A ship declaration collaborative processing system and method based on multi-user roles

Abstract

The application discloses a ship declaration collaborative processing system and method based on multi-user roles, relates to the technical field of ship declaration, and comprises the following steps: instantiating a digital twin software object for a new ship declaration request; extracting core semantic assertions, metadata and semantic potential parameters, collectively encapsulating the same into semantic cell data units, traversing all semantic cells in the digital twin, extracting core assertions of the semantic cells and nodes in a domain knowledge graph for semantic matching, finding a corresponding knowledge graph standard concept node, determining a force rule, placing all semantic cells into a semantic field, and assigning each cell a charge attribute obtained by mapping from the knowledge graph according to the core assertion; and the system runs a physical simulation cycle until the system reaches a steady state or a quasi-steady state, the total potential energy of the entire cell system is calculated, the total potential energy of the system is compared with a preset threshold interval through a normalization function, and a decision branch is automatically triggered.

Description

Technical Field

[0001] This invention relates to the field of ship declaration technology, specifically to a collaborative processing system and method for ship declaration based on multiple user roles. Background Technology

[0002] Traditional data validation primarily focuses on compliance within a single modality or simple formatting, such as whether fields are empty or values ​​are within range. It lacks the ability to correlate, compare, and identify contradictions in the deep semantics of cross-modal information. Multimodal data exhibits significant heterogeneity; the data structures and semantic spaces of text, images, and time-series trajectories are completely different, making unified semantic understanding and alignment technically challenging. Existing solutions often process data in isolation: OCR recognition of MSDS text followed by keyword matching, or simple boundary checks on AIS trajectories. These methods fail to grasp the logical conflict between "non-toxic" and "harmful to the aquatic environment" within the context of environmental regulations. The processing flow is fragmented, the validation granularity is coarse, and it heavily relies on manual review to uncover deeper inconsistencies.

[0003] Chinese invention application CN114020248A discloses a method and system for constructing a digital twin platform for the shipbuilding industry, including: S1, constructing a digital twin service platform; S2, publishing a digital twin on the digital twin service platform that meets the requirements of actual engineering applications in the shipbuilding industry; S3, registering the service functions of the digital twin published on the digital twin service platform; S4, obtaining the physical entity's requirement description for the digital twin; S5, performing similarity matching between the digital twin requirement description and the registered digital twin service functions, and jumping to step S6 when matching occurs; S6, performing adaptive modifications on the digital twin selected and matched by the user for the actual working environment of the physical entity, and jumping to step S7; S7, performing functional verification on the determined digital twin to determine whether it meets the actual engineering application requirements, and jumping to step S8 if it meets the requirements; S8, applying the digital twin that meets the actual engineering requirements in engineering.

[0004] In the above invention applications, the platform-based construction method enables the management and service provision of digital twins on the one hand, and provides common support for the construction of digital twins in the shipbuilding industry on the other hand, reducing the difficulty of digital twin development. However, it cannot detect minor semantic contradictions, which may lead to dangerous goods being misjudged as ordinary goods, causing serious environmental leakage accidents, huge economic compensation and irreversible damage to brand reputation.

[0005] Therefore, the present invention provides a collaborative processing system and method for ship declaration based on multiple user roles. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a collaborative processing system and method for ship declarations based on multiple user roles. By integrating scattered multimodal data into a unified digital twin, it solves the data silo problem in traditional systems. It precisely maps semantic assertions of cells to knowledge graph concepts and determines the rules of interaction. The system transforms previously isolated, static data points into "physical entities" with clear interactive behaviors in a semantic field. This allows subsequent consistency verification to move beyond simple True / False judgments based on symbolic logic, enabling more nuanced and reliable results through simulating dynamic interactions. When the system detects contradictions, its root cause can be clearly traced. This completely avoids the subjective judgment differences and uncertainties caused by factors such as personal experience and fatigue in traditional audits. Decisions are based on unified, objective numerical standards, ensuring the consistency and fairness of processing results. System behavior is no longer fixed but possesses strategic flexibility. For example, thresholds can be raised during special periods to adopt stricter audit standards, allowing the system to flexibly adapt to constantly changing business environments, thereby solving the technical problems described in the background art.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution: a collaborative processing method for ship declaration based on multiple user roles, comprising the following steps:

[0010] Instantiate a digital twin of the new ship declaration request. Software objects, extracting core semantic assertions, metadata, and semantic potential parameters. Together they are encapsulated into semantic cells Data units will generate semantic cells Inject digital twins sequentially ;

[0011] System traversal of digital twins All semantic cells The core assertions are extracted and semantically matched with nodes in the domain knowledge graph to find the corresponding standard concept nodes in the knowledge graph, determine the force rules, and provide a basis for the current digital twin. Create an empty semantic field instance. Load all force rules into this field instance. middle;

[0012] Will All semantic cells within Inserting semantic field Each cell is assigned a charge attribute mapped from the knowledge graph based on its core assertion; the system runs a physical simulation loop until the system reaches a steady state or quasi-steady state, calculates the total potential energy of the entire cell system, calculates the semantic stability score of the total potential energy through a normalization function, compares it with a preset threshold range, and automatically triggers the corresponding decision branch.

[0013] The process involves instantiating a uniquely identified digital twin of the new ship declaration request. Software objects, specifically:

[0014] The system receives the application request through the interface, parses out the key information, calls the digital twin management service, executes the new application digital twin operation, and creates a new object of that class in memory. The system automatically invokes this newly effective object. The initialization method; within this method, the following is executed programmatically:

[0015] Assign a system-generated unique value to the twin identifier attribute, assign the application business number attribute to the received application ID, set the current status attribute to the initial state, set the timestamp attribute to the current system time, initialize the semantic cell container to an empty data structure, and initialize the derived attribute set to an empty key-value pair set.

[0016] The process involves parsing text descriptions, document images, and time-series trajectory multimodal data to extract core semantic assertions, metadata, and semantic potential parameters. Specifically:

[0017] The text parsing engine processes the declaration description text and outputs structured assertions and confidence levels; the document parsing engine analyzes the MSDS image, identifies key sections through layout recognition, and extracts assertions and source locations; the trajectory parsing engine processes AIS time-series data and identifies behavioral pattern assertions.

