A ship offshore emergency repair decision method and system
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN UNIV OF TECH
- Filing Date
- 2026-04-16
- Publication Date
- 2026-07-21
Smart Images
Figure CN122434489A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ship emergency repair technology, and in particular to a ship maritime emergency repair decision-making method and system. Background Technology
[0002] Modern ocean-going vessels rely heavily on equipment reliability. When a ship is sailing in the deep sea, if core components such as the main engine and steering gear fail, the crew must take action in a very short time. At the same time, they face extreme external environments such as high sea states and low visibility, as well as internal constraints such as a lack of spare parts and isolated technical support.
[0003] Existing emergency maintenance technologies suffer from key deficiencies: maintenance decisions lack environmental adaptability assessments, focusing solely on fault repair while ignoring the constraints of external environmental factors such as wind, waves, and currents on maintenance operations, rendering plans unenforceable in adverse sea conditions; ship equipment technical manuals, maintenance procedures, and historical failure case studies are stored in a vast amount of unstructured PDF documents, making manual retrieval extremely time-consuming, and fault diagnosis heavily reliant on personnel experience; ship inventory databases have high query thresholds and low information exchange efficiency, requiring specialized operation and making it impossible to quickly obtain inventory status using natural language, thus failing to meet the requirements for emergency response speed; environmental data, inventory data, document data, and other multi-source heterogeneous data are independent, forming information silos, lacking a unified collaborative decision-making mechanism, requiring decision-makers to manually summarize information from multiple sources, which is prone to oversights and errors in judgment. Summary of the Invention
[0004] The main objective of this invention is to achieve a shift from qualitative to quantitative maintenance decision-making, global optimal resource scheduling, and adaptive system optimization through multi-agent collaboration, integrating simulation, resource scheduling, and knowledge retrieval.
[0005] The technical solution adopted in this invention is: a method for decision-making in emergency repairs at sea, comprising: Acquire multi-source information related to the vessel to be repaired and its fault; the multi-source information includes the vessel's motion status, environmental monitoring data, natural language fault description, and historical status data. Based on the analysis of the multi-source information, the maintenance intention is determined, and a task sequence is generated. The parallel task sequence includes: performing Monte Carlo simulation of the sea conditions of the vessel to be repaired and outputting the operational safety probability; using a contract network protocol mechanism to perform dynamic game calculations based on a pre-built material database and generate a resource allocation plan; and retrieving and extracting maintenance guidance information that matches the fault of the vessel to be repaired from pre-stored technical documents. The pre-built multi-dimensional decision evaluation model is used to integrate and calculate the operation safety probability, resource allocation plan and maintenance guidance information to generate a maintenance decision plan.
[0006] According to the above technical solution, the method for analyzing maintenance intentions based on the multi-source information includes: A time-series backtracking mechanism is established by leveraging the long memory characteristics of large language models to associate the faults of ships to be repaired with historical status data; By analyzing the causal logic between the historical state data and the faults of the vessel to be repaired using the thought chain reasoning technique, the natural language fault description is decomposed into maintenance intentions with temporal dependencies.
[0007] According to the above technical solution, the Monte Carlo simulation of the sea state of the vessel to be repaired, and the output of the operational safety probability, includes: The pre-built ship hydrodynamic model is used to perform Monte Carlo simulation of the sea state in which the ship is located, and a dual convergence control strategy that limits accuracy and timeliness is adopted to control the simulation process. The simulated ship motion attitude is compared with the preset safe attitude limits to determine whether a single simulation is effective. Based on all valid simulation results, the probability of job safety is calculated.
[0008] According to the above technical solution, the step of using the Contract Network protocol mechanism for dynamic game calculation to generate a resource allocation scheme includes: Broadcast material demand information to the vessel to be repaired and at least one accompanying node, wherein the accompanying node refers to a cooperating vessel or shore-based support unit used to provide material support to the vessel to be repaired. Dynamic game calculations are performed based on a comprehensive cost model, which is the sum of transportation costs and waiting costs. The transportation cost is determined based on real-time sailing distance, energy consumption cost per unit mile, and fixed operating costs; the waiting cost is determined based on estimated delivery time, the ship's unit time benchmark operating loss value, and an urgency weighting coefficient based on the fault level.
