It consists of several evidence records, each of which includes at least: the element identifier corresponding to the evidence. Types of evidence sources and the confidence level of the evidence , can be represented as: (4) Among them, the types of evidence sources It should include at least two types of evidence: accident database evidence and large model knowledge base evidence. Accident database evidence includes historical accident / hazard records and their structured fields. The structured fields should at least include event identifier, event description, and involved elements. The large model knowledge base generates search requests based on candidate scenarios through a network information retrieval interface to obtain external publicly available information result sets. The large language model then extracts evidence from the result sets to form structured evidence items. The structured evidence items should at least include source identifier, evidence content, and involved elements in the candidate scenarios.

[0053] Step S4: Based on the preset prompt word template, structured representation, and preset reliability verification rule base, generate prompt words, and input the prompt words into the large language model to perform multiple inferences to obtain the corresponding inference results; the preset reliability verification rule base includes: phase advancement structure rules, polar water operation manual mapping consistency rules, and accident classification consistency rules; the inference results include at least the judgment label and confidence level.

[0054] As an optional implementation, in step S4, prompt words are generated based on a preset prompt word template, a structured representation, and a preset reliability verification rule base, specifically including: Step S41: Obtain a preset prompt word template; the preset prompt word template includes: constraint condition field, output format field, rejection rule field, and reference basis field; Among them, the constraint field restricts the large language model to make judgments only based on the structured representation and the reliability verification rule base; the output format field is configured to output the judgment label, confidence level, citation basis, and conflict item; the rejection rule field is configured to output unreliable or undecidable when the information is contradictory or the evidence is insufficient; the citation basis field is configured to allow the citation basis to be mapped to the directed propagation relation set or the entry of the reliability verification rule base.

[0055] Step S42: Generate corresponding prompt words according to the preset prompt word template.

[0056] Specifically, the preset reliability verification rule base It should include at least: a phased advancement structure rule base. PWOM mapping consistency rule base and accident classification consistency rule base ,in: Used to verify any propagation connection in the scenario. Does it meet the stage advancement constraints? and If the conditions are not met, the scenario is deemed to have failed the validation. Used for verification , , Consistency with the PWOM safety measure failure reasoning It should at least include the mapping relationship between security measure failure clause number - functional cause classification - failure element code (in layman's terms, it is to first deduce the security measure requirements in PWOM in reverse: if these measures fail, what functional failure reasons will they manifest, and then classify these failure reasons into specific failure element codes in the network). Used for verification , Logical consistency within the framework of IHS accident classification and IMO accident investigation rules. At least the following executable rules should be included: 1) When the consequence is (In cases of severe environmental damage) (Pollution) should exist; otherwise, the scenario is considered to have an accident logic contradiction; 2) When the consequence is (When the ship is totaled) (Machinery damage / ship damage) or (Loss of watertight integrity) At least one of these must be present; otherwise, the scenario lacks the necessary chain of events to support the conclusion.

[0057] Furthermore, preset prompt word templates are invoked to generate prompt words for candidate scenarios (also called prompt input). The prompt word template includes: 1) a constraint field, which refers to limiting the context to only the set of scenario elements. Propagation Relationship Set Collection of evidence Target Consequence Set and rule base 1) Judgment; 2) Output format field, requiring the output to include: Judgment label (reliable / unreliable / undeterminable), confidence level (0~1), citation basis (which evidence or rules were cited), and conflict item (reasons for the judgment being unreliable / undeterminable); 3) Rejection rule field, setting the output to unreliable / undeterminable when there is contradiction or insufficient evidence; 4) Citation basis field, setting the citation to be mappable to... or rule base entries .

[0058] Furthermore, the prompt words Input a large language model and execute it independently under a fixed sampling configuration. The reasoning process generates a set of reasoning results. This is represented as follows, and the judgment label is extracted from each inference result set. and confidence level ,in .

[0059] (5) in, For the first The set of reasoning results corresponding to each candidate scenario; For the first The first reasoning under the next inference The reasoning results corresponding to each candidate scenario; For the first The first reasoning under the next inference The judgment label corresponding to each candidate scenario; For the first The first reasoning under the next inference The confidence level corresponding to each candidate scenario; This is the index for the number of inferences.

[0060] To ensure the reproducibility of the inference process, the fixed sampling configuration should include at least a temperature parameter, a cumulative probability truncation threshold, a maximum generation length, and a random seed, wherein: the temperature parameter is preferably selected as follows. The cumulative probability cutoff threshold is preferably selected. The maximum generation length is preferably selected. The random seed is fixed or selected from a preset seed list; It is a positive integer and preferred .

[0061] Step S5: Based on the judgment label and confidence level, calculate the reliability score of the current scenario, and match the current scenario with the reliability verification rule base to generate verification factors.