[0018] Structured assertions, assertions, and behavioral pattern assertions are collectively defined as core assertions;

[0019] The core assertions and their metadata are input into the potential energy parameter prediction network, which dynamically calculates the physical simulation parameters based on the assertion semantics and the current declaration context. Specifically:

[0020] Assign charge values ​​with opposite signs to core assertions that have opposing relationships; and assign mass values ​​based on the authority of the data source, outputting a value containing both charge and mass parameters. gather.

[0021] The extraction of core semantic assertions, metadata, and semantic potential parameters Together they are encapsulated into a standard semantic cell. The data unit is specifically:

[0022] When any parsing engine in the preceding sequence successfully produces a core semantic assertion, the encapsulation process is immediately started. The system calls a globally unique identifier generation service to generate a unique string sequence for each semantic cell, which is denoted as the cell identifier.

[0023] The system instantiates a predefined software object in memory; this object acts as a data container and has several predefined fields for storing specific information, including core assertion fields, potential energy parameter fields, and metadata fields.

[0024] The system performs an integrity check on the fully populated data container to ensure that all necessary fields have been correctly assigned values; subsequently, the container is marked as encapsulated, and a semantic cell is finally generated. Immediately available data entities.

[0025] The process involves extracting the core assertions and performing semantic matching with nodes in the domain knowledge graph. For each semantic cell's core assertion, a corresponding standard concept node from the knowledge graph is found. Specifically:

[0026] System traversal of digital twins All semantic cells Extract its core assertions and perform semantic matching with nodes in the domain knowledge graph. For each semantic cell, find a corresponding standard concept node in the knowledge graph for the core assertion.

[0027] The system converts the core assertions and concept node names in the knowledge graph into semantic vectors, and then finds the closest graph node in the vector space, which is a semantic search based on vector similarity.

[0028] The specific rules for determining the force are as follows:

[0029] Using the set of concept nodes obtained from the mapping as the query condition, the knowledge graph is traversed to retrieve all relation edges with force rule labels that connect these nodes;

[0030] The system transforms these retrieved relational rules, described in natural language or logical labels, into quantized parameters that the physics engine can understand.

[0031] Each cell is specifically assigned a charge attribute derived from the knowledge graph mapping based on its core assertion, as follows:

[0032] The system maps core assertions to standard concept nodes in the domain knowledge graph. The system traverses the knowledge graph, retrieves all relation edges marked with interaction rules that start or end with the concept node, and outputs one or more qualitative logic rules.

[0033] The system converts logical rules into charge symbols based on a predefined logical-physical mapping table and assigns a reference strength value to different types of logical rules; thus, the charge symbols and reference values ​​are obtained.

[0034] System read The mass parameter is used to weight the reference value, and the final charge value = reference value × mass.

[0035] The system runs a simplified physics simulation loop as follows:

[0036] In an abstract two-dimensional or three-dimensional simulation space, the system analyzes the core assertions of all semantic cells, evaluates the initial semantic association strength between each pair of cells through a lightweight semantic similarity calculation model, and assigns an initial coordinate to each cell.

[0037] The system runs a simplified physical simulation loop until it reaches a steady state or quasi-steady state, specifically:

[0038] The system All semantic cells within the cell are paired up, and for each pair of cells... and Perform the following calculations:

[0039] Obtaining cells and charge value and and their current distance in the simulation space. ;

[0040] According to the simplified form of Coulomb's law = k * ( * ) / ( ^2) Calculate the scalar electrostatic force between them. ;

[0041] Where k is a constant coefficient, if and If they have the same sign, then A positive value indicates repulsive force; if and If the signs are different, then A negative value indicates attractiveness;

[0042] Calculated scalar force According to cells and The direction of the line connecting them can be decomposed into the forces acting on them. and Force vectors in opposite directions are added to the resultant force vector acting on each cell.

[0043] Based on the resultant force vector acting on each cell in the current iteration, its acceleration and velocity are calculated according to the simplified Newton's laws of motion, and its coordinates in the simulation space are updated accordingly.

[0044] After each iteration, the system calculates the total kinetic energy of the entire cellular system and the sum of the magnitudes of the net external force vectors acting on all cells; when both of these values ​​decrease and stabilize below a preset threshold, the system is considered to have reached a steady state or quasi-steady state.

[0045] A collaborative processing system for ship declaration based on multiple user roles, comprising:

[0046] The semantic cell generation module instantiates a digital twin from the new ship declaration request. Software objects, extracting core semantic assertions, metadata, and semantic potential parameters. Together they are encapsulated into semantic cells Data units will generate semantic cells Inject digital twins sequentially ;

[0047] The cell rule analysis module systematically traverses the digital twin. All semantic cells The core assertions are extracted and semantically matched with nodes in the domain knowledge graph to find the corresponding standard concept nodes in the knowledge graph, determine the force rules, and provide a basis for the current digital twin. Create an empty semantic field instance. Load all force rules into this field instance. middle;

[0048] The cell system simulation module will All semantic cells within Inserting semantic field Each cell is assigned a charge attribute mapped from the knowledge graph based on its core assertion; the system runs a physical simulation loop until the system reaches a steady state or quasi-steady state, calculates the total potential energy of the entire cell system, calculates the semantic stability score of the total potential energy through a normalization function, compares it with a preset threshold range, and automatically triggers the corresponding decision branch.

[0049] (III) Beneficial Effects

[0050] This invention provides a collaborative processing system and method for ship declaration based on multiple user roles, which has the following beneficial effects:

[0051] 1. By integrating scattered multimodal data into a unified digital twin, the problem of data silos in traditional systems is solved. This provides a complete and structured operational object for subsequent consistency verification, avoiding the complexity of cross-data source queries. Through semantic cell encapsulation, raw data is transformed into semantically rich intelligent units. The introduction of semantic potential parameters endows data with dynamic interaction capabilities, surpassing traditional static data storage. The digital twin updates its state attributes in real time as cells are injected, and the system has adaptive cognitive capabilities, reflecting changes in the integrity and consistency of declared information. Standardized semantic cell interfaces facilitate the rapid access of new data modalities. Modular design supports functional expansion without affecting the core architecture. Object-level data transfer reduces serialization / deserialization overhead. Pre-structured data provides ready-to-use input for subsequent analysis, significantly reducing computational latency. Dynamic state attributes provide real-time indicators for risk assessment. Rich contextual information supports more accurate consistency judgments. The encapsulation mechanism ensures data integrity and reduces the impact of external interference. The state evolution logic provides anomaly detection and fault tolerance capabilities.