[0009] According to the above technical solution, maintenance guidance information matching the fault of the vessel to be repaired is retrieved from the pre-stored technical documents, specifically including: Based on the retrieval enhancement generation framework, retrieval is performed by calculating the semantic similarity between technical document fragments and the description of the ship's faults to be repaired; During the retrieval process, the device model identifier contained in the metadata of the document fragment is simultaneously verified to match the device corresponding to the current fault. Based on the dual verification results of semantic similarity and equipment model identification, the maintenance guidance information is located and extracted.
[0010] According to the above technical solution, the step of using a pre-built multi-dimensional decision evaluation model for fusion calculation includes: The multidimensional decision evaluation model is used to calculate the confidence score of the maintenance plan, which is a weighted sum of the urgency of the operation, the adaptability of the environment, and the resource matching degree. Wherein, the urgency of the operation is a value obtained based on the criticality level mapping of the fault; the environmental adaptability corresponds to the safety probability of the operation; and the resource matching degree is a reverse normalized index obtained based on the comprehensive cost calculation of the resource allocation scheme.
[0011] According to the above technical solution, in the multidimensional decision evaluation model, the weight coefficients of each dimension are dynamically adjusted based on a closed-loop feedback mechanism, and the dynamic adjustment includes: After executing the maintenance task based on the maintenance decision plan, the actual maintenance time is obtained; Calculate the deviation between the actual repair time and the predicted time in the repair decision plan; If the deviation exceeds a preset threshold, a weight adjustment mechanism is triggered to update the weight coefficients corresponding to the task urgency, environmental adaptability, and resource matching degree, respectively.
[0012] Another aspect of the present invention provides a shipboard emergency repair decision-making system, comprising: The fault information collection module is used to acquire multi-source information related to the vessel to be repaired and its faults; the multi-source information includes the vessel's motion status, environmental monitoring data, natural language fault descriptions, and historical status data. The intent analysis and task planning module is used to analyze maintenance intents based on the multi-source information and generate task sequences. The task execution module is used to perform the following tasks in parallel: perform Monte Carlo simulation of the sea conditions of the vessel to be repaired and output the operational safety probability; based on the pre-built material database, use the contract network protocol mechanism to perform dynamic game calculations and generate a resource allocation plan; and retrieve and extract maintenance guidance information that matches the fault of the vessel to be repaired from the pre-stored technical documents. The maintenance decision-making module is used to integrate and calculate the operation safety probability, resource allocation plan and maintenance guidance information using a pre-built multi-dimensional decision evaluation model to generate a maintenance decision-making plan.
[0013] Another aspect of the present invention provides a computer storage medium storing a computer program executable by a processor, the computer program performing the above-described shipboard emergency repair decision-making method.
[0014] Another aspect of the present invention 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 steps of the above-described shipboard emergency repair decision-making method.
[0015] The beneficial effects of this invention are as follows: By acquiring multi-source information related to the malfunction of the vessel to be repaired, analyzing the repair intention, and generating an evaluation task execution sequence, the causes of the malfunction can be accurately located, the task execution logic can be standardized, and the bias of human judgment and process chaos can be reduced; Monte Carlo simulation of sea conditions can be carried out to output the operational safety probability, quantitatively evaluate the feasibility of repair operations in harsh environments, and significantly improve the environmental adaptability and operational safety of decision-making; a contract network protocol mechanism can be used to perform dynamic game calculations to generate resource allocation schemes, achieve global optimal scheduling of materials, and shorten the time for resource location and replenishment; matching maintenance guidance information can be retrieved and extracted from the technical document vector database, quickly obtaining accurate handling basis from massive unstructured documents and reducing the dependence on personnel experience; a multi-dimensional decision evaluation model can be used to integrate and calculate the operational safety probability, resource allocation scheme, and maintenance guidance information, break down the barriers of multi-source heterogeneous data, form a comprehensive and feasible maintenance decision scheme, and comprehensively improve the response efficiency, execution reliability, and decision-making scientificity of ship maritime emergency repair.
[0016] Furthermore, a time-series backtracking mechanism is established by adopting the long memory characteristics of large language models to accurately associate the current fault with historical state data. The causal logic between historical data and the current fault is analyzed through the thinking chain reasoning technology, and the fuzzy natural language fault description is decomposed into a clear maintenance intention with time-series dependence, thereby improving the accuracy of fault location and the rationality of task planning, and reducing human understanding bias.