[0062] As an optional implementation, step S5 specifically includes: Step S51: Calculate the consistency index of the judgment labels in the results of multiple inferences.

[0063] Step S52: Calculate the average confidence level among multiple inference results, and use it as the mean confidence level.

[0064] Step S53: The consistency index and the mean confidence level are weighted and summed to obtain the reliability score of the current scenario.

[0065] Step S54: Match the set of directed propagation relationships and the set of evidence in the current scenario with all rules in the reliability verification rule base to obtain the matching result.

[0066] Step S55: If the matching result satisfies all rules, a valid validation factor is generated; otherwise, an invalid validation factor is generated.

[0067] Specifically, based on the set of reasoning results Calculate the reliability score of the candidate scenarios. The calculation formula is as follows: (6) in, For the first Consistency index for each candidate scenario For the first The mean confidence level of each candidate scenario. For preset weighting coefficients and . and The calculation is as follows: (7) (8) in, For candidate label values, Indicates the first Each candidate scenario corresponds to In this inference, the output label equals The number of times; For the first The total number of independent reasoning attempts for each candidate scenario; This means selecting the tag that appears most frequently among all tag categories; Preferred selection Increase when the overall quality of evidence is high. To emphasize consistency; reduce when the model output confidence is more reliable. When in When two or more labels appear with the highest frequency in the inference results, it is preferable to calculate the mean confidence score of the output corresponding to each tied label and determine the label with the higher mean confidence score as the consistency label; alternatively, to improve the robustness of the judgment, the final judgment label of the candidate scenario can be marked as "undecidable", and the consistency index can be set accordingly. The ratio of the highest number of occurrences in a given sequence to the total number of reasoning attempts is taken.

[0068] Furthermore, and Mapped to rule base Get the matching set And generate validation factors When there are violations of phase progression, mutual exclusion of side directions, or insufficient reference, take... Otherwise take Among them, for the current candidate scenario, from the overall rule base The rules that are relevant and match it are selected from the data: A step-by-step comparison method is used, firstly, by... match The first is the rule regarding "hierarchical advancement and allowing edges"; the second is, using... and the scenarios involved / / Node to match The "Terms - Functional Reasons - Node Coding" rule; thirdly, the use of scenarios... / match The key is to determine whether there are any violations of the rules, and based on this, to determine... The value of .

[0069] Step S6: Based on the reliability score and verification factor, all candidate scenarios are screened to obtain a set of reliable scenarios.

[0070] As an optional implementation, step S6 specifically includes: Step S61: For each candidate scenario, determine whether the corresponding reliability score is higher than the preset reliability threshold and whether the corresponding verification factor is valid.

[0071] Step S62: If yes, the corresponding candidate scenario is included in the reliable scenario set, and the reliable scenario set is finally obtained.

[0072] Specifically, based on a preset reliability threshold Filter the set of reliable scenarios It outputs a structured record for each reliable scenario, containing at least [number of data points]. Among them, the reliable scenario set satisfy: (9) in, Preferred .

[0073] By selecting scenarios that simultaneously meet the requirements of a reliability score higher than the threshold and a valid verification factor, a dual check of intelligent scoring and logical verification is achieved.

[0074] Step S7: Calculate the weight of each disaster-causing element based on the reliable scenario set; wherein, the weight of any disaster-causing element is determined based on the frequency of occurrence of the disaster-causing element in the reliable scenario set, the reliability score of the scenario in which it is located, and whether there is supporting evidence.

[0075] As an optional implementation method, the formula for calculating the weight of any disaster recovery element is: (10) in, Disaster-causing factors The weights; Candidate scenarios The following disaster-causing factors The presence of an indicator variable, when a disaster-causing factor... Candidate scenario The set of scenario elements hour, ,otherwise ; Candidate scenarios The reliability score; Candidate scenarios The following disaster-causing factors Evidence factors; Candidate scenarios The structured representation of .

[0076] Specifically, based on a reliable scenario set For each element Calculate weights The reliability is determined by the occurrence of a reliable scenario, the reliability of the evidence, and the supporting evidence. The formula is given in equation (10), where... And when hour ,otherwise ; As evidence factors, when elements Appear in the evidence collection China Times, ,otherwise ,in For preset coefficients and ; Preferred selection This is to enhance the weight differentiation of nodes with supporting evidence relative to nodes without supporting evidence.

[0077] Step S8: Based on the weights of each disaster-causing element, calculate the cumulative weight of each candidate disaster-causing chain, sort the cumulative weights in descending order, and output the top N candidate disaster-causing chains as key disaster-causing chains; N is a positive integer.