[0052] 2. By precisely mapping cellular semantic assertions to knowledge graph concepts and determining the rules of interaction, the system transforms previously isolated, static data points into "physical entities" with explicit interactive behaviors (attraction / repulsion) within the semantic field. This allows subsequent consistency verification to move beyond simple True / False judgments based on symbolic logic, enabling more nuanced and reliable results through simulating dynamic interactions. When the system detects a contradiction, its root cause can be clearly traced.

[0053] 3. By using physical simulation, complex multimodal information contradictions are transformed into a single, quantifiable "semantic stability score." This completely avoids the subjective judgment differences and uncertainties caused by factors such as personal experience and fatigue in traditional review processes. Decisions are based on unified, objective numerical standards, ensuring the consistency and fairness of the processing results. The system's behavior is no longer fixed but possesses strategic flexibility. For example, during special periods, thresholds can be raised to adopt stricter review standards, enabling the system to flexibly adapt to constantly changing business environments. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating a collaborative processing method for ship declaration based on multiple user roles according to the present invention.

[0055] Figure 2 This is a schematic diagram of the structure of a ship declaration collaborative processing system based on multiple user roles according to the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see Figure 1 This invention provides a collaborative processing method for ship declaration based on multiple user roles, comprising the following steps:

[0058] Step 1: Instantiate a uniquely identified digital twin of the new ship declaration request. A software object is generated, and initial state attributes and basic metadata are pre-defined for this digital twin software object. Text descriptions, document images, and time-series trajectory multimodal data are parsed to extract core semantic assertions, metadata, and semantic potential parameters. Together they are encapsulated into a standard semantic cell. Data units will generate semantic cells Inject digital twins sequentially , According to the injection and It dynamically updates its internal state attribute set.

[0059] Step one includes the following steps:

[0060] Step 101: Instantiate a uniquely identified digital twin of the new ship declaration request. The software object is pre-defined, and its initial state attributes and basic metadata are pre-defined.

[0061] Define a software class during the system design phase. This class serves as a template for all declared digital twins, including key attributes and key methods.

[0062] Key attributes include twin identifier, application business number, timestamp, current status, semantic cell container, and derived attribute set.

[0063] The twin identifier is a unique identifier, a globally unique string generated by the system using a specific algorithm, such as a distributed ID generated using the Snowflake algorithm or a standard UUID. This is the unique identity credential of $D_t$ throughout the entire system. The application business number is the associated application business ID, used for precise association and reverse tracing with the original application record in the business database. The timestamp records the twin's creation time, last status update time, etc., for lifecycle tracking. The current status is a dynamically changing attribute used to identify and track the twin's lifecycle stage. Its value comes from a predefined enumeration set, such as: "Building", "Data Injection", "Semantic Field Simulation", "Verification Passed", "Alarm", etc. The semantic cell container is a collection of semantic cells used to store and manage all semantic cells generated and injected in subsequent steps. The derived attribute set is a key-value pair set used to store high-level, comprehensive attributes calculated through comprehensive analysis of all semantic cells. For example: {"Environmental Risk Index": 0.85, "Overall Consistency Score": 0.45}.

[0064] Key methods include initialization methods, semantic cell injection methods, state update methods, and derived attribute computation methods.

[0065] The initialization method is used to set the initial state and basic metadata; the inject semantic cell method is used to receive a semantic cell object, store it in the semantic cell container, and trigger the update of the internal state; the update state method is used to change the value of the current state; the calculate derived attribute method calculates or updates the derived attribute set based on the current state of all cells in the semantic cell container.

[0066] The instantiation process is as follows:

[0067] When the system's access layer receives a new ship declaration request, the following instantiation process will be automatically triggered:

[0068] The system receives the application request through the interface, parses out key information such as the application ID submitted by the user, calls the digital twin management service, executes the new application digital twin operation, and creates a new object of that class in memory. The system automatically invokes this newly effective object. The initialization method.

[0069] Within this method, the following is executed programmatically:

[0070] Assign a system-generated unique value to the twin's identifier attribute, assign the application business number attribute to the received application ID, set the current status attribute to the initial state (e.g., "Under Construction"), set the timestamp attribute to the current system time, initialize the semantic cell container to an empty data structure, and initialize the derived attribute set to an empty set of key-value pairs. At this point, a live, uniquely identified twin is formed. The software object has been successfully created in the system's memory.

[0071] It should be noted that: newly instantiated The object will be immediately serialized and persistently saved to the database. Meanwhile, the The object's reference is published as a message event to the internal message bus or directly passed to the next processing module, thereby driving the entire verification process. Afterward, the... The entire lifecycle of an object, including state transitions, cell injection, and simulation result writing, will be uniformly managed and maintained by the system until the final completion of the application process, such as verification and archiving.

[0072] Step 102: Utilize a domain-pre-trained large model to parse multimodal data such as text descriptions, document images, and time-series trajectories; extract the core semantic assertions, metadata, and a set of dynamically adjustable semantic potential parameters. Together they are encapsulated into a standard semantic cell. Data unit.

[0073] The specific steps for processing input data from different modalities using three independent domain-specific pre-trained models are as follows:

[0074] The text parsing engine processes the declaration description text and outputs structured assertions and confidence levels such as "Toxicity: Non-toxic"; the document parsing engine analyzes images such as MSDS, identifies key sections through layout recognition, such as "Hazard Statement", and extracts assertions such as "Environmental Hazard: Harmful to the aquatic environment" and their source locations; the trajectory parsing engine processes AIS time-series data and identifies behavioral pattern assertions such as "Boarding in sensitive areas".

[0075] The text parsing engine transforms unstructured declaration text into structured semantic assertions through an ordered, three-tiered processing architecture. This architecture consists of: a domain knowledge injection layer, a semantic parsing and annotation layer, and a structured assertion generation layer.

[0076] The goal of the domain knowledge injection layer is to create a text understanding model with expertise in the maritime domain.

[0077] The system collects and preprocesses authoritative texts in the maritime field to form a training dataset. This dataset includes: international maritime conventions and national standards, such as SOLAS and MARPOL; dangerous goods lists and classification manuals, such as the full text of the International Maritime Dangerous Goods Code; chemical safety data sheets (MSDS) documents containing a large amount of compliant text; and a historical declaration text database, consisting of accumulated, manually reviewed ship declaration descriptions.