[0017] Furthermore, Monte Carlo simulations are conducted using a ship hydrodynamic model, and the simulation process is constrained by a dual convergence control strategy that emphasizes both accuracy and timeliness. This ensures computational reliability while meeting the requirements for rapid emergency response. The ship's motion attitude is compared with the preset safe attitude limits to determine the valid sample, thereby achieving a quantitative assessment of the feasibility of sea state maintenance and significantly improving the safety and feasibility of maintenance decisions in harsh environments.
[0018] Furthermore, the system broadcasts material requirements to the ships awaiting repair and surrounding nodes, conducts dynamic game theory based on a comprehensive cost model that combines transportation costs and waiting costs, and accurately calculates costs by combining real-time sailing distance, energy consumption costs, fixed costs, delivery time, and a weighted coefficient for the urgency of the fault. It automatically selects the resource allocation path with the lowest cost and best timeliness, achieving global optimal scheduling of materials across nodes.
[0019] Furthermore, based on the retrieval enhancement generation framework, document fragments are quickly filtered by semantic similarity, and equipment model identification is verified simultaneously for double verification. This allows for accurate matching of maintenance guidance information corresponding to faults in a massive amount of unstructured technical documents, significantly improving the efficiency and accuracy of knowledge extraction and reducing the time cost of manually consulting manuals.
[0020] Furthermore, a multi-dimensional decision evaluation model that uses a weighted summation of operational urgency, environmental adaptability, and resource matching is adopted to calculate the confidence score, thereby achieving efficient integration of multi-source information and optimal solution selection.
[0021] Furthermore, by comparing the deviation between the actual maintenance time and the predicted time through a closed-loop feedback mechanism, the weight coefficients are automatically dynamically adjusted when the threshold is exceeded, so that the multi-dimensional decision evaluation model can continuously adapt to the actual operating characteristics of the ship and has the ability to learn and iterate. Attached Figure Description
[0022] Figure 1 This is a flowchart of the ship's maritime emergency repair decision-making method according to an embodiment of the present invention; Figure 2 This is a flowchart of another ship emergency repair decision-making method according to an embodiment of the present invention; Figure 3 This is a structural diagram of the ship's maritime emergency repair decision-making system according to an embodiment of the present invention; Figure 4 This is a multi-agent collaboration diagram of the ship maritime emergency repair decision-making system according to an embodiment of the present invention. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0024] Example 1 This embodiment provides a multi-agent-based decision-making method for ship emergency repair at sea, as shown in Figure 1. This method is based on multi-source information fusion, intelligent collaboration, and quantitative evaluation, and is used to achieve automated and precise decision-making for emergency repairs at sea. Specifically, it includes the following steps: T1. Acquire multi-source information related to the fault of the vessel to be repaired. This information specifically includes the vessel's motion status (such as six-degree-of-freedom attitude) collected by shipborne sensors, real-time environmental monitoring data (such as wind speed and current speed), natural language fault descriptions input by the crew, and historical status data recorded by the vessel's systems.
[0025] T2. Based on the multi-source information, perform in-depth analysis of the maintenance intention and generate an execution sequence containing multiple specialized assessment tasks.
[0026] Furthermore, the key to intent analysis lies in leveraging the long memory characteristics of large language models such as DeepSeek-V3-0324 to establish a temporal backtracking mechanism. This mechanism links the current fault phenomenon with historical alarms, parameter changes, and other state data to trace the cause of the fault. Simultaneously, through thought chain reasoning technology, the causal logic between historical and current data is analyzed, breaking down and refining the crew's vague verbal descriptions into clear, specific maintenance intentions with sequential dependencies between steps.
[0027] T3. The system will execute various evaluation tasks in parallel according to the above-planned execution sequence.
[0028] Furthermore, it mainly includes three aspects: A pre-built ship hydrodynamic model is invoked to perform Monte Carlo simulations of current and forecast sea conditions to assess operational risks. A dual convergence control strategy that balances accuracy and response speed is employed to manage the simulation process. The validity of each simulation is determined by comparing the ship's motion attitude obtained with preset safe attitude limits (such as maximum roll and pitch angles). Finally, based on the results of all valid simulations, a quantified operational safety probability is calculated and output as the proportion of safe samples to the total number of simulations.