[0078] Specifically, based on the weights of each disaster-causing factor Calculate the first Candidate chain Cumulative weight As a basis for identifying the critical chain, it is represented as follows: (11) in, And satisfy ; Indicates the chain at the 1st Elements at each level The weights; The larger the value, the more likely the chain is composed of high-weight nodes, and therefore the more critical the chain. By accumulating the weights of candidate chains in descending order and outputting the set of critical disaster-causing chains, the transformation from qualitative screening to quantitative ranking is achieved.

[0079] Optionally, to avoid numerical underflow caused by multiplication, the logarithmic cumulative weight can be calculated in the logarithmic field: (12) in, Indicates the first The logarithmic cumulative weight of each chain, It is the natural logarithm function. When sorting by chain, it uses... Size substitution Compare them.

[0080] Beneficial effects: (1) High screening efficiency, effectively solving the problem of scenario combination explosion. Related technologies are often helpless in the face of multi-level and multi-element combinations, and manual screening is time-consuming and laborious. This application filters out a large number of invalid scenarios that do not conform to topological constraints at the generation stage by expanding the real-time verification mechanism layer by layer; and uses a large language model to perform automated batch reasoning screening, replacing manual review of each scenario, which greatly improves the efficiency of processing massive candidate scenarios and makes chain analysis of large-scale complex systems possible.

[0081] (2) High scenario recognition accuracy and strong chain mechanism logic. Related data-driven methods are prone to uncovering spurious correlations or reverse causal relationships. This application constructs a phased advancement network architecture, which enforces constraints on the direction and hierarchical order of disturbance propagation from the underlying logic; combined with the rule base cross-validation mechanism, it ensures that the retained scenarios not only conform to the statistical laws of data, but also strictly conform to the physical mechanisms and operational specifications of polar shipping such as PWOM, thereby significantly improving the accuracy and logical credibility of the recognition results.

[0082] (3) The results are highly reliable, with good noise resistance and reproducibility. The direct output of a large model may contain illusions or randomness. This application adopts a multiple independent inference strategy under a fixed sampling configuration, and comprehensively scores by calculating the consistency index and mean confidence, which effectively smooths out the random noise of a single inference. At the same time, the validation factor introduced in this point serves as a "hard constraint" to eliminate erroneous scenarios where the model scores high but violates the core rules, ensuring the robustness and reliability of the final results.

[0083] (4) The identification of critical links is evidentiarily interpretable and has high decision-making reference value. Most related ranking methods are based on a single frequency. When calculating the weights, this application not only considers the frequency of scenario occurrence, but also integrates the reliability score of the scenario itself and whether there is clear evidence to support it, such as accident records and regulatory provisions. This makes the identified critical links not only high-frequency, but also highly credible, providing a more persuasive and targeted decision-making basis for polar shipping safety management.

[0084] In one exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for identifying critical chains of disaster-causing polar maritime traffic disturbance propagation.

[0085] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements a method for identifying critical chains of disaster-causing propagation of polar maritime traffic disturbances.

[0086] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements a method for identifying critical chains of disaster-causing propagation of polar maritime traffic disturbances.

[0087] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 3As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for identifying the critical chain of disaster-causing propagation of polar maritime traffic disturbances.

[0088] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0089] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0090] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0091] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based maritime traffic safety and artificial intelligence logic devices, etc., and are not limited to these.

[0092] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying the critical chain of disaster-causing propagation of polar maritime traffic disturbances, characterized in that, The method includes: The disturbance propagation and disaster-causing process is divided into seven levels, and a set of inter-level connections that are only allowed from the current level to the next level or the level after that is defined, forming a stage-progression constraint. This constructs a disturbance propagation and disaster-causing chain network architecture with disaster-causing elements as nodes and allowed connections as directed edges. Specifically, this includes: defining an element-level mapping function and mapping each disaster-causing element to a unique level among the seven levels; defining a set of allowed connections between levels; the allowed connection set only contains level pairs that point from the current level to the next level or the level after that; and constructing a directed graph representation of the disturbance propagation and disaster-causing chain network architecture based on the mapped disaster-causing elements and the allowed connection set. The seven levels are, in order: disturbance source, disturbance absorption failure, system alarm, disturbance adaptation failure, hazard and severe accident, disturbance recovery failure, and extremely severe accident. Under the constraints of the network architecture, the disaster-causing elements contained in each level are expanded layer by layer, and when expanding each level, paths that do not meet the stage advancement constraints are discarded to obtain a set of candidate scenarios; one candidate scenario corresponds to a complete disturbance propagation disaster-causing chain from the first level to the seventh level. Take any candidate scenario from the candidate scenario set as the current scenario, and for the current scenario: Generate a structured representation of the current scenario; the structured representation includes: a set of scenario elements, a set of directed propagation relationships, a set of evidence, and a set of target consequences; Based on preset prompt word templates, structured representations, and a preset reliability verification rule base, prompt words are generated and input into a large language model to perform multiple inferences to obtain corresponding inference results. The preset reliability verification rule base includes: phased advancement structure rules, polar water operation manual mapping consistency rules, and accident classification consistency rules. The inference results include at least a judgment label and confidence level. Based on the judgment label and confidence level, the reliability score of the current scenario is calculated, and the current scenario is matched with the reliability verification rule base to generate verification factors; Based on reliability scores and verification factors, all candidate scenarios are screened to obtain a set of reliable scenarios; Based on a set of reliable scenarios, the weight of each disaster-causing factor is calculated; the weight of any disaster-causing factor is determined based on the frequency of occurrence of the disaster-causing factor in the set of reliable scenarios, the reliability score of the scenario in which it is located, and whether there is supporting evidence. Based on the weights of each disaster-causing factor, the cumulative weight of each candidate disaster-causing chain is calculated, and the cumulative weights are sorted in descending order. The top N candidate disaster-causing chains are output as the key disaster-causing chains; N is a positive integer.

