Multi-department collaborative decision-making method and system based on large model and water transport safety risk event grading

Abstract

The invention relates to the field of emergency decision-making, and particularly discloses a multi-department collaborative decision-making method and system based on a large model and water transportation safety risk event grading, which is used for constructing ship navigation comprehensive situation characteristics by carrying out multi-dimensional parameter joint perception on ship AIS data and ship navigation environment data. Then, all historical risk event records related to the current ship navigation comprehensive situation in a professional knowledge base in the water transportation safety field are inquired, deep context modeling is carried out on the historical risk event records to capture safety risk modes and coping strategies possibly existing in current ship navigation, and then the safety risk modes and coping strategies are obtained; semantic reasoning is carried out on the historical risk event context reference features and the current ship navigation comprehensive situation features by means of a large model, so that risk assessment and collaborative decision suggestions for the current ship are generated. The method not only can adapt to real-time navigation situation evolution, but also has credible support of historical disposal experience, and provides effective guarantee for timeliness and collaboration of cross-department emergency response.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of emergency decision-making, and more specifically, to a multi-department collaborative decision-making method and system based on large models and water transport safety risk event classification. BACKGROUND

[0002] As an important part of global trade and domestic logistics, the safe and efficient operation of waterway transportation has a crucial impact on national economy and social development. However, the water transport system has high inherent complexity, involving ships, waterways, ports, weather, hydrology and many other dynamic factors, and sudden events often have a chain effect, making safety risk prevention and control often involve multi-department collaborative work of maritime supervision, weather warning, port scheduling, etc.

[0003] Currently, in the field of water transport safety risk management and emergency decision-making, the water transport safety risk management mode still relies on independent information collection, analysis and judgment of each department, supplemented by traditional manual consultation mechanisms. In the face of growing ship traffic flow and complex and variable navigation environment, it gradually exposes problems such as information barriers, response delays, strong subjectivity of decision-making, and difficulty in effectively passing on experience. On the one hand, the fragmented management of heterogeneous data such as AIS, weather, and waterways leads to fragmented situational awareness, making it difficult to capture complex risk factors in a timely manner. On the other hand, although existing research attempts to use large models to realize water transport safety risk warning, due to the high degree of specialization, complex terminology, large number of industry standards and processes, and much non-public or scattered expert tacit knowledge in the field of water transport safety, the large model is prone to poor industry adaptability when generating analysis or suggestions, giving unrealistic or contradictory to actual operation solutions, thereby reducing its reliability and usability in actual decision-making.

[0004] Therefore, an optimized multi-department collaborative decision-making method and system based on large models and water transport safety risk event classification is expected. SUMMARY

[0005] To solve the above technical problems, the present application is proposed. Embodiments of the present application provide a multi-department collaborative decision-making method and system based on large models and water transport safety risk event classification. The method constructs a comprehensive situation feature of ship navigation containing dynamic characteristics of the ship and constraints of hydrological and meteorological conditions of the channel by multi-dimensional parameter joint perception of ship AIS data and ship navigation environment data. Then, all historical risk event records related to the current comprehensive situation of ship navigation in the professional knowledge base of the water transport safety field are queried, and the queried historical risk event records are deeply context modeled to capture the possible safety risk patterns and coping strategies in the current ship navigation, forming a context reference basis. Further, with the help of large models, semantic reasoning is performed on the historical risk event context reference features and the comprehensive situation features of the current ship navigation to generate risk assessment and collaborative decision-making suggestions for the current ship. The method can adapt to the evolution of real-time navigation situation and has reliable support of historical disposal experience, providing effective guarantee for the timeliness and collaboration of cross-department emergency response.

[0006] According to one aspect of the present application, a multi-department collaborative decision-making method based on large models and water transport safety risk event classification is provided, which includes:

[0007] Obtaining ship AIS data from a ship automatic identification system and obtaining ship navigation environment data from a meteorological department;

[0008] Joint perception fusion of the ship AIS data and the ship navigation environment data to obtain a comprehensive situation feature of ship navigation;

[0009] Constructing a professional knowledge base of the water transport safety field, which stores a plurality of water transport safety historical risk event records;

[0010] Querying water transport safety historical risk event records related to the comprehensive situation feature of ship navigation from the professional knowledge base of the water transport safety field to obtain a set of water transport safety historical risk event records;

[0011] Aggregating and analyzing the set of water transport safety historical risk event records to obtain water transport safety historical risk event context reference features;

[0012] Generating multi-department collaborative decision-making suggestions based on the water transport safety historical risk event context reference features and the comprehensive situation features of ship navigation.

[0013] According to another aspect of the present application, a multi-department collaborative decision-making system based on large models and water transport safety risk event classification is provided, which includes:

[0014] The navigation environment data acquisition module is configured to acquire ship AIS data from a ship automatic identification system and acquire ship navigation environment data from a meteorological department;

[0015] The joint perception fusion module is configured to perform joint perception fusion on the ship AIS data and the ship navigation environment data to obtain ship navigation comprehensive situation features;

[0016] The professional knowledge base construction module is configured to construct a water transportation safety field professional knowledge base, and the water transportation safety field professional knowledge base stores a plurality of water transportation safety historical risk event records;

[0017] The historical risk query module is configured to query water transportation safety historical risk event records related to the ship navigation comprehensive situation features from the water transportation safety field professional knowledge base to obtain a set of water transportation safety historical risk event records;

[0018] The risk record aggregation analysis module is configured to perform aggregation analysis on the set of water transportation safety historical risk event records to obtain water transportation safety historical risk event context reference features;

[0019] The decision suggestion generation module is configured to generate multi-department collaborative decision suggestions based on the water transportation safety historical risk event context reference features and the ship navigation comprehensive situation features.

[0020] Compared with the prior art, the multi-department collaborative decision method and system based on a large model and water transportation safety risk event classification provided in the application can perform multi-dimensional parameter joint perception on ship AIS data and ship navigation environment data to construct ship navigation comprehensive situation features containing ship dynamic characteristics and channel hydro-meteorological condition constraints. Then, all historical risk event records related to the current ship navigation comprehensive situation in the water transportation safety field professional knowledge base are queried, and the queried historical risk event records are subjected to deep context modeling to capture possible safety risk patterns and coping strategies in the current ship navigation, form context reference basis, and then, with the aid of a large model, semantic reasoning is performed on the historical risk event context reference features and the current ship navigation comprehensive situation features to generate risk assessment and collaborative decision suggestions for the current ship. The method can adapt to real-time navigation situation evolution and has credible support of historical disposal experience, and effectively guarantees the timeliness and collaboration of cross-department emergency response. BRIEF DESCRIPTION OF DRAWINGS

[0021] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 This is a flowchart of a multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to an embodiment of the present application.