[0029] Based on the clearly defined maintenance intention, a contract network protocol mechanism is used for resource allocation calculation. Material requirements are broadcast to the vessel requiring maintenance and at least one accompanying node (i.e., a nearby supporting vessel or shore-based support unit). A dynamic game is then played based on a comprehensive cost model, defined as the sum of transportation costs and waiting costs. Transportation costs are determined by real-time sailing distance, energy consumption per unit mile of the transport vehicle, and fixed operating expenses. Waiting costs are determined by the estimated delivery time, the vessel's unit-time baseline operational loss, and an urgency-weighted coefficient set according to the fault level. This process seeks the globally optimal solution between rapid and economical options.
[0030] Precise maintenance guidance information is retrieved and extracted from a pre-stored technical document vector database. This is achieved through a retrieval enhancement generation framework, which includes preliminary screening by calculating the semantic similarity between document fragments and fault descriptions, while simultaneously verifying whether the device model identifier in the document fragment metadata is a complete match. Based on the results of this dual semantic and entity verification, key maintenance steps or diagnostic suggestions are located and extracted.
[0031] T4. Using a pre-built multi-dimensional decision evaluation model, the above-mentioned parallel obtained operation safety probability, resource allocation plan and maintenance guidance information are integrated and calculated to generate the final maintenance decision plan.
[0032] Furthermore, the model quantifies the evaluation of each potential maintenance path by calculating a confidence score, which is a weighted sum of three factors: operational urgency, environmental adaptability, and resource matching. Operational urgency is a value directly mapped from the criticality level of the faulty component; environmental adaptability corresponds to the previously calculated operational safety probability; and resource matching is an indicator obtained by inversely normalizing the overall cost of resource allocation schemes. The model compares the confidence scores of each scheme and selects the scheme with the highest score as the output.
[0033] Furthermore, to achieve continuous system optimization, a closed-loop feedback mechanism is used to dynamically adjust the weight coefficients corresponding to each dimension in the multi-dimensional decision evaluation model. Specifically, after each maintenance task is actually completed, the actual maintenance time is obtained, and the deviation between this time and the original predicted time in the decision plan is calculated. If this deviation exceeds a preset threshold, the weight adjustment mechanism is triggered, automatically updating the weight coefficients corresponding to the task urgency, environmental adaptability, and resource matching degree, thereby making the model's evaluation more accurate in the future.
[0034] This embodiment also provides a shipboard emergency repair decision-making system, corresponding to the above-described decision-making method, including: The fault information collection module is used to acquire multi-source information related to the fault of the vessel to be repaired; the multi-source information includes the vessel's motion status, environmental monitoring data, natural language fault description, and historical status data.
[0035] The intent analysis and task planning module is used to parse the maintenance intent based on the multi-source information, generate a time-series execution sequence containing multiple evaluation tasks, and schedule the execution of the task sequence.
[0036] The task execution module is used to perform Monte Carlo simulation of the sea conditions of the vessel to be repaired and output the operational safety probability; based on the pre-built material database, it uses the contract network protocol mechanism to complete dynamic game calculation and generate resource allocation plan; and it retrieves maintenance guidance information matching the current fault from the pre-stored technical document vector database through double verification.
[0037] The maintenance decision generation module is used to perform weighted fusion calculations on the probability of work safety, resource allocation plan and maintenance guidance information using a pre-built multi-dimensional decision evaluation model, to select the optimal path and generate the final maintenance decision.
[0038] This embodiment also provides a computer storage medium storing a computer program that can be executed by a processor, the computer program executing the above-described ship emergency repair decision-making method at sea.
[0039] This embodiment also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described ship emergency repair decision-making method.
[0040] Example 2 Based on Example 1, this example provides another method for decision-making in emergency repairs at sea, the process of which is as follows: Figure 2 As shown, the steps include: S1. Acquire multi-source information related to ship malfunctions, including real-time time-series data of the ship's six degrees of freedom motion attitude collected by sensors, environmental monitoring data, natural language descriptions of malfunctions by crew members, and historical status data extracted from the ship's system logs.