[0078] First, the general language model is further trained unsupervised on the aforementioned training dataset. In this stage, the model learns and masters professional vocabulary, terminology collocations, and syntactic structures in the maritime field through a cloze test task, transferring its prior knowledge to a specific domain.

[0079] Next, we prepare a dataset annotated by domain experts, where key entities and relationships in the text have been identified. On this dataset, we conduct supervised training on the model through a joint entity and relationship extraction task. The training objective of this task is to enable the model to simultaneously recognize named entities in text and determine the specific semantic relationships that exist between these entities.

[0080] The semantic parsing and annotation layer utilizes the domain-adaptive model output from the first level to perform deep semantic analysis on the input new application text.

[0081] The engine receives the declaration text and inputs it into a finely tuned model. The model first converts the text into an internal context vector representation. Then, it performs a pre-trained joint extraction task, traversing the text sequence and identifying and labeling all entities belonging to predefined categories. For example, it labels "chemical products" as a <material> entity and "toxicity" as an <attribute> entity. While identifying entities, the model analyzes the semantic relationships between them in parallel. It determines that there is a [descriptive] relationship between "toxicity" and "non-toxic," thus forming an "attribute-value" pair.

[0082] The structured assertion generation layer transforms the analysis results from the second level into a system-defined standardized output format.

[0083] The system captures the attribute-value pairs identified by the model. For the "toxicity-non-toxicity" pair, the system maps and formats it into a standard structured assertion object: {attribute: "toxicity", value: "non-toxicity"}.

[0084] When classifying each extracted assertion, the model outputs a Softmax probability distribution. The confidence level is determined by the probability value of the target class in this distribution. For example, if the model classifies the relationship between "toxicity" and "non-toxicity" as [description], and the probability of this classification is 0.95, then the confidence level of this assertion is 0.95. This value is attached to the assertion object as metadata.

[0085] The text parsing engine ultimately outputs one or more core semantic assertions, such as {attribute: "toxicity", value: "non-toxic", confidence: 0.95}, along with their text source information, and delivers them to the subsequent semantic cell encapsulation process.

[0086] The document parsing engine transforms visual and textual information from document images into structured semantic assertions through an ordered, three-tiered processing architecture. This architecture consists of: a document structure parsing layer, a key information localization and extraction layer, and an assertion synthesis and confidence fusion layer.

[0087] The goal of the document structure parsing layer is to decode the input document image into a structured document object containing visual, textual, and layout information.

[0088] The system receives document images and uses a multimodal model, such as LayoutLMv3, pre-trained on a large-scale document image dataset to extract pixel-level features of the image through a convolutional neural network. It perceives the visual layout of lines, table boxes, seals, and text lines. The model outputs a document object model, in which each identified text fragment is associated with its precise coordinates on the page and its visual context.

[0089] The key information location and extraction layer uses the document object model produced by the first level to locate the preset key columns and extract their text content.

[0090] The input is a Document Object Model (DOM) and a predefined list of key headings, such as ["Hazard Statement", "Ingredients / Composition Information", "First Aid Measures"]. Within the DOM, text blocks semantically highly similar to the key heading names are searched; for example, text titled "Hazard Statement" or "Hazard Identification" is found. Once the heading title is located, the model infers the boundaries of the content area belonging to that heading based on page layout, such as chapter indentation and text blocks in the lower area. One or more target content areas are identified, each corresponding to a key heading. The system focuses on each identified target content area, extracting all text sequences within the bounding box of that area from the DOM. The clean, heading-specific text content is obtained. For example, for the "Hazard Statement" heading, the output is the content string: "May have long-term harmful effects on the aquatic environment."

[0091] The assertion synthesis and confidence fusion layer transforms the column content extracted in the second level into standardized assertions defined by the system and calculates the overall confidence level.

[0092] The system maps column names to assertion attributes and uses the extracted column content strings as assertion values ​​to generate structured core semantic assertions, such as: {Attribute: "Environmental Hazard", Value: "May cause long-term harmful effects on the aquatic environment"}. The system records which logical part of the document the assertion originates from (e.g., "Chapter 2.3") and / or its physical coordinate range in the original image.

[0093] The final confidence score is a fusion value, calculated by weighting the confidence scores of category detection and text recognition. Category detection confidence score is the model's confidence score in locating key category titles and content areas. Text recognition confidence score is the average confidence score of the OCR engine in recognizing text within the content area.

[0094] The trajectory parsing engine transforms raw, low-level AIS temporal location data into high-level, semantically meaningful behavioral pattern assertions through an ordered, three-layer processing architecture. This architecture consists of: a trajectory cleansing and semantic enhancement layer, a multi-scale behavioral feature extraction layer, and a pattern recognition and assertion generation layer.

[0095] The goal of the trajectory cleansing and semantic enhancement layer is to clean the original AIS data and integrate it with the geographic information system to give the trajectory a preliminary semantic label.

[0096] The system receives AIS data streams from ships over a continuous period of time. Data packets typically include fields such as timestamps, latitude and longitude, speed over land, and heading over land. Anomalies that clearly do not conform to the ship's motion patterns are identified and eliminated using a discrimination method based on speed / acceleration thresholds or a Kalman filter algorithm. Data gaps caused by brief signal loss are appropriately filled using spline interpolation or motion model-based prediction methods to form a continuous and smooth trajectory sequence.

[0097] Each cleaned trajectory point is spatially overlaid and analyzed against a semantic layer in an electronic nautical chart or geographic information system. The system pre-defines various types of polygonal regions, such as "ecological protection zones," "designated anchorages," and "port operation areas." Each trajectory point is then labeled with a semantic tag indicating its location within the ecological protection zone. For example, one point might be marked as located in an ecological protection zone, while another might be marked as located in open water.

[0098] The multi-scale behavioral feature extraction layer calculates a series of quantitative behavioral feature indicators based on the semantic trajectory produced by the first level.

[0099] For each predefined sensitive area, the system calculates entry and departure timestamps based on continuous changes in the area's semantic labels. The total duration of stay within the area is the difference between the departure and entry times. The centroid of the stopping point is the average latitude and longitude of the ship's position across all points within the area.

[0100] At both global and regional scales, time-series statistics on airspeed are calculated, such as mean, variance, and the percentage of time spent below the "stopping threshold." The rate of change of course is calculated by determining the difference in course between adjacent track points to identify behaviors such as hesitancy and turning. Track curvature analyzes the degree of curvature of the track to distinguish between straight-line navigation and detours.

