CN-122024453-A - Intelligent identification method and system for ship navigation risk alarm information

CN122024453ACN 122024453 ACN122024453 ACN 122024453ACN-122024453-A

Abstract

The invention relates to the technical field of port safety management, in particular to an intelligent identification method and system for ship navigation risk alarm information; the method comprises the steps of carrying out iterative updating of a directional entity perceived graph annotation force network on an initial node feature matrix to obtain a final high-order node feature matrix when the entity and the relation are extracted, inputting the final high-order node feature matrix into a classifier to carry out entity identification and relation classification, and outputting a brand new final high-order node feature matrix in an iterative updating link, wherein each row in the matrix is a depth feature which is rich in global context information and directional entity perception and is customized for entity identification and relation classification. The depth features are directly input to a subsequent classifier to finish a final extraction task, so that the extraction precision and robustness of complex alarm sentences are remarkably improved.

Inventors

SUI HAICHEN
YANG LIU
LI JING
WANG FANGZHENG
SUN HUI
YANG KUN
MENG FANWEI
WANG WENJIE
DONG XUHUA
ZHANG KAIHUA

Assignees

天津水运工程研究院有限公司
交通运输部天津水运工程科学研究所

Dates

Publication Date: 20260512
Application Date: 20260407

Claims (10)

1. The intelligent identification method for the ship navigation risk alarm information is characterized by comprising the following steps of: S1, collecting multi-source alarm information; s2, carrying out data preprocessing operation on the multi-source alarm information; s3, carrying out data processing on the preprocessed multi-source alarm information, and extracting an entity and a relation; the step S3 is specifically as follows: S3.1, performing preliminary coding operation on the preprocessed multi-source alarm information to obtain an initial node feature matrix, S3.2, constructing a full-connection graph structure, S3.3, performing iterative updating of a graph attention network for directional entity perception on the initial node feature matrix based on a graph neural network model of a graph attention mechanism to obtain a final high-order node feature matrix, S3.4, inputting the final high-order node feature matrix into a classifier, and performing entity recognition and relationship classification; s4, constructing a ship navigation risk knowledge graph according to the entity and the relationship; and S5, carrying out intelligent identification on the ship navigation risk alarm information according to the ship navigation risk knowledge graph.
2. The intelligent identification method for ship navigation risk warning information according to claim 1, wherein, In S3.3, the iterative process is repeated from the first The input of the graph neural network model based on the graph attention mechanism is the layer L from the beginning of the layer L to the end of the layer L Node feature matrix of all nodes of layer, for layer 1, the input of the graph neural network model based on graph attention mechanism is the initial node feature matrix ; Firstly, performing linear transformation on the input of the graph neural network model based on the graph attention mechanism, and then calculating the importance of the node j to the node i as a basic attention score The calculation formula is as follows: ; In the formula, For the vector concatenation operation, Is a learnable weight vector for mapping the spliced high-dimensional vector into a scalar attention score, For the transpose operation, In order to activate the function, Is the ith node after linear transformation The node feature vector of the layer, Is the jth node after linear transformation Node feature vectors of the layers; Calculating and adding directional entity perception bias terms ; The directional entity aware bias term The expression of (2) is: ; where lambda is a coefficient, sigmoid is a normalization function, As the relative distance between the words of node i and node j, As the entity vector of the node i, An entity vector for node j; the attention score of the node i and the node j is the sum of the basic attention score and the directional entity perception bias term, and the expression is: ; for the node i, carrying out softmax normalization on the attention scores of all the neighbor nodes j to obtain the final attention weight Then, feature polymerization is carried out; Performing the attention score calculation and feature aggregation on all nodes in the fully connected graph in parallel to obtain the first node Node characteristic matrix of all nodes of layer Executing an iteration process, and finally, outputting a final high-order node characteristic matrix of each node after the iteration of the L layer is finished The expression is: ; In the formula, Is a node feature matrix iterated to the ith node of the L-th layer.
3. The intelligent identification method for ship navigation risk warning information according to claim 2, wherein, In S3.1, the preprocessed multi-source alarm information is converted into a word vector S, where an expression of the word vector S is: In which, in the process, The i-th word in the pretreated multi-source alarm information is used, and n is the total number of words in the pretreated multi-source alarm information; Each word in the word vector S is encoded through a BERT-Marine-Domain model to obtain a context vector Meanwhile, position coding is added to preserve the sequence order information of each word, and the word vector S is expressed as an initial node characteristic matrix The mathematical expression is as follows: 。
4. The intelligent identification method for ship navigation risk warning information according to claim 3, wherein, In S3.2, each word in the word vector S Defined as nodes in a fully connected graph structure, thereby forming a node set V having the mathematical expression: In which, in the process, For the ith node, corresponding to the word At any two different nodes And Establishing a undirected edge between the two to form a connection edge set E, thereby forming the full connection graph structure 。
5. The intelligent identification method for ship navigation risk warning information according to claim 2, wherein, In S3.4, the classifier includes an entity identification header for predicting an entity tag corresponding to each word, and a relationship classification header for predicting a relationship type of any pair of entities.
6. The intelligent identification method for ship navigation risk warning information according to claim 1, wherein, In the step S1, the multi-source alarm information is derived from a ship navigation and safety system, a cabin and power monitoring system, a safety and fire monitoring system and a weather and platform system.
7. The intelligent identification method for ship navigation risk warning information according to claim 6, wherein, The ship navigation and safety system comprises a ship automatic identification system, a radar, an electronic chart display and information system, a global positioning system, a depth finder and a log, wherein the cabin and power monitoring system comprises a cabin centralized monitoring system, a host remote control system, a generator control system, a boiler control system and a pump control system, the safety and fire protection monitoring system comprises a fire detection and alarm system, a general alarm system, a carbon dioxide release alarm system and an emergency cutting-off system, and the weather and platform system comprises an anemoscope and a compass.
8. The intelligent identification method for ship navigation risk warning information according to claim 1, wherein, In the step S2, the data preprocessing operation includes data cleansing, data deduplication, and time stamp alignment.
9. The intelligent identification method for ship navigation risk warning information according to claim 1, wherein, In the step S4, a map node is created for each identified unique entity, classification is carried out according to the types of the map nodes, edges with labels are created between the corresponding nodes according to the extracted relation, and the numerical value and the real-time data extracted from the preprocessed multi-source alarm information are added into the attribute table of the corresponding nodes as attributes, so that the ship navigation risk knowledge map is constructed.
10. An intelligent recognition system for ship navigation risk warning information, characterized in that the system adopts the intelligent recognition method for ship navigation risk warning information according to any one of claims 1-9, and the system comprises: the data acquisition module is used for acquiring multi-source alarm information; the data preprocessing module is used for performing data preprocessing operation on the multi-source alarm information; The entity and relation extraction module is used for carrying out data processing on the preprocessed multi-source alarm information and extracting the entity and relation; the knowledge graph construction module is used for constructing a ship navigation risk knowledge graph according to the entity and the relationship; And the intelligent recognition module is used for intelligently recognizing the ship navigation risk alarm information according to the ship navigation risk knowledge graph.