[0023] Figure 2 This is a data flow diagram of a multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to an embodiment of the present application.

[0024] Figure 3 This is a flowchart of sub-step S2 of the multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to an embodiment of the present application.

[0025] Figure 4 This is a flowchart of sub-step S5 of the multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to an embodiment of the present application.

[0026] Figure 5 This is a flowchart of sub-step S52 of the multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to an embodiment of the present application.

[0027] Figure 6 This is a flowchart of sub-step S522 of the multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to an embodiment of the present application.

[0028] Figure 7 This is a block diagram of a multi-department collaborative decision-making system based on a large model and water transport safety risk event classification according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0030] Although the present application makes various references to certain modules in the system according to embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server. The modules are merely illustrative and different aspects of the system and method can use different modules.

[0031] Flowcharts are used in the present application to illustrate the operations performed by the system according to embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in sequence. Instead, various steps can be processed in reverse order or simultaneously, as desired. Other operations can also be added to or removed from these processes, or one or more steps can be removed from these processes.

[0032] In the following, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It is obvious that the described embodiments are only a part of the embodiments of the present application, and not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0033] It is worth noting that in the present application, all actions of obtaining data are carried out in compliance with the corresponding data protection regulations and policies of the country where the device is located, and with the authorization given by the owner of the corresponding device.

[0034] To solve the technical problems described in the above background, the present application proposes a multi-department collaborative decision-making method based on large models and water transport safety risk event classification. The method performs multi-dimensional parameter joint perception on ship AIS data and ship navigation environment data to construct a comprehensive situation feature of ship navigation containing ship dynamic characteristics and channel hydrological and meteorological condition constraints. Then, by querying all historical risk event records related to the current ship navigation comprehensive situation in the water transport safety field knowledge base, and performing deep context modeling on the queried historical risk event records, the method captures the possible safety risk patterns and coping strategies in the current ship navigation to form a context reference basis. Further, by means of a large model, the method performs semantic reasoning on the historical risk event context reference features and the current ship navigation comprehensive situation features to generate risk assessment and collaborative decision-making suggestions for the current ship. The method can adapt to the evolution of real-time navigation situation, and has the credible support of historical disposal experience, providing effective guarantee for the timeliness and collaboration of cross-department emergency response.

[0035] Figure 1 Flowchart of the multi-department collaborative decision-making method based on large models and water transport safety risk event classification according to embodiments of the present application. Figure 2 Data flow diagram of the multi-department collaborative decision-making method based on large models and water transport safety risk event classification according to embodiments of the present application. As Figure 1 and Figure 2As shown, the multi-department collaborative decision-making method based on large models and water transportation safety risk event classification includes the following steps: S1, obtaining ship AIS data from a ship automatic identification system and obtaining ship navigation environment data from a meteorological department; S2, jointly perceiving and fusing the ship AIS data and the ship navigation environment data to obtain ship navigation integrated situation features; S3, constructing a water transportation safety field professional knowledge base, the water transportation safety field professional knowledge base stores a plurality of water transportation safety historical risk event records; S4, querying water transportation safety historical risk event records related to the ship navigation integrated situation features from the water transportation safety field professional knowledge base to obtain a set of water transportation safety historical risk event records; S5, aggregating and analyzing the set of water transportation safety historical risk event records to obtain water transportation safety historical risk event context reference features; and S6, generating multi-department collaborative decision-making suggestions based on the water transportation safety historical risk event context reference features and the ship navigation integrated situation features.

[0036] In the above multi-department collaborative decision-making method based on large models and water transportation safety risk event classification, the step S1, obtaining ship AIS data from a ship automatic identification system and obtaining ship navigation environment data from a meteorological department. It should be understood that the formation of water transportation safety risks often results from the complex interaction of ship dynamic characteristics and environmental parameters, and single-dimensional data is difficult to fully reflect the potential risks of the navigation situation. Therefore, in order to realize the fine perception of the ship navigation state, the present application is based on the multi-source heterogeneous data fusion theory, by integrating the real-time dynamic data of the ship automatic identification system (AIS) and the static environmental data of the meteorological and channel departments, to obtain multi-dimensional parameters covering ship maneuvering performance, cargo carrying characteristics and channel constraint conditions. Specifically, the ship AIS data includes ship type, navigation state, ship position, ground heading, ground speed, ship tonnage and loaded cargo type, etc., and the ship navigation environment data includes channel geometric parameters (depth, width), wind speed, wind direction, visibility, sea flow direction and speed, wave height and wave period, etc. Hydro-meteorological conditions, both of which constitute the basic data set for water transportation safety risk analysis, providing full-factor input for subsequent ship navigation situation modeling.

[0037] Specifically, the Automatic Identification System (AIS) as one of the core data sources of modern maritime traffic management, can provide high-frequency, wide-coverage ship operation state information. The data obtained through this system includes but is not limited to ship type, navigation state, position coordinates, ground heading, ground speed, ship tonnage, and loaded cargo type, etc. These data are continuously sent at a certain update period, usually refreshed every few seconds to tens of seconds, ensuring continuous tracking ability of ship behavior. Due to the strong penetration and stability of AIS signals, they can still maintain high data availability in complex sea conditions, so they have become an indispensable information source in waterway transportation monitoring.

[0038] In the process of specific implementation, in order to realize the efficient acquisition of AIS data, the system needs to access the AIS base station network deployed by the national or regional maritime management department, or obtain the AIS data stream processed by the third-party data service provider. Considering the coverage range and signal density difference of different waters, the system also needs to have a flexible data access mechanism to adapt to the application requirements in various scenarios such as coastal, inland, and port. In addition, in order to ensure the timeliness and accuracy of the data, the system should introduce a data cleaning and verification module to eliminate abnormal records caused by equipment failure, signal interference or data format error, so as to ensure the quality of the basic data relied on subsequent analysis.

[0039] At the same time, the safety of ship navigation is not only affected by its own motion characteristics, but also closely related to external environmental conditions. Therefore, in addition to the AIS data of the ship itself, relevant data about the navigation environment need to be obtained from the meteorological department. These data cover wind speed, wind direction, visibility, sea flow direction and speed, wave height and period, etc., which are important factors affecting the ship maneuvering performance and navigation risk. For example, strong wind may cause the ship to deviate or lose control; low visibility will significantly increase the risk of collision; and complex water flow and wave conditions may cause the ship to lose stability, thus causing safety accidents.