[0101] The pattern recognition and assertion generation layer utilizes the feature sequences extracted in the second stage to perform advanced behavioral pattern recognition through a machine learning model, and generates the final structured assertions.

[0102] The multi-dimensional features extracted in the second stage, such as dwell time, speed sequence, and heading changes, are organized into a feature sequence in chronological order and input into a temporal classification model. This model (such as LSTM, Transformer, or temporal convolutional network) has been trained on a large amount of manually labeled trajectory data and can identify complex behavioral patterns such as "stopping," "loitering," "illegal intrusion," and "normal navigation." The model analyzes the feature sequence and outputs a behavior classification result and its corresponding confidence score.

[0103] The system fuses the model's classification results with the semantic labels of the regions obtained at the first level to generate the final behavioral pattern assertion. For example: {Behavior: "Parking", Location Type: "Sensitive Area", Object: "Ecological Reserve", Confidence: 0.92}.

[0104] The core assertions and their metadata are input into the potential energy parameter prediction network, which dynamically calculates the physical simulation parameters based on the assertion semantics and the current declaration context. Specifically:

[0105] Assign opposite charges to assertions with conflicting relationships, such as assigning a positive charge to "non-toxic" and a negative charge to "harmful"; and assign mass values ​​based on the authority of the data source, outputting parameters including charge and mass. gather.

[0106] The output of the preceding steps is encapsulated into a complete data unit, specifically:

[0107] When any parsing engine in the preceding sequence successfully produces a core semantic assertion, the encapsulation process is immediately initiated. The system calls a globally unique identifier generation service, which uses a specific algorithm (such as the snowflake algorithm or UUID generation algorithm) to ensure that a unique string sequence is generated for each semantic cell in a distributed system environment, denoted as the cell identifier.

[0108] The system instantiates a predefined software object in memory. This object acts as a data container, with several pre-defined fields for storing specific information, including but not limited to: core assertion fields, potential parameter fields, and metadata fields.

[0109] The "semantic content" and "interaction behavior" attributes of the cells are systematically populated into the aforementioned data framework. For example, the structured assertions output by the parsing engine (e.g., {attribute: "toxicity", value: "non-toxic"}) are written into the core assertion field of the data container as the core data body of the cell. The set of physical parameters calculated from the "potential energy parameter prediction network" (e.g., {charge: +0.8, mass: 1.0}) is written into the potential energy parameter field of the data container. The data source of the assertions (e.g., "declaration text"), the parsing confidence level (e.g., 0.95), and other contextual information are written into the metadata field of the data container as descriptive data.

[0110] The system performs an integrity check on the fully populated data container to ensure that all necessary fields have been correctly assigned values. Then, the container is marked as "encapsulated," and a semantic cell is generated. A readily available data entity. Logically, this entity is a composite data object containing identity, core semantics, interaction parameters, and descriptive metadata. This complete... Outputted by the system and transmitted to the downstream "semantic field simulation" component, it awaits injection into the corresponding digital twin. It participates in the consistency calculation.

[0111] Step 103: Generate semantic cells Inject digital twins sequentially ; According to the injection and It dynamically updates its internal state attribute set, such as updating the "environmental risk index" to "high".

[0112] Digital twin Internally, a semantic cell registry is maintained. The system does not randomly insert cells, but determines the priority of their injection based on cell type, source authority, or timestamp, and records their information in the registry.

[0113] Whenever there are new semantic cells When successfully injected into the registry, or when the cell collection reaches a certain critical point, The internal state update engine is automatically triggered. This engine first retrieves all current state information from the registry. Complete information, including its core assertions and potential parameters. This constructs a temporary, complete snapshot of the internal context as input for state evaluation.

[0114] The state update engine feeds a context snapshot into a state evaluation model, which contains a set of predefined state mapping rules.

[0115] Example of environmental risk index rules:

[0116] IF has a semantic cell whose core assertion is [Environmental Hazard: Harmful to the aquatic environment] AND its potential parameter charge < -0.5 AND has another cell whose behavior assertion is [Dock: Sensitive area].

[0117] THEN calls a weighted evaluation function. This function combines the potential parameters of these relevant cells (such as the intensity of charge and the weight of mass) with their confidence levels to calculate a numerical result.

[0118] FINALLY updates the value of the "Environmental Risk Index" status attribute to "High" based on the range in which the value falls (e.g., 0.0-0.3 is "Low", 0.3-0.7 is "Medium", and 0.7-1.0 is "High").

[0119] It should be noted that: It can be understood as a complete "report" or a "biological organism," while These are the "paragraphs and sentences" that make up this report, or the "organs and cells" of a living organism. Digital twin. As a container and manager, it holds all the semantic cells extracted from this application. It provides a unified context, enabling semantic cells that originally came from different sources (text, images, trajectories) to be associated together.

[0120] It should be noted that each cell is responsible for carrying only one specific semantic content, contains only one core semantic assertion, and has one and only one set of semantic potential parameters based on that content. .

[0121] The subscript 'i' in the code is an index variable used to identify and distinguish different semantic cells. In a system, there may be multiple semantic cells, each with a unique index 'i', similar to an array index in programming or a primary key in a database, used to represent the instance identifier of the cell.

[0122] The subscript 's' in the table is a type identifier representing "semantic". It indicates that this set of parameters are semantically related potential parameters used to define the interactive behavior of cells in the semantic field. It is not an index, but a fixed set of parameter names, which contains a specific set of values ​​(such as charge and mass) for each cell.

[0123] That is, i is an instance-level identifier, pointing to a single cell, and s is a type-level identifier, pointing to the type (semantic potential) of the parameter.

[0124] When using this method, refer to steps 101 to 103:

[0125] By integrating scattered multimodal data into a unified digital twin, the problem of data silos in traditional systems is solved. This provides a complete and structured operational object for subsequent consistency verification, avoiding the complexity of cross-data source queries. Through semantic cell encapsulation, raw data is transformed into semantically rich intelligent units. The introduction of semantic potential parameters endows data with dynamic interaction capabilities, surpassing traditional static data storage. The digital twin updates its state attributes in real time as cells are injected, and the system has adaptive cognitive capabilities, reflecting changes in the integrity and consistency of declared information. Standardized semantic cell interfaces facilitate the rapid access of new data modalities. Modular design supports functional expansion without affecting the core architecture. Object-level data transfer reduces serialization / deserialization overhead. Pre-structured data provides ready-to-use input for subsequent analysis, significantly reducing computational latency. Dynamic state attributes provide real-time indicators for risk assessment. Rich contextual information supports more accurate consistency judgments. The encapsulation mechanism ensures data integrity and reduces the impact of external interference. State evolution logic provides anomaly detection and fault tolerance capabilities.

