Full life cycle management method and system based on nuclear power construction element data
Through natural language processing technology and ontological construction language, a map of nuclear power construction elements is constructed, which solves the problem of low knowledge acquisition and update efficiency in traditional technologies, and realizes efficient integration of nuclear power construction knowledge and decision-making support, improving the level of automation and intelligence.
Patent Information
- Application Number
- CN202410599627.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-15
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-05-15
AI Technical Summary
Traditional textual materials and experiences are difficult to fully express the knowledge of nuclear power construction elements, it is difficult to efficiently obtain and integrate knowledge, and the knowledge update and model maintenance are inefficient, lack of interpretability and transparency, and it is difficult to meet the needs of multilingual understanding and complex semantic processing.
By extracting nuclear power construction knowledge from unstructured texts, a unified knowledge model is built, and natural language processing technologies such as entity recognition, dependency analysis, semantic role annotation and relationship perception machines are used to construct nuclear power construction feature maps in combination with ontology construction language, and extract tasks Dependency and resource occupation state, obtain the optimal scheduling scheme based on the constraint optimization model.
It realizes automatic and efficient extraction of nuclear power construction knowledge from massive unstructured texts, integrates knowledge from different channels, provides interpretable decision support, improves the efficiency of knowledge acquisition and the scientificity and automation of decision-making, and reduces the cost of manual participation.
Smart Images

Figure CN118364968B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data management, and in particular to a full life cycle management method and system based on nuclear power construction element data. Background Art
[0002] Nuclear power construction involves complex engineering design, construction organization and project management, which requires full consideration of the interrelationships and constraints between various elements. Traditional text materials and experience reliance are difficult to fully express all knowledge. How to efficiently obtain knowledge of nuclear power construction elements and make scientific decisions based on this is an urgent problem to be solved.
[0003] Traditional text data management is usually trained for a single task, with poor generalization ability, and it is difficult to meet the needs of understanding multiple languages at the same time. The model has limited processing capabilities for long-distance dependencies and complex semantic phenomena, and requires a large amount of high-quality annotated data, which is relatively expensive to obtain. The construction of knowledge graphs mainly relies on the experience and rules of domain experts. The construction process is time-consuming and labor-intensive, with poor scalability and reusability, and it is difficult to fully cover all knowledge elements. When knowledge updates and model maintenance face high-dimensional and complex constraints, the solution efficiency is low, and the solution process lacks explainability and transparency.
[0004] Therefore, there is an urgent need for a new technical solution that can automatically and efficiently extract nuclear power construction knowledge from massive unstructured texts, integrate and build a unified knowledge model, and provide efficient and explainable decision support based on this to guide the scientific construction of nuclear power projects. Summary of the invention
[0005] The embodiments of the present invention provide a full life cycle management method and system based on nuclear power construction element data, which can at least solve some problems in the prior art.
[0006] According to a first aspect of the embodiments of the present invention,
[0007] Provides a full life cycle management method based on nuclear power construction element data, including:
[0008] Collect nuclear power construction element data, determine element text, construct a word vector sequence based on words, parts of speech, and word context in the element text, input the word vector sequence into an entity recognizer, and determine element entities; based on the element text, determine syntactic dependency relationships through a dependency parser, construct a syntactic dependency graph, and simultaneously determine semantic relationships through a semantic role tagger in combination with the element entities, construct a semantic graph, input the syntactic dependency graph and the semantic graph into a trained relational perceptron, and determine element relationship types;
[0009] Based on the requirements of nuclear power construction, an ontology model is constructed using an ontology construction language, the classes, attributes and relationships corresponding to the ontology model are determined, the types corresponding to the element entities are mapped and aligned with the classes of the ontology model, and the element relationship types are matched and connected with the attributes and relationships of the ontology model, a structured knowledge representation is constructed, and a nuclear power construction element map is determined;
[0010] Based on the nuclear power construction project, the project data is determined, the project data is imported into the nuclear power construction element map, the task dependencies and resource occupancy status are extracted, the decision goals and constraints are determined, and based on the pre-built target constraint solving model, the optimal scheduling plan that meets the constraints is obtained. According to the optimal scheduling plan, the construction scheduling of the nuclear power construction project is executed.