[0040] The acquisition path of meteorological data mainly includes government meteorological service agencies, professional weather forecasting platforms, and marine monitoring agencies. These data sources provide various forms of information, including regular weather reports based on fixed time intervals, and high-resolution real-time environmental data generated by satellite remote sensing, radar detection, etc. To meet the system's requirement for data timeliness, API interface calling or data subscription is usually used for automated collection to ensure data acquisition and transmission in the shortest time. At the same time, considering the significant differences in weather change rules in different sea areas and different seasons, the system should also have the ability to dynamically adjust the data collection strategy according to geographical location and time to improve the overall sensing accuracy.

[0041] It is worth noting that AIS data and meteorological environmental data come from different industry systems, and there are great differences in their collection standards, update frequency, spatial resolution, etc. For example, AIS data records the moving track of the ship with second-level precision, while meteorological data is usually published with hourly or longer time scale. This heterogeneity makes the fusion of the two face certain challenges, so the needs of later processing should be fully considered in the data acquisition stage, and data sources with higher spatio-temporal matching degree should be selected as much as possible. In addition, for some specific parameters such as wind direction and wind speed, attention should be paid to whether the measurement height or depth conforms to the actual affected level of the ship, so as to avoid risk misjudgment caused by data deviation.

[0042] In terms of data structure, AIS data is usually stored in text or binary format, containing fields such as timestamp, ship MMSI identification code, latitude and longitude, speed, heading, etc., which is convenient for parsing and processing; while meteorological data may exist in NetCDF, GRIB, CSV format, involving multiple variables and their changes over time and space. In order to facilitate unified management and subsequent modeling, the system needs to convert the original data into a standardized data format after obtaining it, and establish a unified metadata description system for cross-system sharing and calling. In addition, a data caching mechanism should be set up to deal with sudden communication interruption or data delay, etc., to ensure the stable operation of the whole system.

[0043] In the above method for multi-department collaborative decision-making based on large models and water transport safety risk event classification, the step S2 is to perform joint perception fusion on the ship AIS data and the ship navigation environment data to obtain ship navigation comprehensive situation features. Specifically, since the generation of ship navigation risk has parameter correlation, the traditional isolated analysis of AIS or environmental data is easy to ignore key risk factors (such as the stability degradation of a specific tonnage ship encountering cross wind in a narrow channel). Therefore, in order to construct a global perspective of navigation risk representation, the present application further performs joint perception modeling on the ship AIS data and the ship navigation environment data to comprehensively consider the multi-dimensional risk factors in the ship navigation process, and forms the ship navigation comprehensive situation features. Among them, Figure 3 The flow chart of the sub-step S2 of the method for multi-department collaborative decision-making based on large models and water transport safety risk event classification according to the embodiments of the present application. As shown in Figure 3As shown, the step S2 comprises steps of: S21, embedding and encoding the ship AIS data and the ship navigation environment data respectively to obtain a ship AIS data embedding feature encoding vector and a ship navigation environment data embedding feature encoding vector; S22, inputting the ship AIS data embedding feature encoding vector and the ship navigation environment data embedding feature encoding vector into a multi-layer perception based navigation situation joint perception module to obtain a ship navigation comprehensive situation feature vector as the ship navigation comprehensive situation feature.

[0044] Specifically, the step S21, embedding and encoding the ship AIS data and the ship navigation environment data respectively to obtain a ship AIS data embedding feature encoding vector and a ship navigation environment data embedding feature encoding vector. It can be understood that, due to the significant differences in data dimension and distribution characteristics between the ship AIS data (such as ship type, tonnage, speed) and the navigation environment data (such as sea wave height, visibility), direct fusion may cause feature conflict or information loss. For example, the ship type as a discrete classification variable lacks comparability with the numerical physical parameters such as sea wave height in the numerical space. Therefore, in order to eliminate the semantic gap between heterogeneous data and extract high-dimensional abstract features, the ship AIS data and the ship navigation environment data are first mapped to a unified semantic space through embedding and encoding technology. Specifically, for discrete features (such as ship type, navigation state, loaded cargo type) in AIS data, One-Hot Encoding is used for conversion to map it to a vector form; and for numerical features (such as ship tonnage, speed, etc.), linear transformation is performed through normalization and full connection layer; finally, the vector representations of various parameters are spliced to form a ship AIS data embedding feature encoding vector. Similarly, for the navigation environment data, the same embedding and encoding strategy as the above ship AIS data is used to form a ship navigation environment data embedding feature encoding vector. In this way, the ship AIS data and the ship navigation environment data can represent the deep semantic association between ship behavior and environmental constraints in the same vector space, providing an aligned semantic basis for subsequent feature interaction fusion.

[0045] Specifically, the step S22, the ship AIS data embedding feature code vector and the ship navigation environment data embedding feature code vector are input into the multi-layer perception based navigation situation joint perception module to obtain a ship navigation comprehensive situation feature vector as the ship navigation comprehensive situation feature. It should be understood that, since the formation of the ship navigation risk depends on the nonlinear interaction of the ship characteristics and the environmental parameters (for example, when a large ship is affected by cross flow in a narrow channel, its maneuvering response is essentially different from that of a small ship), simple feature splicing or linear weighting cannot capture such complex relationships. Therefore, in order to model the nonlinear coupling effect between multi-modal features, the present application is based on the deep neural network architecture of multi-layer perception (MLP), and the nonlinear transformation and feature interaction of the ship AIS data embedding feature code vector and the ship navigation environment data embedding feature code vector are performed to realize multi-dimensional information joint perception, so as to capture the ship navigation comprehensive situation feature. Specifically, first, the ship AIS data embedding feature code vector and the ship navigation environment data embedding feature code vector are concatenated and input into the MLP module, which is stacked by multiple fully connected layers and activation functions (such as GELU). Each layer performs affine transformation on the input features through a weight matrix and introduces nonlinearity, and gradually extracts high-order cross features (such as the joint influence mode of "ship tonnage-channel width-sea wave period"). Finally, through the iterative learning and feature abstraction of the multi-layer network, a ship navigation comprehensive situation feature vector is obtained, which integrates the ship navigation state and the hydro-meteorological condition constraints of the channel, and is used to represent the comprehensive risk situation of the ship in the current navigation environment.