Description

Intelligent identification method and system for ship navigation risk alarm information Technical Field The invention relates to the technical field of port safety management, in particular to an intelligent identification method and system for ship navigation risk alarm information. Background The navigation environment of ships is increasingly complex, and various risks such as collision, stranding, fire, equipment failure, bad weather and the like are faced. Modern vessels are equipped with advanced navigation systems, cabin monitoring systems, weather instruments, etc., which trigger a large amount of alarm information when a potential hazard is detected. In the intelligent alarming field of ships, intelligent identification of alarming information is realized by adopting a knowledge graph as a more deeply studied technical branch, when the knowledge graph of the ship alarming information is constructed, most of traditional entity and relation extraction methods adopt a scheme of identifying the entity first and classifying the entity pair, however, the traditional scheme is easy to cause error propagation, the tiny errors of the entity identification in the previous step can be directly transmitted to a relation classification stage, and meanwhile, the traditional scheme is used for seeing the relation between the entity pair in isolation, and ignoring the inherent grammar structure and logic framework of an alarming sentence. Disclosure of Invention In order to solve the technical problems, the invention provides an intelligent identification method and system for ship navigation risk alarm information, which are used for solving the problems in the prior art. The invention provides an intelligent identification method for ship navigation risk alarm information, which comprises the following steps: S1, collecting multi-source alarm information; s2, carrying out data preprocessing operation on the multi-source alarm information; s3, carrying out data processing on the preprocessed multi-source alarm information, and extracting an entity and a relation; the step S3 is specifically as follows: S3.1, performing preliminary coding operation on the preprocessed multi-source alarm information to obtain an initial node characteristic matrix; S3.2, constructing a full connection diagram structure; S3.3, performing iterative update of a directional entity perceived graph attention network on the initial node feature matrix based on a graph neural network model of a graph attention mechanism to obtain a final high-order node feature matrix; S3.4, inputting the final high-order node feature matrix into a classifier to perform entity identification and relationship classification; s4, constructing a ship navigation risk knowledge graph according to the entity and the relationship; and S5, carrying out intelligent identification on the ship navigation risk alarm information according to the ship navigation risk knowledge graph. Preferably, in S3.3, the iterative process is from the firstThe input of the graph neural network model based on the graph attention mechanism is the layer L from the beginning of the layer L to the end of the layer LNode feature matrix of all nodes of layer, for layer 1, the input of the graph neural network model based on graph attention mechanism is the initial node feature matrix; Firstly, performing linear transformation on the input of the graph neural network model based on the graph attention mechanism, and then calculating the importance of the node j to the node i as a basic attention scoreThe calculation formula is as follows: ; In the formula, For the vector concatenation operation,Is a learnable weight vector for mapping the spliced high-dimensional vector into a scalar attention score,For the transpose operation,In order to activate the function,Is the ith node after linear transformationThe node feature vector of the layer,Is the jth node after linear transformationNode feature vectors of the layers; Calculating and adding directional entity perception bias terms ; The directional entity aware bias termThe expression of (2) is: ; where lambda is a coefficient, sigmoid is a normalization function, As the relative distance between the words of node i and node j,As the entity vector of the node i,An entity vector for node j; the attention score of the node i and the node j is the sum of the basic attention score and the directional entity perception bias term, and the expression is: ; for the node i, carrying out softmax normalization on the attention scores of all the neighbor nodes j to obtain the final attention weight Then, feature polymerization is carried out; Performing the attention score calculation and feature aggregation on all nodes in the fully connected graph in parallel to obtain the first node Node characteristic matrix of all nodes of layerExecuting an iteration process, and finally, outputting a final high-order node characteristic matrix of each node after the