[0041] S2. The intelligent agent manager performs in-depth processing on the multi-source information to generate maintenance intentions, specifically including: using the long memory characteristics of the large language model to perform time-series backtracking to associate historical alarms with current faults; analyzing the causal chain through thought chain reasoning to generate clear maintenance intentions; and planning a sub-task execution sequence with strict parameter dependencies accordingly.
[0042] S3, plan the execution sequence of subtasks.
[0043] S4. Schedule multiple agents to perform special evaluation tasks in parallel: Simulation agent: Calls a pre-set ship hydrodynamic model and performs multiple Monte Carlo simulations based on the current wave spectrum.
[0044] To resolve the conflict between computational complexity and real-time emergency response, this intelligent agent employs a dual control strategy focusing on accuracy and timeliness. This strategy includes: real-time monitoring of the statistical variance of the simulation output and automatically terminating the computation process when its stability reaches a predetermined threshold; simultaneously, setting a hard maximum computation time threshold, which immediately terminates the simulation regardless of whether the iteration is complete when the computation time reaches this threshold, outputting statistical results based on the completed iterations, thus ensuring that the decision-making process is not blocked by excessive computational delays.
[0045] Based on this, complex dynamic feedback is transformed into quantified operational safety probabilities through multi-dimensional attitude constraint logic: a set of safety limits (including the maximum roll angle) is preset for specific operations. Maximum pitch angle and maximum vertical velocity In a single simulation, if the ship's motion state throughout the entire time domain never exceeds any of the above limits (i.e., and and If a sample is found to be a valid safe sample, then it is recorded as a valid safe sample. This is the ship's current roll angle. This is the ship's current pitch angle. This represents the ship's current vertical speed.
[0046] Ultimately, the probability of safe operation It can be calculated using the following formula:
[0047] in, To satisfy the above constraints on the number of safe samples, This represents the total number of effective simulations actually completed. This metric provides quantifiable probabilistic support for helicopter hovering or small boat approach.
[0048] The intelligent agent for material inquiry automatically broadcasts information to the vessel requiring repair and surrounding nodes using a contract network protocol. It utilizes formulas... Dynamically play the game of (the sum of transportation costs and waiting costs) to lock in the material supply chain with the best cost and fastest delivery.
[0049] The costs in the formula are calculated using the following logic: Transportation costs ( (): Represents the direct cost of the physical transfer of materials. The calculation formula is:
[0050] in, This refers to the real-time navigation distance between the vessel awaiting repair and the supply node. The energy cost per unit mile for the selected transport vehicle (such as a helicopter or transport boat). Fixed operating costs for the release and lifting of materials from the warehouse.
[0051] Waiting cost ( ): This represents the risk of performance loss due to maintenance delays, and the calculation formula is as follows:
[0052] in, For estimated delivery time, This represents the baseline operating loss value for a ship per unit of time. An urgency-weighted coefficient dynamically generated based on the fault level (e.g., during a propulsion system failure). Take the higher value; this is generally for lighting malfunctions. Take the lower value). By introducing The system can automatically calculate the overall cost between "expensive but fast helicopter airlift" and "cheap but slow small boat transfer" based on the urgency of the mission. The minimum value is obtained to achieve global optimality in resource scheduling.
[0053] The document retrieval agent operates based on the Retrieval Enhancement Generation (RAG) framework. This agent accesses a vector database to locate key segments of technical documents related to the fault. For the selection of "key segments," a dual "semantic-entity" verification standard is employed: First, the cosine similarity between the document slice vector and the fault description vector is calculated, and semantically relevant content with a similarity higher than a preset threshold (0.85 in this embodiment) is selected. Simultaneously, the device model identifier recorded in the metadata of the selected slice is forcibly verified to ensure complete consistency with the model of the current fault object. Based on this dual verification standard, multiple sets of potential "diagnostic suggestions" and "operational procedures" candidate sets are directly extracted from massive amounts of unstructured documents, eliminating the lengthy manual review and screening process.
[0054] T4. The agent manager collects intermediate results (including material matching status, simulation success rate, etc.) from the various functional agents. Based on the multiple candidate solutions provided by the document retrieval agent, it constructs multiple potential maintenance paths and uses a multi-dimensional emergency decision evaluation model to calculate the confidence score P for each path.