[0046] In the above multi-department collaborative decision-making method based on large models and water transportation safety risk event classification, the step S3 of constructing a water transportation safety field professional knowledge base stores a plurality of water transportation safety historical risk event records. It should be understood that the present application takes into account the high professionalism and complex industry standards and processes in the field of water transportation safety, so that the direct use of large models for risk assessment and decision suggestion generation may face the lack and misunderstanding of industry knowledge, resulting in limited feasibility and accuracy of the suggestions. Therefore, in order to improve the accuracy and reliability of the decision, the present application introduces a water transportation safety field professional knowledge base as an auxiliary. Specifically, the construction of the water transportation safety field professional knowledge base is based on long-term water transportation safety practice, and through expert review and experience summary, a plurality of water transportation safety historical risk event records are organized and compiled, each of which not only describes the occurrence background, process and consequences of the risk event in detail, but also contains risk level assessment, involved ship type, accident specific reason analysis, relevant department response measures and final processing results. In this way, based on the water transportation safety field professional knowledge base, historical experience can be accumulated and reused, providing effective historical reference and reliable basis for water transportation safety risk level determination and emergency response strategy formulation, and helping to enhance the large model's compliance and understanding of industry standards, ensuring that the generated decision suggestions achieve the optimal technical feasibility.

[0047] In the multi-department collaborative decision-making method based on large models and water transportation safety risk event classification described above, step S4 queries water transportation safety historical risk event records related to the ship navigation comprehensive situation features from the water transportation safety field knowledge base to obtain a set of water transportation safety historical risk event records. In a specific example of the present application, the water transportation safety historical risk event records include historical risk event descriptions, risk levels, involved ship types, accident causes, related department response measures, and processing results. It should be understood that since the similarity of the current ship navigation situation and historical events may imply similar risk evolution rules, in order to filter historical cases highly related to the current ship navigation situation, the present application uses a feature similarity retrieval algorithm to achieve accurate association through vector space matching. Specifically, the ship navigation comprehensive situation feature vector is used as a query vector, and the ship AIS data and navigation environment data in historical cases are extracted from each water transportation safety historical risk event record. The navigation situation features of each historical risk event are constructed as library vectors based on the joint encoding mode of the ship AIS data and navigation environment data, and the cosine similarity index is used to measure the similarity between the query vector and each library vector, thereby selecting the top-N historical risk events similar to the current ship navigation situation to form a set of water transportation safety historical risk event records. In this way, by accurately matching the current navigation situation features with the historical risk event records in the professional knowledge base, similar risk scenarios can be quickly located, and effective response measures can be extracted to provide effective decision support for risk event response management.

[0048] In the multi-department collaborative decision-making method based on large models and water transportation safety risk event classification described above, step S5 aggregates and analyzes the set of water transportation safety historical risk event records to obtain water transportation safety historical risk event context reference features. Specifically, the locality of a single historical risk event may lead to biased decision recommendations, while multi-event aggregation can reveal common risk patterns, potential rules, and lessons learned. Therefore, the present application further aggregates and analyzes the set of water transportation safety historical risk event records to mine common rules in the historical risk event cluster as a context reference for risk decision-making, forming a generalizable risk response paradigm. Among them, Figure 4 The flowchart of sub-step S5 of the multi-department collaborative decision-making method based on large models and water transportation safety risk event classification according to the embodiments of the present application. As Figure 4As shown, the step S5 includes steps of: S51, performing Bert model-based text semantic vectorization processing on each water transport safety historical risk event record in the set of water transport safety historical risk event records to obtain a set of water transport safety historical risk event semantic feature encoding vectors; S52, performing clustering analysis on the set of water transport safety historical risk event semantic feature encoding vectors to obtain a water transport safety historical risk event context reference feature encoding vector as the water transport safety historical risk event context reference feature.

[0049] Specifically, the step S51 performs Bert model-based text semantic vectorization processing on each water transport safety historical risk event record in the set of water transport safety historical risk event records to obtain a set of water transport safety historical risk event semantic feature encoding vectors. It can be understood that since the water transport safety historical risk event records contain a large amount of unstructured text (such as event description, accident cause chain, countermeasures, etc.), the semantic connotation thereof involves complex causal logic and domain terminology, and the traditional bag-of-words model is difficult to capture the implicit association in the event context. Therefore, in order to convert the unstructured text into a computable deep semantic representation, the present application is based on the transfer learning principle of the pre-trained language model, and performs context-aware semantic coding on each water transport safety historical risk event record through the BERT (Bidirectional Encoder Representations from Transformers) model. Specifically, first, the water transport safety historical risk event record is subjected to word segmentation and serialization processing, then the word sequence is input into the BERT model, the bidirectional context dependency relationship between words is captured by using the Transformer architecture in the BERT model, and a deep context embedding representation of each word is generated; finally, the hidden state vector corresponding to the [CLS] mark is extracted as the overall semantic representation of the event to obtain the water transport safety historical risk event semantic feature vector. In this way, each water transport safety historical risk event record is converted into a vector in a high-dimensional semantic space, so that semantically similar or related events are close to each other in the vector space, thereby facilitating subsequent common risk pattern and countermeasure mining.

[0050] Specifically, the step S52, the set of water transportation safety historical risk event semantic feature encoding vectors is subjected to clustering analysis to obtain a water transportation safety historical risk event context reference feature encoding vector as the water transportation safety historical risk event context reference feature. That is, in order to mine the generalizable risk response paradigm from the historical event cluster, the application identifies the core mode in the event cluster by clustering analysis on the set of water transportation safety historical risk event semantic feature vectors, and forms a more refined water transportation safety historical risk event context reference feature encoding vector by aggregating the compensation information of each water transportation safety historical risk event relative to the core risk common mode on the basis of the historical risk event cluster core common mode. In this way, not only the common risk mode of the historical event cluster is summarized, but also the subtle differences between different events and the response strategies in specific situations are revealed, providing more comprehensive and in-depth reference for risk decision-making. Among them, Figure 5 The flowchart of the sub-step S52 of the multi-department collaborative decision-making method based on large models and water transportation safety risk event classification according to the embodiment of the application. As shown in Figure 5 The step S52 includes the steps of: S521, performing ground state semantic aggregation on the set of water transportation safety historical risk event semantic feature encoding vectors to obtain a water transportation safety historical risk event ground state semantic aggregation encoding vector; S522, performing gain modulation on the water transportation safety historical risk event ground state semantic aggregation encoding vector based on the semantic compensation of each water transportation safety historical risk event semantic feature encoding vector in the set of water transportation safety historical risk event semantic feature encoding vectors relative to the water transportation safety historical risk event ground state semantic aggregation encoding vector to obtain the water transportation safety historical risk event context reference feature encoding vector.