[0126] Step 2: Systematically traverse the digital twin. All semantic cells The system extracts its core assertions and performs semantic matching with nodes in the domain knowledge graph. For each semantic cell's core assertion, it finds a corresponding standard concept node in the knowledge graph, determines the force rules, and the system forms the current digital twin. Create an empty semantic field instance. And load all force rules into this field instance. middle.

[0127] Step two includes the following:

[0128] Step 201: Systematically traverse the digital twin. All semantic cells Extract the "attribute-value" pairs from its core assertions. For example, from the cell... The assertion "toxicity: non-toxic" is extracted. The system then performs semantic matching between the extracted assertions and nodes in the domain knowledge graph, finding a corresponding standard concept node in the knowledge graph for the core assertion of each semantic cell.

[0129] The system converts the core assertions and concept node names in the knowledge graph into semantic vectors, and then finds the closest graph node in the vector space. This is not a simple string matching, but a semantic search based on vector similarity.

[0130] Step 202: Using the set of concept nodes obtained from the first stage mapping as the query condition, traverse the knowledge graph and retrieve all relation edges with the "force rule" label that connect these nodes.

[0131] The system transforms these retrieved relational rules, described in natural language or logical labels, into quantifiable parameters that the physics engine can understand. For example, the rule label "mutually exclusive" is directly mapped to "same charge sign," and the "base strength" parameter is mapped to the absolute value of the charge.

[0132] Examples of predefined rules in knowledge graphs:

[0133] There exists a relation edge pointing from [Concept: Non-toxic] to [Concept: Environmental Hazard], with a rule type of "mutually exclusive" and parameters {force type: "repulsive force", base strength: 0.9}.

[0134] There exists a relation edge pointing from [Concept: Toxicity] to [Concept: Environmental Risk], with a rule type of "Related" and parameters {Force Type: "Gravity", Base Strength: 0.7}.

[0135] Step 203: The system is the current digital twin. Create an empty semantic field instance. All force rules retrieved and parameterized in the second stage are loaded into this field instance. At this point, it is no longer an empty shell, but contains a complete system specifically designed for... The "interaction law" of all semantic concepts within.

[0136] The system will use this fully configured semantic field With digital twins Binding signifies that the field is ready for subsequent physics simulations.

[0137] It should be noted that: These are the "intrinsic properties" of a semantic cell, defining how the cell should "behave" in a physical simulation. They can be understood as the cell's inherent physical properties such as "mass" and "charge," just as an electron carries a negative charge e. A cell asserted to be "non-toxic" inherently possesses a set of... parameter.

[0138] It is the "external environment" for cell interactions, and it is a digital twin. The specific domain of action defines the "physical laws" of interactions between cells. For example, the rule that "like charges repel and unlike charges attract" is determined by the field. Decision made Just like the gravitational field we are in, it is not a property of objects themselves, but it stipulates that all objects with mass will be subject to gravity within it.

[0139] When using this method, refer to steps 201 to 203:

[0140] By precisely mapping cellular semantic assertions to knowledge graph concepts and determining the rules of interaction, the system transforms previously isolated, static data points into "physical entities" with explicit interactive behaviors (attraction / repulsion) within the semantic field. This allows subsequent consistency verification to move beyond simple True / False judgments based on symbolic logic, enabling more nuanced and reliable results through simulating their dynamic interactions. When the system detects a contradiction, its root cause can be clearly traced.

[0141] Step 3, All semantic cells within Inserting semantic field Each cell is assigned a charge attribute mapped from the knowledge graph based on its core assertion. The system runs a simplified physical simulation loop until it reaches a steady or quasi-steady state. The total potential energy of the entire cell system is calculated, and the semantic stability score calculated by passing the total potential energy through a preset normalization function is compared with a preset threshold range, automatically triggering the corresponding decision branch.

[0142] Step three includes the following steps:

[0143] Step 301, All semantic cells within Inserting semantic field Each cell is assigned a charge attribute mapped from the knowledge graph based on its core assertion. The system runs a simplified physical simulation loop, calculating the electrostatic force between each pair of cells according to Coulomb's law, and iteratively solving for the net external force and energy state of the entire cellular system until the system reaches a steady state or quasi-steady state.

[0144] System traversal Each semantic cell in Based on the concepts mapped to in the knowledge graph by its core assertions, from the semantic field The rules determine the sign and baseline value of its charge, and then combine this with the cell's own... The parameters were fine-tuned, and finally each cell was assigned a charge attribute value used in this simulation.

[0145] From the semantic field The rules determine the sign and baseline value of its charge, and then combine this with the cell's own... The parameters were fine-tuned, and each cell was ultimately assigned a charge attribute value used in this simulation, specifically as follows:

[0146] The system maps core assertions to standard concept nodes in the domain knowledge graph (e.g., mapping "toxicity: non-toxic" to [concept: non-toxic]). The system traverses the knowledge graph, retrieving all relation edges marked with "interaction rules" that start or end with that concept node. For example, it retrieves an edge from [concept: non-toxic] to [concept: environmental hazard] with the rule "mutually exclusive".

[0147] Output one or more qualitative logical rules. For example: "For the concept of non-toxic, when it interacts with the concept of environmental hazard, the relationship is mutually exclusive."

[0148] The system converts logical rules into charge symbols based on a predefined "logic-physical" mapping table.

[0149] "Mutual repulsion" charges have the same sign and are usually defined as positive [+], while "correlated" or "consistent" charges have opposite signs and are usually defined as negative [-], indicating attraction.

[0150] The same mapping table assigns a baseline strength value to logical rules of different types. For example, the baseline strength for "mutual exclusion" is 0.8 (strong), and the baseline strength for "weak association" is 0.3 (weak).

[0151] At this point, the system has obtained the sign of the charge (e.g., +) and the reference value (e.g., 0.8).