[0055] in, Reflecting the urgency of the task, Represents environmental adaptability. For resource matching, the system will horizontally compare the potential paths. The solution with the highest score that exceeds the safety threshold is selected as the final execution instruction. When logical conflicts arise, such as "the solution is feasible but the environment is high-risk," the system generates a waiting window suggestion or a downgrade plan, which is then determined by the system. Value-triggered hierarchical resolution: When the score is extremely high, high-priority preemption is activated, and a contingency plan is forcibly generated; when the score is in the middle range, a balanced allocation suggestion is generated.
[0056] To ensure that the decision-making model's calculations have physical meaning and are feasible, the above indicators are quantified and defined according to the following rules: Task urgency ( ): Mapping is done by the agent manager based on the criticality level of the faulty object. A level-one fault is set (e.g., host shutdown). Level 2 fault Level 3 fault This indicator does not rely on underlying sub-agents and directly reflects the global impact weight of maintenance tasks on the ship's survivability.
[0057] Environmental adaptability ): The normalized operational safety probability output by the simulated intelligent agent, i.e. This value It directly represents the probability of safe operation under the current sea state.
[0058] Resource matching degree ( ): The comprehensive cost output by the corresponding material query intelligent agent. The inverse normalization index. The calculation formula is:
[0059] in, This is the preset baseline cost for this type of task. This value reflects the superiority or inferiority of the current material plan compared to the historical average.
[0060] Regarding weighting coefficients Configuration: The system employs an "expert knowledge preset strategy" for cold start during initial runtime (this is configured in this embodiment to reflect the "safety first" principle). Based on the system's closed-loop feedback mechanism, these weights are dynamically and iteratively adjusted according to the deviations in the actual effects of historical decisions.
[0061] T5. The system performs closed-loop learning to optimize decision-making: After each task is completed, the execution results are automatically collected. The system sets a time deviation threshold. (In this embodiment, it is set to predict the time consumption) If the actual repair time is... With system prediction time The relative deviation exceeds this threshold (i.e. This immediately triggers the backpropagation mechanism. At this point, if an error is detected in the original decision weight configuration, the backpropagation algorithm is automatically initiated to dynamically adjust the weight coefficients in the evaluation model. This achieves a monotonically increasing performance of the system.
[0062] This embodiment also provides a shipboard emergency repair decision-making system, the structure of which is as follows: Figure 3 As shown, it includes: The user interaction layer, serving as the interface between the system and the crew, is used to receive fault information described in natural language and output the final maintenance decision.
[0063] The intelligent agent management layer, as the core of system control and coordination, is connected below the user interaction layer. It is used to receive input information from the user interaction layer, parse maintenance intentions and plan tasks, and generate execution sequences; to schedule and coordinate tasks of lower-level functional intelligent agents; and to aggregate and fuse the results returned by lower-level functional intelligent agents.
[0064] The functional agent layer, as the core execution unit of the system, is located below the agent management layer and logically contains three parallel cooperating agents: The material query intelligent agent, based on the tasks issued by the intelligent agent management layer, accesses the material database and uses the contract network protocol and comprehensive cost model to calculate the resource allocation plan; The simulated intelligent agent, based on the tasks issued by the intelligent agent management layer, calls the ship hydrodynamic model to perform Monte Carlo simulation and calculates the operational safety probability. The document retrieval agent is configured to retrieve and extract maintenance guidance information from the technical document database based on tasks issued by the agent management layer and the retrieval enhancement generation framework.
[0065] Data exchange and command transmission between different levels are achieved through defined interfaces, jointly realizing the aforementioned ship emergency repair decision-making method at sea.
[0066] The functional intelligent agent layer also incorporates a dynamic context interaction module, which unifies objective environmental data (sensor time-series data) and subjective interaction data (crew commands) into a generalized "context." On one hand, the module relies on DeepSeek... V3 The long memory characteristics of the 0324 large language model are used to build a time-series backtracking mechanism to continuously track and associate the ship's historical status. This allows for the accurate identification of fault causes from subtle vibration alarms from hours ago. On the other hand, the CoT (Cooperation in the Thinking) reasoning technology is used to analyze the causal logic between historical data and current faults, breaking down the crew's ambiguous instructions into a sequence of sub-tasks with strict temporal dependencies.