[0051] More specifically, the step S521 includes: first, inputting the set of water transportation safety historical risk event semantic feature encoding vectors into a K-Means clustering network to obtain K water transportation safety historical risk event semantic clustering center encoding vectors, which can be expressed by the formula:

[0052] X={x1,x2,...,x i ,...,x n}∈R n×d

[0053] C=K_Means(X)={c1,c2,...,c k}

[0054] Wherein, X represents the set of water transportation safety historical risk event semantic feature encoding vectors, x1, x2, x i and x nrespectively represent the first, second, i-th and n-th waterway safety historical risk event semantic feature encoding vector in the set of waterway safety historical risk event semantic feature encoding vectors, n is the number of vectors in the set of waterway safety historical risk event semantic feature encoding vectors, d represents the feature dimension value of the waterway safety historical risk event semantic feature encoding vector, K_Means(·) represents the K-Means clustering network, C represents the set of waterway safety historical risk event semantic clustering center encoding vectors, c1, c2 and ck respectively represent the first, second and k-th waterway safety historical risk event semantic clustering center encoding vector in the set of waterway safety historical risk event semantic clustering center encoding vectors, and k respectively represent the first, second and k-th waterway safety historical risk event semantic clustering center encoding vector in the set of waterway safety historical risk event semantic clustering center encoding vectors;

[0055] Then, the K waterway safety historical risk event semantic clustering center encoding vectors are subjected to ground state aggregation based on the self-attention mechanism to obtain the waterway safety historical risk event ground state semantic aggregation encoding vector, which is expressed by the formula as follows:

[0056] Q = CW Q

[0057] K = CW k

[0058] V = CW v

[0059]

[0060] g = layerNorm(C + Attention(Q, K, V))

[0061] wherein Q, K and V respectively represent a query matrix, a key matrix and a value matrix, W Q , W k and W v respectively represent a query embedding matrix, a key embedding matrix and a value embedding matrix, Attention(·) represents an attention interaction function, Softmax represents a normalized exponential function, layerNorm(·) represents a layer normalization operation, and g represents a waterway safety historical risk event ground state semantic aggregation encoding vector.

[0062] That is, by extracting the water transport safety historical risk event ground state semantic aggregation encoding vector that can reflect the internal structure and core semantics from the set of water transport safety historical risk event semantic feature encoding vectors, representative ground state semantic information support is provided for subsequent analysis to integrate the key semantic features of historical risk events and build a referenceable basic semantic framework. Specifically, the present application identifies the representative centers of the areas with higher data density in the high-dimensional feature space and mines the high-order dependency relationship between the centers through the K-Means clustering network and the self-attention mechanism, forming the water transport safety historical risk event ground state semantic aggregation encoding vector that can reflect the internal structure and core distribution characteristics of the entire set, and laying an effective foundation for subsequent generation of multi-department collaborative decision suggestions combined with the current ship navigation comprehensive situation characteristics.

[0063] Figure 6 The flowchart of sub-step S522 of the multi-department collaborative decision method based on a large model and water transport safety risk event classification according to the embodiment of the present application. As shown in Figure 6 the step S522, it includes steps: S5221, extracting the semantic compensation features of each water transport safety historical risk event semantic feature encoding vector in the set of water transport safety historical risk event semantic feature encoding vectors relative to the water transport safety historical risk event ground state semantic aggregation encoding vector to obtain a set of water transport safety historical risk event semantic compensation feature encoding vectors; S5222, calculating the saliency modulation weight factors of each water transport safety historical risk event semantic compensation feature encoding vector in the set of water transport safety historical risk event semantic compensation feature encoding vectors to obtain a set of water transport safety historical risk event semantic compensation saliency modulation weight factors; S5223, based on the set of water transport safety historical risk event semantic compensation saliency modulation weight factors, weighting and modulating the set of water transport safety historical risk event semantic compensation feature encoding vectors to obtain a set of water transport safety historical risk event semantic compensation feature saliency encoding vectors; S5224, inputting the set of water transport safety historical risk event semantic compensation feature saliency encoding vectors and the water transport safety historical risk event ground state semantic aggregation encoding vector into a feature gain superposition network to obtain the water transport safety historical risk event context reference feature encoding vector.

[0064] In one specific example of the present application, the step S5221 is expressed by the formula:

[0065] e i =Sigmoid(W e [x i ;g]+b e )⊙(x i -g)

[0066] ξ={e1,e2,...,ei ..., e n}

[0067] wherein [·; ·] represents feature concatenation, Sigmoid represents a Sigmoid activation function, W e and b e respectively represent a semantic compensation weight parameter matrix and a semantic compensation bias term, ξ represents a set of semantic compensation feature encoding vectors of water transportation safety historical risk events, e1, e2, e i and e n respectively represent x1, x2, x i and x n corresponding semantic compensation feature encoding vectors of water transportation safety historical risk events.

[0068] That is, the individual difference information not covered by the water transportation safety historical risk event ground state semantic aggregation encoding vector is stripped from the water transportation safety historical risk event semantic feature encoding vector, each water transportation safety historical risk event semantic feature encoding vector is decomposed into a ground state common projection and a semantic compensation component, and a water transportation safety historical risk event semantic compensation feature encoding vector is generated, thereby providing data support for subsequent mining of specific patterns and differentiated response strategies of risk events, enabling the model to accurately capture abnormal features or specific semantic information in historical events that do not conform to the ground state mode, and effectively improving the adaptability of decisions to complex shipping scenarios.

[0069] In particular, the present application considers that in calculating the semantic compensation representation of each water transportation safety historical risk event semantic feature encoding vector relative to the water transportation safety historical risk event ground state semantic aggregation encoding vector, x i -g substantially defines a dynamic induced metric of each geometric metric space of the water transportation safety historical risk event semantic feature encoding vector relative to the ground state reference space of the water transportation safety historical risk event ground state semantic aggregation encoding vector, i.e., the relative high-dimensional water transportation safety historical risk event semantic feature encoding vector x icharacteristic space as a geometric metric to define a low-dimensional geometric metric in the original state reference space of the waterway safety historical risk event ground state semantic aggregation encoding vector by deriving the mapping. In this way, it is necessary to consider the boundary measure problem under the dynamic induced metric, that is, to make the semantic compensation characteristic of each waterway safety historical risk event limited to the topological boundary constraint of the dimension transition under the derived mapping, rather than a simple high-dimensional-low-dimensional manifold mapping. Based on this, in one preferred example of the present application, the step S5222 comprises: first, based on the semantic difference boundary limitation between each waterway safety historical risk event semantic feature encoding vector in the set of waterway safety historical risk event semantic feature encoding vectors and the waterway safety historical risk event ground state semantic aggregation encoding vector, the characteristic modulation is carried out on the semantic compensation feature encoding vector of each waterway safety historical risk event to obtain the set of optimized waterway safety historical risk event semantic compensation feature encoding vectors.