[0011] In an optional implementation, the entity identifier includes:
[0012] Define a state probability matrix and a path pointer matrix, determine that each state probability element in the state probability matrix represents the maximum probability of reaching the target element entity category and outputting a word sequence at a corresponding time, and at the same time determine that each path pointer element in the path pointer matrix records the previous state of the optimal path to reach the target element entity category at the corresponding time;
[0013] Based on the initial state probability vector, determining the first column of the state probability matrix and the first column of the path pointer matrix, and simultaneously initializing the bundle;
[0014] At each moment, calculate all possible transition states and corresponding scores of each state in the bundle, select the K states with the highest scores as the new bundle, and use the state in the bundle as the optimal state at the current moment, and update the path pointer matrix at the same time;
[0015] In the bundle corresponding to the last moment, the state with the highest probability is selected as the terminal state of the approximate optimal path;
[0016] The approximate optimal path is gradually traced back from the terminal state to obtain a feature entity annotation sequence, the feature entity annotation sequence is post-processed to remove illegal annotations, and adjacent feature entities of the same category are merged to determine the final feature entity sequence and complete the construction of the entity identifier.
[0017] In an optional implementation, based on the element text, a dependency parser is used to determine the syntactic dependency relationship, a syntactic dependency graph is constructed, and a semantic role tagger is used to determine the semantic relationship in combination with the element entity, a semantic graph is constructed, and the syntactic dependency graph and the semantic graph are input into a trained relational perceptron. Determining the element relationship type includes:
[0018] Using the dependency parser to perform dependency parsing on the element text to obtain syntactic dependency relations, and constructing a syntactic dependency graph according to the syntactic dependency relations, wherein the nodes of the syntactic dependency graph represent words in the element text, and the edges of the syntactic dependency graph represent dependency relations between words;
[0019] Using the semantic role annotator to semantically annotate the element text, based on the element entity, identifying the semantic relationship between words, obtaining a semantic role annotating result, and constructing a semantic graph according to the semantic role annotating result, wherein the nodes of the semantic graph represent the words in the element text, and the edges of the semantic graph represent the semantic relationship between words;
[0020] The syntactic dependency graph and the semantic graph are input into a relational perceptron, which constructs a graph convolutional network and obtains a structured representation of each node in the syntactic dependency graph and the semantic graph through convolution operations of multiple convolutional layers, and superimposes the structured representation with the context representation obtained by identifying each word in the word vector sequence to determine the comprehensive representation of the words and extract the element relationship type.
[0021] In an optional implementation, the relationship sensor includes:
[0022] Initialize the graph convolutional network according to the preset number of convolutional layers and hidden units;
[0023] Perform convolution operations on each convolution layer separately:
[0024] For each node in the syntactic dependency graph and the semantic graph, determine a node representation, determine a neighbor node corresponding to the node, aggregate the neighbor nodes, and obtain a neighbor representation;
[0025] The neighbor representation is concatenated with the node representation, and a nonlinear activation function is applied to obtain a fused node representation, and a weight matrix is used to perform a linear transformation on the fused node representation to obtain an updated node representation;
[0026] Obtaining structured representations corresponding to the syntactic dependency graph and the semantic graph through a global pooling process;
[0027] Based on LSTM, the context representation of each word in the word vector sequence is identified, the context representation and the structured representation are added to determine the comprehensive representation of the word, the relationship classification is performed on the comprehensive representation of the word, the relationship type between the words is determined, and the element relationship type is obtained.
[0028] In an optional implementation, based on the nuclear power construction requirements, an ontology model is constructed using an ontology construction language, classes, attributes and relationships corresponding to the ontology model are determined, types corresponding to the element entities are mapped and aligned with classes of the ontology model, and at the same time, the element relationship types are matched and connected with attributes and relationships of the ontology model, a structured knowledge representation is constructed, and determining a nuclear power construction element map includes:
[0029] Based on the requirements of nuclear power construction, the ontology model is constructed by using the ontology construction language, by determining the class hierarchy, the domain and value range of the attribute, the type and constraint of the relationship;
[0030] Based on the feature entity, the corresponding fine-grained type is parsed to determine the entity type. The similarity between the entity type and the class in the ontology model is calculated by combining the word vector similarity with the string matching degree, the best matching class is determined, and the mapping alignment from entity to class is established;
[0031] Based on the element relationship type, in combination with the semantic relationship and the dependency relationship, the semantic type and semantic direction corresponding to the element relationship type are captured, the similarity with the attributes and relationships in the ontology model is determined, the most matching attribute or the most matching relationship is determined, and relationship-to-attribute and relationship-to-relationship matching connections are established;
[0032] According to the mapping alignment result from entity to class, each entity is converted into an instance of the corresponding class and a unique identifier is generated; according to the matching connection result from relationship to attribute or relationship to relationship, each relationship is converted into an attribute or relationship between the instances, and based on the simple association relationship constructed from the instance, the attribute and the relationship, an ontology element structure corresponding to the simple association relationship is generated;
[0033] If there is a multi-element association relationship or a nested relationship, an intermediate node and a composite attribute are introduced to construct a multi-element composite association relationship, and an ontology element structure corresponding to the multi-element composite association relationship is generated;
[0034] Based on the ontology element structure and in combination with the unique identifier, a structured element representation is formed to construct an element graph.