[0067] This system constructs a hierarchical and collaborative software architecture, aiming to automate the entire process from fault information input to intelligent maintenance decision generation.
[0068] Furthermore, its information flow and processing logic are as follows: Figure 4 As shown: First, at the data input layer, the system receives descriptions of fault phenomena in natural language from crew members via user terminals. Simultaneously, it collects real-time environmental monitoring data such as the ship's motion attitude, wind speed, and current velocity through sensor interfaces, providing comprehensive multi-source information input for decision-making. This information is then aggregated at the core processing and scheduling layer, the intelligent agent management layer, located at the center of the architecture. This layer, acting as the system's brain, incorporates several key modules, including intent recognition, dynamic memory, conflict resolution, and multi-dimensional evaluation models. It is responsible for parsing fuzzy maintenance instructions, correlating historical data to trace the cause of the fault, and breaking down and planning the core maintenance intent into a series of logically dependent specific task sequences.
[0069] Subsequently, the intelligent agent management layer synchronously schedules the three specialized functional agents below it to execute in parallel according to the planned task sequence. The material query agent receives bidding instructions, accesses the database using the Contract Network protocol and Text2SQL technology, and performs dynamic game calculations through a model that integrates transportation costs and waiting costs to generate the optimal resource allocation plan. The simulation agent, based on the acquired real-time sea state parameters, calls the ship hydrodynamic model to perform thousands of Monte Carlo simulations, and adopts a dual convergence strategy of "accuracy-timeliness" to ensure response speed, ultimately outputting a quantified operational safety probability. At the same time, the document retrieval agent, based on fault characteristics and using the Retrieval Enhancement Generation (RAG) framework, accurately retrieves and extracts relevant maintenance procedures and historical cases from a vectorized technical document database through a dual standard of semantic similarity matching and mandatory equipment model verification.
[0070] Finally, the agent manager aggregates the specific evaluation results returned by the three agents, performs fusion calculations using the built-in multi-dimensional decision evaluation model, generates the comprehensive maintenance decision scheme with the highest confidence level, and outputs it to the user.
[0071] also, Figure 4 The document also describes a closed-loop path from task execution results back to the core layer, representing the system's self-optimization capability: by comparing the deviation between actual maintenance results and predicted values, the parameters of the evaluation model are dynamically adjusted, making its decisions increasingly accurate with the accumulation of cases. The entire architecture, through close collaboration among modules at all levels, achieves intelligent coordination of multi-source information, providing rapid, safe, and reliable decision support for emergency maritime maintenance of ships.
[0072] In summary, this invention provides a method and system for decision-making in marine emergency repairs, which can improve the response efficiency, execution reliability, and scientific nature of marine emergency repairs.
[0073] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.
[0074] The order of the steps in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0075] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.
Claims
1. A method for decision-making in emergency repairs at sea, characterized in that, include: Acquire multi-source information related to the vessel to be repaired and its fault; the multi-source information includes the vessel's motion status, environmental monitoring data, natural language fault description, and historical status data. Based on the analysis of the multi-source information, the maintenance intention is determined, and a task sequence is generated. The parallel task sequence includes: performing Monte Carlo simulation of the sea conditions of the vessel to be repaired and outputting the operational safety probability; using a contract network protocol mechanism to perform dynamic game calculations based on a pre-built material database and generate a resource allocation plan; and retrieving and extracting maintenance guidance information that matches the fault of the vessel to be repaired from pre-stored technical documents. The pre-built multi-dimensional decision evaluation model is used to integrate and calculate the operation safety probability, resource allocation plan and maintenance guidance information to generate a maintenance decision plan.
2. The ship maritime emergency repair decision-making method according to claim 1, characterized in that, The method for analyzing maintenance intentions based on the aforementioned multi-source information includes: A time-series backtracking mechanism is established by leveraging the long memory characteristics of large language models to associate the faults of ships to be repaired with historical status data; By analyzing the causal logic between the historical state data and the faults of the vessel to be repaired using the thought chain reasoning technique, the natural language fault description is decomposed into maintenance intentions with temporal dependencies.