[0070] Specifically, first, calculate x i Low-rank activation weight factor relative to g:

[0071] α =‖x i ‖1-‖g‖1

[0072] β =‖x i ‖2-‖g‖2

[0073] Where ‖·‖1 represents the L1 norm, ‖·‖2 represents the L2 norm, and α and β represent different low-rank activation weight factors, respectively.

[0074] Next, introduce low-rank tangent space projection to decompose the derived mapping in a tangent space decomposition mode, so that it is projected onto the orthogonal basis in the original state reference space:

[0075] x 1i = x i - α

[0076] x 2i = x i - β

[0077] Where x 1i and x 21 represent different dimension reduction tangent space projection vectors, respectively.

[0078] Then, the low-rank tangent space projection is taken as a geometric metric under the local sub-manifold structure to perform edge-manifold geometric reconstruction to convert the activation state quantization parameter complexity from being determined by the high-dimensional set of waterway safety historical risk event semantic feature encoding vectors to being determined by the topological boundary constraint of the tangent space geometric metric:

[0079]

[0080] where ω is an adaptive compensation factor for compensating the too large manifold curvature modification factor n n-1 The impact of e (·) denotes the exponential function with base natural constant, x′ i denotes x i corresponding corrected water transportation safety historical risk event semantic feature encoding vector.

[0081] Finally, x′ i is modified as e i = x′ i ⊙ e i ⊙ e i , where e′ i denotes e i corresponding optimized water transportation safety historical risk event semantic compensation feature encoding vector, i.e., the i-th optimized water transportation safety historical risk event semantic compensation feature encoding vector in the set of optimized water transportation safety historical risk event semantic compensation feature encoding vectors. In this way, by making the excitation metric dependent on the boundary rather than the spatial transformation in the case of dynamically induced metrics for submanifold geometry mapping to the original state, the fixed points in the tangent geometry flow can be regularized to achieve effective quantization of the topological boundary and avoid feature representation redundancy suppression of each water transportation safety historical risk event semantic compensation feature encoding vector.

[0082] Then, each optimized water transportation safety historical risk event semantic compensation feature encoding vector in the set of optimized water transportation safety historical risk event semantic compensation feature encoding vectors is input into a feature transformation compression module based on the ReLU function and then globally normalized to obtain a set of water transportation safety historical risk event semantic compensation saliency modulation weight factors, which is expressed by the formula as follows:

[0083]

[0084] A = {α1, α2,..., α i ,...,α n}

[0085] where w denotes the attention score conversion vector, W s denotes a learnable weight parameter matrix, ReLU(·) denotes the ReLU activation function, exp(·) denotes the exponential function with base e, A denotes the set of water transportation safety historical risk event semantic compensation saliency modulation weight factors, α1, α2,..., and α i denote the corresponding water transportation safety historical risk event semantic compensation saliency modulation weight factors. n i n ​​​

[0086] That is, by feature transformation compression and global normalization processing, the importance of the optimized water transportation safety historical risk event semantic compensation feature encoding vector is quantified, the key difference features with actual value for risk analysis are screened out, the interference of irrelevant noise or minor details is suppressed, and more reasonable weight allocation basis is provided for subsequent weighted modulation. Specifically, the application uses the nonlinear transformation of the ReLU function to highlight the significance of effective features and filter negative or invalid information, and through global normalization, the importance weights of each optimized water transportation safety historical risk event semantic compensation feature encoding vector have cross-dimension comparability, and finally a set of water transportation safety historical risk event semantic compensation significance modulation weight factors are formed, which can accurately reflect the influence degree of unique features of different historical risk events on current decision-making.

[0087] In one specific example of the application, the step S5223 is represented by the formula:

[0088]

[0089] wherein, represents a set of water transportation safety historical risk event semantic compensation feature significant encoding vectors, and respectively represent e1, e2, e i and e n corresponding water transportation safety historical risk event semantic compensation feature significant encoding vectors.

[0090] That is, based on the importance difference of each water transportation safety historical risk event semantic compensation feature encoding vector, the water transportation safety historical risk event semantic compensation significance modulation weight factor is used for differential weighting processing, so that the features with more value for current risk analysis and decision-making are strengthened, and the influence of secondary or irrelevant features is reasonably weakened, thereby forming a set of water transportation safety historical risk event semantic compensation feature significant encoding vectors focusing on key semantic compensation information. Through this weight-driven feature modulation mechanism, the specific semantic features with actual decision-making reference value in historical risk events are accurately extracted and enhanced.

[0091] In one specific example of the application, the step S5224 is represented by the formula:

[0092]

[0093] wherein, MLP(·) represents a multi-layer perceptron, v f represents a water transportation safety historical risk event context reference feature encoding vector.

[0094] That is, through the weight-driven feature screening and enhancement mechanism, the specific semantic features highly related to the current shipping safety risk assessment in historical risk events are accurately extracted and highlighted, making the generated set of water transportation safety historical risk event context reference feature encoding vectors not only embody the uniqueness of historical risk events relative to the ground state, but also highlight the actual impact of different features on cross-departmental collaborative decision-making, effectively improving the collaboration of multi-department emergency response and the industry adaptability of decision-making schemes.