[0152] System read The quality parameter (which reflects the authority of the data source, such as an MSDS being of higher quality than plain text) is used to weight the baseline value. Higher quality contributes more to the final charge value, and its assertions carry more weight in the simulation. Final charge value = Baseline value × Quality

[0153] The system assigns an initial coordinate to each cell in an abstract two-dimensional or three-dimensional simulation space. This coordinate is only used to simulate the relative position between cells and has no actual geographical meaning.

[0154] Assigning an initial coordinate to each cell is specifically as follows:

[0155] The system analyzes the core assertions of all semantic cells and evaluates the initial semantic association strength between each pair of cells using a lightweight semantic similarity calculation model.

[0156] Cells describing different properties of the same entity (such as “toxicity” and “flammability” describing the same shipment) are considered to be strongly correlated and are assigned a smaller initial distance.

[0157] Cells from the same data source (such as those from the same MSDS document) are considered to have a moderate association and are assigned a moderate initial distance.

[0158] Cells that are semantically unrelated are assigned a large initial distance.

[0159] The system All semantic cells within the cell are paired up, and for each pair of cells... and Perform the following calculations:

[0160] Obtaining cells and charge value and and their current distance in the simulation space. .

[0161] According to the simplified form of Coulomb's law = k * ( * ) / ( ^2) Calculate the scalar electrostatic force between them. .

[0162] Where k is a constant coefficient, if and If they have the same sign, then A positive value indicates repulsive force; if and If the signs are different, then A negative value indicates attractiveness.

[0163] Calculated scalar force According to cells and The direction of the line connecting them can be decomposed into the forces acting on them. and Force vectors in opposite directions are applied to each cell. These force vectors are then added to the resultant force vector acting on each cell.

[0164] Based on the net force vector acting on each cell in the current iteration, its acceleration and velocity are calculated according to simplified Newton's laws of motion, and its coordinates in the simulation space are updated accordingly. Cells subjected to net repulsive forces will move away from each other, while cells subjected to net attractive forces will move closer to each other.

[0165] A velocity damping coefficient is introduced during the update process to simulate the "friction" effect of the system. This is done to dissipate the system's kinetic energy, prevent the cells from oscillating indefinitely, and thus ensure that the simulation eventually converges to a steady state.

[0166] After each iteration, the system calculates the total kinetic energy of the entire cellular system and the sum of the magnitudes of the net external force vectors acting on all cells. When both values ​​decrease and stabilize below a preset threshold, the system is considered to have reached a steady state or quasi-steady state.

[0167] Step 302: After the system reaches steady state, calculate the total potential energy of the entire cellular system based on the final relative positions of all cell pairs and the forces between them. Map the total potential energy of the system to a semantic stability score between 0 and 1 using a preset normalization function. The higher the score, the more consistent the submitted content.

[0168] A consistent declaration will result in a cellular system that converges to a low potential energy state; while a contradictory declaration will result in a system that is in a high potential energy state due to internal stress.

[0169] Step 303: The system compares the calculated semantic stability score with the preset threshold range and automatically triggers the corresponding decision branch:

[0170] The high score range of 0.8 - 1.0 is judged as "highly consistent and low risk". The system automatically marks the application status as "verification passed" and allows it to flow into the next business process stage.

[0171] The range of 0.4 - 0.8 is considered "suspicious, medium risk," and the system marks the application as "requiring manual review" and places it in the reviewer's pending work queue. Simultaneously, the system pre-prepares the visual diagnostic report generated in step two for the reviewer, greatly improving the efficiency of manual review.

[0172] A score in the low range of 0.0-0.4 is considered "serious contradiction, high risk," and the system automatically marks the declaration as "rejected and issues an alert." In addition to halting the process, the system also sends immediate alerts to high-risk control positions (such as customs inspection officers).

[0173] When using this method, refer to steps 301 to 303:

[0174] By using physical simulation, complex multimodal information contradictions are transformed into a single, quantifiable "semantic stability score." This completely avoids the subjective judgment differences and uncertainties caused by factors such as personal experience and fatigue in traditional review processes. Decisions are based on unified, objective numerical standards, ensuring consistency and fairness in processing results. System behavior is no longer fixed but possesses strategic flexibility. For example, during special periods, thresholds can be raised to adopt stricter review standards, enabling the system to flexibly adapt to constantly changing business environments.

[0175] Please see Figure 2 This invention provides a collaborative processing system for ship declaration based on multiple user roles, comprising:

[0176] The semantic cell generation module instantiates a digital twin from the new ship declaration request. Software objects, extracting core semantic assertions, metadata, and semantic potential parameters. Together they are encapsulated into semantic cells Data units will generate semantic cells Inject digital twins sequentially ;

[0177] The cell rule analysis module systematically traverses the digital twin. All semantic cells The core assertions are extracted and semantically matched with nodes in the domain knowledge graph to find the corresponding standard concept nodes in the knowledge graph, determine the force rules, and provide a basis for the current digital twin. Create an empty semantic field instance. Load all force rules into this field instance. middle;

[0178] The cell system simulation module will All semantic cells within Inserting semantic field Each cell is assigned a charge attribute mapped from the knowledge graph based on its core assertion; the system runs a physical simulation loop until the system reaches a steady state or quasi-steady state, calculates the total potential energy of the entire cell system, calculates the semantic stability score of the total potential energy through a normalization function, compares it with a preset threshold range, and automatically triggers the corresponding decision branch.

[0179] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0180] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0181] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A collaborative processing method for ship declaration based on multiple user roles, characterized in that: Includes the following steps: Instantiate a digital twin of the new ship declaration request. Software objects, extracting core semantic assertions, metadata, and semantic potential parameters. Together they are encapsulated into semantic cells Data units will generate semantic cells Inject digital twins sequentially ; System traversal of digital twins All semantic cells The core assertions are extracted and semantically matched with nodes in the domain knowledge graph to find the corresponding standard concept nodes in the knowledge graph, determine the force rules, and provide a basis for the current digital twin. Create an empty semantic field instance. Load all force rules into this field instance. middle; The extraction of core semantic assertions, metadata, and semantic potential parameters Specifically: The text parsing engine processes the declaration description text and outputs structured assertions and confidence levels; the document parsing engine analyzes the MSDS image, identifies key sections through layout recognition, and extracts assertions and source locations; the trajectory parsing engine processes AIS time-series data and identifies behavioral pattern assertions. Structured assertions, assertions, and behavioral pattern assertions are collectively defined as core assertions; The core assertions and their metadata are input into the potential energy parameter prediction network, which dynamically calculates the physical simulation parameters based on the assertion semantics and the current declaration context. Specifically: Assign charge values ​​with opposite signs to core assertions that have opposing relationships; and assign mass values ​​based on the authority of the data source, outputting a value containing both charge and mass parameters. gather; Will All semantic cells within Insert semantic field Each cell is assigned a charge attribute mapped from the knowledge graph based on its core assertion; the system runs a physical simulation loop until the system reaches a steady state or quasi-steady state, calculates the total potential energy of the entire cell system, calculates the semantic stability score of the total potential energy through a normalization function, compares it with a preset threshold range, and automatically triggers the corresponding decision branch.