[0035] In an optional implementation, based on a nuclear power construction project, determining project data, importing the project data into the nuclear power construction element map, extracting task dependencies and resource occupancy status, determining decision goals and constraints, and obtaining an optimal scheduling solution that meets the constraints based on a pre-built target constraint solving model includes:
[0036] Acquire task data, resource data and dependency data of a nuclear power construction project, construct a resource description triple, and import the resource description triple into the nuclear power construction element map;
[0037] For each of the tasks, create a task instance and fill in corresponding attribute values;
[0038] For each resource, create a resource instance and fill in corresponding attribute values;
[0039] For the dependency relationship between tasks, create a dependency instance, associate the corresponding predecessor task with the corresponding successor task, determine the end time of the predecessor task and the start time of the successor task, and determine the first constraint condition that the start time of the successor task is greater than or equal to the end time of the predecessor task;
[0040] For the resources occupied by the task, an occupation relationship between the task instance and the resource instance is created to determine whether the task occupies the resource and the available quantity of the resource, and to determine a second constraint condition that the sum of the resource occupation quantities of all tasks occupying the same resource is less than or equal to the available quantity of the resource;
[0041] Based on the task data, the resource data and the dependency data, the decision objectives and constraints are determined, and a constraint satisfaction model is constructed by combining the decision variables and the objective function. The first constraint and the second constraint are integrated into the constraint satisfaction model, and the constraint satisfaction model is solved based on a preset solver to obtain an optimal scheduling solution that satisfies the constraints.
[0042] According to a second aspect of the embodiments of the present invention,
[0043] Provide a full life cycle management system based on nuclear power construction element data, including:
[0044] The first unit is used to collect nuclear power construction element data, determine element text, construct a word vector sequence based on words, parts of speech, and word context in the element text, input the word vector sequence into an entity recognizer, and determine the element entity; based on the element text, determine the syntactic dependency relationship through a dependency parser, construct a syntactic dependency graph, and at the same time, determine the semantic relationship through a semantic role tagger in combination with the element entity, construct a semantic graph, input the syntactic dependency graph and the semantic graph into a trained relational perceptron, and determine the element relationship type;
[0045] The second unit is used to construct an ontology model based on the nuclear power construction requirements using an ontology construction language, determine the classes, attributes and relationships corresponding to the ontology model, map and align the types corresponding to the element entities with the classes of the ontology model, and match and connect the element relationship types with the attributes and relationships of the ontology model, construct a structured knowledge representation, and determine the nuclear power construction element map;
[0046] The third unit is used to determine project data based on the nuclear power construction project, import the project data into the nuclear power construction element map, extract task dependencies and resource occupancy status, determine decision goals and constraints, obtain the optimal scheduling plan that meets the constraints based on a pre-built target constraint solving model, and execute the construction scheduling of the nuclear power construction project according to the optimal scheduling plan.
[0047] According to a third aspect of the embodiments of the present invention,
[0048] An electronic device is provided, comprising:
[0049] processor;
[0050] a memory for storing processor-executable instructions;
[0051] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0052] According to a fourth aspect of the embodiments of the present invention,
[0053] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.
[0054] Through natural language processing technology, knowledge elements such as entities, facts and semantic relationships related to nuclear power construction are automatically extracted from a large amount of unstructured text, greatly improving the efficiency of knowledge acquisition. Compared with the traditional manual rule method, the heterogeneous knowledge obtained from different channels is integrated into a unified structured representation through steps such as building an ontology model, knowledge mapping alignment and knowledge graph construction. This fusion method is conducive to fully exploring and utilizing all knowledge resources in the field.
[0055] At the same time, based on the constructed knowledge graph and specific project data, a constrained optimization model is proposed to formalize the modeling decision objectives and constraint rules, and an efficient algorithm is designed to obtain the optimal decision solution that meets multiple constraints. Compared with traditional experience-driven decision-making, it provides explainable, efficient and scientific decision support, which runs through the entire process from text to knowledge to decision-making, greatly improves the automation and intelligence level of related links, and reduces the cost of manual participation. The natural language processing model and knowledge modeling paradigm have certain domain independence and can be migrated and reused in other fields after appropriate adjustments, which is conducive to reducing development and maintenance costs and improving the universality and sustainable development capabilities of related technologies. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 It is a flowchart of a full life cycle management method based on nuclear power construction element data according to an embodiment of the present invention;
[0057] Figure 2 It is a structural schematic diagram of a full life cycle management system based on nuclear power construction element data according to an embodiment of the present invention. DETAILED DESCRIPTION
[0058] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0059] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.