2. The method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation according to claim 1, characterized in that, Under the constraints of the network architecture, the disaster-causing elements contained in each level are expanded layer by layer. When expanding each level, paths that do not satisfy the stage progression constraints are discarded, resulting in a set of candidate scenarios, specifically including: Starting from the first level, select one disaster-causing element in each level in turn, and form a directed connection with the disaster-causing element selected in the previous level in the currently expanded path; Verify whether the directed connection belongs to the set of allowed connections; If so, retain the current path and continue expanding to the next level; If not, discard the current path and select other disaster-causing elements at the current level to continue expanding until the seventh level is reached; Each complete path from level 1 to level 7 is treated as a candidate scenario, thus obtaining a set of candidate scenarios.

3. The method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation according to claim 1, characterized in that, Generate a structured representation of the current context, specifically including: Extract all disaster-causing factors contained in the current scenario to form a scenario element set; Extract the directed propagation relationships between the various disaster-causing factors from the current scenario to form a set of directed propagation relationships; Acquire evidence records that are relevant to the current situation to form an evidence set; the evidence records include: the element identifier corresponding to the evidence, the type of evidence source, and the confidence level of the evidence. Determine the consequences corresponding to the most serious accident level that the current scenario will ultimately lead to, and form a set of target consequences.

4. The method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation according to claim 1, characterized in that, Based on preset prompt word templates, structured representations, and a preset reliability verification rule base, prompt words are generated, specifically including: Retrieve preset prompt word templates; the preset prompt word templates include: constraint condition field, output format field, rejection rule field, and reference basis field; Among them, the constraint field restricts the large language model to make judgments only based on the structured representation and the reliability verification rule base; the output format field is configured to output the judgment label, confidence level, citation basis, and conflict item; the rejection rule field is configured to output unreliable or undecidable when the information is contradictory or the evidence is insufficient; the citation basis field is configured to allow the citation basis to be mapped to the directed propagation relation set or the entry of the reliability verification rule base. Generate corresponding prompt words based on the preset prompt word template.

5. The method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation according to claim 1, characterized in that, Based on the judgment label and confidence level, a reliability score for the current scenario is calculated, and the current scenario is matched with a reliability verification rule base to generate verification factors, specifically including: The consistency of labels is determined by statistical analysis of multiple inference results, and a consistency index is calculated. Calculate the average confidence level among multiple inference results, and use it as the mean confidence level; The reliability score of the current scenario is obtained by weighting and summing the consistency index with the mean confidence level. The set of directed propagation relationships and the set of evidence in the current scenario are matched with all the rules in the reliability verification rule base to obtain the matching results; If the matching result satisfies all the rules, a valid validation factor is generated; otherwise, an invalid validation factor is generated.

6. The method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation according to claim 1, characterized in that, Based on reliability scores and validation factors, all candidate scenarios are screened to obtain a set of reliable scenarios, which specifically includes: For each candidate scenario, determine whether the corresponding reliability score is higher than the preset reliability threshold and whether the corresponding verification factor is valid; If so, the corresponding candidate scenario is included in the reliable scenario set, and the reliable scenario set is finally obtained.

7. The method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation according to claim 1, characterized in that, The formula for calculating the weight of any disaster-causing factor is: ； in, Disaster-causing factors The weights; Candidate scenarios The following disaster-causing factors The presence of an indicator variable, when a disaster-causing factor... Candidate scenario The set of scenario elements hour, ,otherwise ; Candidate scenarios The reliability score; Candidate scenarios The following disaster-causing factors Evidence factors; Candidate scenarios The structured representation; This is a set of reliable scenarios.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method for identifying the critical chain of polar maritime traffic disturbance propagation causing disasters, as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for identifying the critical chain of disaster-causing polar maritime traffic disturbance propagation as described in any one of claims 1-7.

Images