2. The collaborative processing method for ship declaration based on multiple user roles according to claim 1, characterized in that: The process of instantiating a digital twin of the new ship declaration request. Software objects, specifically: The system receives the application request through the interface, parses out the key information, calls the digital twin management service, executes the new application digital twin operation, and creates a new object of that class in memory. The system automatically invokes this newly effective object. The initialization method; within this method, the following is executed programmatically: Assign a system-generated unique value to the twin identifier attribute, assign the application business number attribute to the received application ID, set the current status attribute to the initial state, set the timestamp attribute to the current system time, initialize the semantic cell container to an empty data structure, and initialize the derived attribute set to an empty key-value pair set.

3. The collaborative processing method for ship declaration based on multiple user roles according to claim 1, characterized in that: The common encapsulation is a semantic cell. The data unit is specifically: When any parsing engine in the preceding sequence successfully produces a core semantic assertion, the encapsulation process is immediately started. The system calls a globally unique identifier generation service to generate a unique string sequence for each semantic cell, which is denoted as the cell identifier. The system instantiates a predefined software object in memory; this object acts as a data container and has several predefined fields for storing specific information, including core assertion fields, potential energy parameter fields, and metadata fields. The system performs an integrity check on the fully populated data container to ensure that all necessary fields have been correctly assigned values; subsequently, the container is marked as encapsulated, and a semantic cell is finally generated. Immediately available data entities.

4. The collaborative processing method for ship declaration based on multiple user roles according to claim 1, characterized in that: The process involves extracting the core assertions and performing semantic matching with nodes in the domain knowledge graph to find the corresponding standard concept nodes in the knowledge graph. System traversal of digital twins All semantic cells Extract its core assertions and perform semantic matching with nodes in the domain knowledge graph. For each semantic cell, find a corresponding standard concept node in the knowledge graph for the core assertion. The system converts the core assertions and concept node names in the knowledge graph into semantic vectors, and then finds the closest graph node in the vector space, which is a semantic search based on vector similarity.

5. The collaborative processing method for ship declaration based on multiple user roles according to claim 1, characterized in that: The specific rules for determining the force are as follows: Using the set of concept nodes obtained from the mapping as the query condition, the knowledge graph is traversed to retrieve all relation edges with force rule labels that connect these nodes; The system transforms these retrieved relational rules, described in natural language or logical labels, into quantized parameters that the physics engine can understand.

6. The collaborative processing method for ship declaration based on multiple user roles according to claim 1, characterized in that: Each cell is specifically assigned a charge attribute derived from the knowledge graph mapping based on its core assertion, as follows: The system maps core assertions to standard concept nodes in the domain knowledge graph. The system traverses the knowledge graph, retrieves all relation edges marked with interaction rules that start or end with the concept node, and outputs one or more qualitative logic rules. The system converts logical rules into charge symbols based on a predefined logical-physical mapping table and assigns a reference strength value to different types of logical rules; thus, the charge symbols and reference values ​​are obtained. System read The mass parameter is used to weight the reference value, and the final charge value = reference value × mass.

7. The collaborative processing method for ship declaration based on multiple user roles according to claim 1, characterized in that: The specific physical simulation cycle of the system operation is as follows: In an abstract two-dimensional or three-dimensional simulation space, the system analyzes the core assertions of all semantic cells, evaluates the initial semantic association strength between each pair of cells through a lightweight semantic similarity calculation model, and assigns an initial coordinate to each cell.

8. The collaborative processing method for ship declaration based on multiple user roles according to claim 1, characterized in that: The system runs a physical simulation cycle until it reaches a steady state or quasi-steady state, specifically: The system All semantic cells within the cell are paired up, and for each pair of cells... and Perform the following calculations: Obtaining cells and charge value and and their current distance in the simulation space. ; According to the simplified form of Coulomb's law = k ( ) / ( ^2) Calculate the scalar electrostatic force between them. ; Where k is a constant coefficient, if and If they have the same sign, then A positive value indicates repulsive force; if and If the signs are different, then A negative value indicates attractiveness; Calculated scalar force According to cells and The direction of the line connecting them is decomposed into the forces acting on them. and Force vectors in opposite directions are added to the resultant force vector acting on each cell. Based on the resultant force vector acting on each cell in the current iteration, its acceleration and velocity are calculated according to the simplified Newton's laws of motion, and its coordinates in the simulation space are updated accordingly. After each iteration, the system calculates the total kinetic energy of the entire cellular system and the sum of the magnitudes of the net external force vectors acting on all cells; when both of these values ​​decrease and stabilize below a preset threshold, the system is considered to have reached a steady state or quasi-steady state.

9. A collaborative processing system for ship declaration based on multiple user roles, used to implement the method described in any one of claims 1 to 8, characterized in that: include: The semantic cell generation module instantiates a digital twin from the new ship declaration request. Software objects, extracting core semantic assertions, metadata, and semantic potential parameters. Together they are encapsulated into semantic cells Data units will generate semantic cells Inject digital twins sequentially ; The cell rule analysis module systematically traverses the digital twin. All semantic cells The core assertions are extracted and semantically matched with nodes in the domain knowledge graph to find the corresponding standard concept nodes in the knowledge graph, determine the force rules, and provide a basis for the current digital twin. Create an empty semantic field instance. Load all force rules into this field instance. middle; The cell system simulation module will All semantic cells within Insert semantic field Each cell is assigned a charge attribute mapped from the knowledge graph based on its core assertion; The system runs a physical simulation loop until it reaches a steady state or quasi-steady state. It calculates the total potential energy of the entire cellular system, uses a normalization function to calculate the semantic stability score, compares it with a preset threshold range, and automatically triggers the corresponding decision branch.

Images