[0095] In the above multi-department collaborative decision-making method based on large models and water transportation safety risk event classification, the step S6 generates multi-department collaborative decision-making suggestions based on the water transportation safety historical risk event context reference features and the ship navigation comprehensive situation features. In one specific example of the present application, the step S6 includes inputting the water transportation safety historical risk event context reference features and the ship navigation comprehensive situation features into a risk event decision generation model based on a LLM large model to obtain the multi-department collaborative decision-making suggestions, which include event risk level, related department responsibilities, recommended response measures, and expected processing results. It should be understood that the ship navigation comprehensive situation features and the water transportation safety historical risk event context reference features respectively carry current ship navigation risk perception and historical experience rules. In order to fully utilize historical experience to deeply understand the current risk situation and assist decision-making, the present application inputs the water transportation safety historical risk event context reference features and the ship navigation comprehensive situation features into a LLM (Large Language Model) large model, and based on a prompt engineering to guide it to generate multi-dimensional collaborative decisions conforming to the field specifications. Specifically, the LLM large model can deeply analyze the input features based on the extensive knowledge and logical reasoning ability learned from large-scale corpus training, and generate risk event decision-making suggestions conforming to the prompt requirements. In the implementation process, first, a prompt template is embedded in the LLM large model (such as a GPT-3 model pre-trained based on maritime regulations, emergency plans, etc.), for example, “According to the current navigation situation: [ship navigation comprehensive situation features] and historical risk event reference information: [water transportation safety historical risk event context reference features], generate decision-making suggestions including risk level, department responsibilities, response measures, and expected processing results”. Then, the LLM large model analyzes the input features according to the prompt template, wherein the ship navigation comprehensive situation features provide real-time navigation state and potential risk points, and the water transportation safety historical risk event context reference features provide response strategies and lessons from similar historical events. The LLM large model uses its built-in reasoning mechanism and knowledge base to output structured decision text according to the guidance of the prompt template, to clearly define the risk level of the current risk event, the related departments involved and their responsibility division, the recommended response measures, and the expected processing results, which helps decision-makers better grasp the decision-making effect and make reasonable choices. In this way, by introducing historical context reference information, the decision-making process of the large model is constrained within the boundaries of domain knowledge, avoiding suggestions that violate industry norms, while retaining dynamic adaptability to real-time situations, thereby improving the accuracy and practicality of decision-making suggestions.

[0096] In summary, the multi-department collaborative decision-making method based on large models and water transport safety risk event classification according to the embodiments of the present application is illustrated, which constructs a ship navigation comprehensive situation feature containing ship dynamic characteristics and channel hydrological and meteorological condition constraints by multi-dimensional parameter joint perception of ship AIS data and ship navigation environment data. Then, by querying all historical risk event records related to the current ship navigation comprehensive situation in the water transport safety field knowledge base, and performing deep context modeling on the queried historical risk event records, the possible safety risk patterns and coping strategies in the current ship navigation are captured to form a context reference basis. Further, with the help of large models, semantic reasoning is performed on the historical risk event context reference features and the current ship navigation comprehensive situation features to generate risk assessment and collaborative decision-making suggestions for the current ship. This method can adapt to real-time navigation situation evolution and has reliable support of historical disposal experience, providing effective guarantee for the timeliness and collaboration of cross-department emergency response.

[0097] Further, a multi-department collaborative decision-making system based on large models and water transport safety risk event classification is also provided.

[0098] Figure 7 A block diagram of the multi-department collaborative decision-making system based on large models and water transport safety risk event classification according to the embodiments of the present application is shown. As shown in Figure 7 The multi-department collaborative decision-making system based on large models and water transport safety risk event classification 100 according to the embodiments of the present application includes: a navigation environment data acquisition module 110 for acquiring ship AIS data from a ship automatic identification system and acquiring ship navigation environment data from a meteorological department; a joint perception fusion module 120 for joint perception fusion of the ship AIS data and the ship navigation environment data to obtain a ship navigation comprehensive situation feature; a professional knowledge base construction module 130 for constructing a water transport safety field professional knowledge base, which stores a plurality of water transport safety historical risk event records; a historical risk query module 140 for querying water transport safety historical risk event records related to the ship navigation comprehensive situation feature from the water transport safety field professional knowledge base to obtain a set of water transport safety historical risk event records; a risk record aggregation analysis module 150 for aggregating and analyzing the set of water transport safety historical risk event records to obtain water transport safety historical risk event context reference features; and a decision-making suggestion generation module 160 for generating multi-department collaborative decision-making suggestions based on the water transport safety historical risk event context reference features and the ship navigation comprehensive situation features.

[0099] Here, those skilled in the art can understand that the specific operations of each module in the above multi-department collaborative decision-making system based on large models and water transport safety risk event classification have been described above with reference to Figures 1 to 6The description of the multi-department collaborative decision-making method based on a large model and water transportation safety risk event classification is described in detail in the description of the first aspect, and therefore, the repeated description thereof will be omitted.

[0100] The above describes the basic principles of the present application in combination with specific embodiments, but it should be noted that the advantages, advantages, effects, etc. mentioned in the present application are only examples and not limitations, and these advantages, advantages, effects, etc. cannot be considered as necessary for each embodiment of the present application. In addition, the specific details of the above embodiments are only for the purpose of example and for the purpose of understanding, and are not limited to the above specific details. The present application must be implemented using the above specific details.

[0101] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in a certain embodiment can be referred to the related description of other embodiments. In the several embodiments provided by the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the unit division is only a logical function division, and there can be other division ways in actual implementation. The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place or distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0102] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered as exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be considered as limiting the claims involved.

[0103] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The plurality of units stated in the system claim can also be implemented by one unit through software or hardware.

[0104] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solutions of the present application and are not limited. Although the technical solutions are modified or replaced by equivalents with reference to the preferred embodiments, they do not deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. A multi-department collaborative decision-making method based on a large model and water transport safety risk event classification, characterized by: include: Obtain ship AIS data from the ship automatic identification system and ship navigation environment data from the meteorological department; Performing joint perception fusion on the ship AIS data and the ship navigation environment data to obtain comprehensive ship navigation situation characteristics; Building a professional knowledge base in the field of water transport safety, wherein the professional knowledge base in the field of water transport safety stores a plurality of historical water transport safety risk event records; Searching the water transport safety field professional knowledge database for water transport safety historical risk event records related to the comprehensive navigation situation characteristics of the ship to obtain a collection of water transport safety historical risk event records; Performing aggregate analysis on the set of water transport safety historical risk event records to obtain context reference features of the water transport safety historical risk events; Based on the contextual reference features of the historical water transport safety risk events and the comprehensive navigation situation features of the ship, multi-department collaborative decision-making recommendations are generated.

2. The multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to claim 1 is characterized in that: The ship AIS data includes ship type, navigation status, ship position, course over ground, speed over ground, ship tonnage and loaded cargo type; the ship navigation environment data includes channel depth and width, wind speed, wind direction, visibility, sea water direction and speed, wave height and wave period; the water transport safety historical risk event record includes historical risk event description, risk level, type of ship involved, cause of accident, response measures of relevant departments and handling results.