3. The ship maritime emergency repair decision-making method according to claim 1, characterized in that, The Monte Carlo simulation of the sea conditions of the vessel to be repaired outputs the operational safety probability, including: The pre-built ship hydrodynamic model is used to perform Monte Carlo simulation of the sea state in which the ship is located, and a dual convergence control strategy that limits accuracy and timeliness is adopted to control the simulation process. The simulated ship motion attitude is compared with the preset safe attitude limits to determine whether a single simulation is effective. Based on all valid simulation results, the probability of job safety is calculated.
4. The ship maritime emergency repair decision-making method according to claim 1, characterized in that, The method of using the Contract Network protocol mechanism for dynamic game calculation to generate resource allocation schemes includes: Broadcast material demand information to the vessel to be repaired and at least one accompanying node, wherein the accompanying node refers to a cooperating vessel or shore-based support unit used to provide material support to the vessel to be repaired. Dynamic game calculations are performed based on a comprehensive cost model, which is the sum of transportation costs and waiting costs. The transportation cost is determined based on real-time sailing distance, energy consumption cost per unit mile, and fixed operating costs; the waiting cost is determined based on estimated delivery time, the ship's unit time benchmark operating loss value, and an urgency weighting coefficient based on the fault level.
5. The ship maritime emergency repair decision-making method according to claim 1, characterized in that, Retrieve and extract maintenance guidance information matching the fault of the vessel to be repaired from pre-stored technical documents, specifically including: Based on the retrieval enhancement generation framework, retrieval is performed by calculating the semantic similarity between technical document fragments and the description of the ship's faults to be repaired; During the retrieval process, the device model identifier contained in the metadata of the document fragment is simultaneously verified to match the device corresponding to the current fault. Based on the dual verification results of semantic similarity and equipment model identification, the maintenance guidance information is located and extracted.
6. The ship maritime emergency repair decision-making method according to claim 1, characterized in that, The fusion calculation is performed using a pre-built multidimensional decision evaluation model, including: The multidimensional decision evaluation model is used to calculate the confidence score of the maintenance plan, which is a weighted sum of the urgency of the operation, the adaptability of the environment, and the resource matching degree. Wherein, the urgency of the operation is a value obtained based on the criticality level mapping of the fault; the environmental adaptability corresponds to the safety probability of the operation; and the resource matching degree is a reverse normalized index obtained based on the comprehensive cost calculation of the resource allocation scheme.
7. The ship maritime emergency repair decision-making method according to claim 6, characterized in that, In the multidimensional decision evaluation model, the weight coefficients of each dimension are dynamically adjusted based on a closed-loop feedback mechanism, and the dynamic adjustment includes: After executing the maintenance task based on the maintenance decision plan, the actual maintenance time is obtained; Calculate the deviation between the actual repair time and the predicted time in the repair decision plan; If the deviation exceeds a preset threshold, a weight adjustment mechanism is triggered to update the weight coefficients corresponding to the task urgency, environmental adaptability, and resource matching degree, respectively.
8. A shipboard emergency repair decision-making system, characterized in that, include: The fault information collection module is used to acquire multi-source information related to the vessel to be repaired and its faults; the multi-source information includes the vessel's motion status, environmental monitoring data, natural language fault descriptions, and historical status data. The intent analysis and task planning module is used to analyze maintenance intents based on the multi-source information and generate task sequences. The task execution module is used to perform the following tasks in parallel: perform Monte Carlo simulation of the sea conditions of the vessel to be repaired and output the operational safety probability; based on the pre-built material database, use the contract network protocol mechanism to perform dynamic game calculations and generate a resource allocation plan; and retrieve and extract maintenance guidance information that matches the fault of the vessel to be repaired from the pre-stored technical documents. The maintenance decision-making module is used to integrate and calculate the operation safety probability, resource allocation plan and maintenance guidance information using a pre-built multi-dimensional decision evaluation model to generate a maintenance decision-making plan.
9. A computer storage medium, characterized in that, It contains a computer program that can be executed by a processor, which performs the ship maritime emergency repair decision-making method as described in any one of claims 1-7.
10. A computer device, characterized in that, It includes 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 steps of the ship maritime emergency repair decision-making method according to any one of claims 1-7.