3. The multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to claim 2 is characterized in that: Performing joint perception fusion on the ship AIS data and the ship navigation environment data to obtain comprehensive ship navigation situation characteristics, including: Embedding the ship AIS data and the ship navigation environment data to obtain a ship AIS data embedding feature coding vector and a ship navigation environment data embedding feature coding vector respectively; The ship AIS data embedded feature coding vector and the ship navigation environment data embedded feature coding vector are input into a navigation situation joint perception module based on a multi-layer perceptron to obtain a ship navigation comprehensive situation feature vector as the ship navigation comprehensive situation feature.

4. The multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to claim 1 is characterized in that: Based on the contextual reference features of the historical water transport safety risk events and the comprehensive navigation situation features of the ship, a multi-department collaborative decision-making recommendation is generated, including: The context reference features of the historical risk events of water transport safety and the comprehensive situation features of the ship navigation are input into the risk event decision generation model based on the LLM large model to obtain the multi-department collaborative decision-making recommendations, which include the event risk level, responsibilities of relevant departments, recommended response measures and expected processing results.

5. The multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to claim 1 is characterized in that: The set of water transport safety historical risk event records is aggregated and analyzed to obtain contextual reference features of the water transport safety historical risk events, including: Performing text semantic vectorization processing based on the Bert model on each water transport safety historical risk event record in the set of water transport safety historical risk event records to obtain a set of water transport safety historical risk event semantic feature encoding vectors; A cluster analysis is performed on the set of semantic feature coding vectors of the water transport safety historical risk events to obtain a context reference feature coding vector of the water transport safety historical risk events as the context reference feature of the water transport safety historical risk events.

6. The multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to claim 5 is characterized in that: Cluster analysis is performed on the set of semantic feature coding vectors of the water transport safety historical risk events to obtain a context reference feature coding vector of the water transport safety historical risk events as the context reference feature of the water transport safety historical risk events, including: Performing base state semantic aggregation on the set of semantic feature coding vectors of the historical water transport safety risk events to obtain a base state semantic aggregation coding vector of the historical water transport safety risk events; Based on the semantic compensation of each semantic feature coding vector of water transport safety historical risk events in the set of semantic feature coding vectors of water transport safety historical risk events relative to the ground state semantic aggregation coding vector of water transport safety historical risk events, the ground state semantic aggregation coding vector of water transport safety historical risk events is gain modulated to obtain the context reference feature coding vector of water transport safety historical risk events.

7. The multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to claim 6 is characterized in that: Performing base state semantic aggregation on the set of semantic feature coding vectors of the historical water transport safety risk events to obtain a base state semantic aggregation coding vector of the historical water transport safety risk events, including: Inputting the set of semantic feature encoding vectors of the water transport safety historical risk events into a K-Means clustering network to obtain K semantic clustering center encoding vectors of the water transport safety historical risk events; The K semantic clustering center encoding vectors of the water transport safety historical risk events are subjected to ground state aggregation based on the self-attention mechanism to obtain the ground state semantic aggregation encoding vector of the water transport safety historical risk events.

8. The multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to claim 7 is characterized in that: Based on semantic compensation of each semantic feature coding vector of the water transport safety historical risk event in the set of semantic feature coding vectors of the water transport safety historical risk event relative to the ground state semantic aggregate coding vector of the water transport safety historical risk event, gain modulation is performed on the ground state semantic aggregate coding vector of the water transport safety historical risk event to obtain the context reference feature coding vector of the water transport safety historical risk event, including: Extracting semantic compensation features of each semantic feature coding vector of the historical water transport safety risk events in the set of semantic feature coding vectors of the historical water transport safety risk events relative to the base state semantic aggregation coding vector of the historical water transport safety risk events to obtain a set of semantic compensation feature coding vectors of the historical water transport safety risk events; Calculating the significance modulation weight factor of each semantic compensation feature coding vector of the water transport safety historical risk event in the set of semantic compensation feature coding vectors of the water transport safety historical risk event to obtain a set of significance modulation weight factors of semantic compensation for the water transport safety historical risk event; Based on the set of semantic compensation significance modulation weight factors of the water transport safety historical risk events, weighted modulation is performed on the set of semantic compensation feature coding vectors of the water transport safety historical risk events to obtain a set of semantic compensation feature significance coding vectors of the water transport safety historical risk events; The set of semantic compensation feature significant coding vectors of the water transport safety historical risk events and the ground state semantic aggregation coding vector of the water transport safety historical risk events are input into the feature gain superposition network to obtain the context reference feature coding vector of the water transport safety historical risk events.

9. The multi-department collaborative decision-making method based on a large model and water transport safety risk event classification according to claim 8 is characterized in that: Calculating the significance modulation weight factor of each water transport safety historical risk event semantic compensation feature coding vector in the set of water transport safety historical risk event semantic compensation feature coding vectors to obtain a set of water transport safety historical risk event semantic compensation significance modulation weight factors, including: Based on the semantic difference boundary definition between each semantic feature coding vector of the water transport safety historical risk event in the set of semantic feature coding vectors of the water transport safety historical risk event and the ground state semantic aggregation coding vector of the water transport safety historical risk event, feature modulation is performed on each semantic compensation feature coding vector of the water transport safety historical risk event to obtain a set of optimized semantic compensation feature coding vectors of the water transport safety historical risk event; Each optimized semantic compensation feature coding vector of historical water transport safety risk events in the set of optimized semantic compensation feature coding vectors of historical water transport safety risk events is input into a feature transformation compression module based on the ReLU function and then globally normalized to obtain a set of significance modulation weight factors of semantic compensation for historical water transport safety risk events.

10. A multi-department collaborative decision-making system based on a large model and water transport safety risk event classification, characterized by: include: The navigation environment data acquisition module is used to obtain ship AIS data from the ship automatic identification system and obtain ship navigation environment data from the meteorological department; A joint perception and fusion module is used to perform joint perception and fusion on the ship AIS data and the ship navigation environment data to obtain comprehensive navigation situation characteristics of the ship; A professional knowledge base construction module is used to construct a professional knowledge base in the field of water transport safety, wherein the professional knowledge base in the field of water transport safety stores a plurality of historical water transport safety risk event records; A historical risk query module is used to query the water transport safety historical risk event records related to the comprehensive navigation situation characteristics of the ship from the water transport safety professional knowledge database to obtain a collection of water transport safety historical risk event records; a risk record aggregation analysis module, configured to perform aggregation analysis on the set of water transport safety historical risk event records to obtain contextual reference features of water transport safety historical risk events; The decision-making suggestion generation module is used to generate multi-department collaborative decision-making suggestions based on the context reference characteristics of the historical water transport safety risk events and the comprehensive navigation situation characteristics of the ship.

Images