[0060] Figure 1 FIG. 1 is a flow chart of a method for managing the entire life cycle of nuclear power construction element data according to an embodiment of the present invention. Figure 1 As shown, the method includes:
[0061] S101. Collect nuclear power construction element data, determine element text, construct a word vector sequence based on words, parts of speech, and word context in the element text, input the word vector sequence into an entity recognizer, and determine element entities; based on the element text, determine syntactic dependency relationships through a dependency parser, construct a syntactic dependency graph, and simultaneously determine semantic relationships through a semantic role tagger in combination with the element entities, construct a semantic graph, input the syntactic dependency graph and the semantic graph into a trained relational perceptron, and determine element relationship types;
[0062] The entity recognizer comprises:
[0063] Define a state probability matrix and a path pointer matrix, determine that each state probability element in the state probability matrix represents the maximum probability of reaching the target element entity category and outputting a word sequence at a corresponding time, and at the same time determine that each path pointer element in the path pointer matrix records the previous state of the optimal path to reach the target element entity category at the corresponding time;
[0064] Based on the initial state probability vector, determining the first column of the state probability matrix and the first column of the path pointer matrix, and simultaneously initializing the bundle;
[0065] At each moment, calculate all possible transition states and corresponding scores of each state in the bundle, select the K states with the highest scores as the new bundle, and use the state in the bundle as the optimal state at the current moment, and update the path pointer matrix at the same time;
[0066] In the bundle corresponding to the last moment, the state with the highest probability is selected as the terminal state of the approximate optimal path;
[0067] The approximate optimal path is gradually traced back from the terminal state to obtain a feature entity annotation sequence, the feature entity annotation sequence is post-processed to remove illegal annotations, and adjacent feature entities of the same category are merged to determine the final feature entity sequence and complete the construction of the entity identifier.
[0068] The state probability matrix means that each element represents the maximum probability of reaching the target feature entity category and outputting a word sequence at the corresponding time. The path pointer matrix means that each element records the previous state of the optimal path to reach the target feature entity category at the corresponding time. According to the initial state probability vector, the first column of the state probability matrix and the first column of the path pointer matrix are determined.
[0069] The initialization bundle represents the optimal candidate path retained at that moment. When performing the bundle search iteration, all possible transition states and corresponding scores of each state in the bundle are calculated at each moment, and the K states with the highest scores are selected as the new bundle. The state in the new bundle is then used as the optimal state at the current moment, and the path pointer matrix is updated.
[0070] Backtracking the approximate optimal path means selecting the state with the highest probability in the corresponding beam at the last moment as the terminal state of the approximate optimal path, and gradually backtracking the path pointer matrix from the terminal state to obtain the feature entity annotation sequence. After removing illegal annotations and merging adjacent and identical feature entities, the final feature entity sequence is determined to complete the construction of the entity recognizer. This process utilizes the beam search algorithm to approximate the optimal path through a limited beam width, thereby efficiently completing the sequence annotation task and obtaining the feature entity recognition result.
[0071] In this embodiment, based on the element text, the syntactic dependency relationship is determined by a dependency parser, and a syntactic dependency graph is constructed. At the same time, a semantic role annotator is used to determine the semantic relationship in combination with the element entity, and a semantic graph is constructed. The syntactic dependency graph and the semantic graph are input into a trained relational perceptron. Determining the element relationship type includes:
[0072] Using the dependency parser to perform dependency parsing on the element text to obtain syntactic dependency relations, and constructing a syntactic dependency graph according to the syntactic dependency relations, wherein the nodes of the syntactic dependency graph represent words in the element text, and the edges of the syntactic dependency graph represent dependency relations between words;
[0073] Using the semantic role annotator to semantically annotate the element text, based on the element entity, identifying the semantic relationship between words, obtaining a semantic role annotating result, and constructing a semantic graph according to the semantic role annotating result, wherein the nodes of the semantic graph represent the words in the element text, and the edges of the semantic graph represent the semantic relationship between words;
[0074] The syntactic dependency graph and the semantic graph are input into a relational perceptron, which constructs a graph convolutional network and obtains a structured representation of each node in the syntactic dependency graph and the semantic graph through convolution operations of multiple convolutional layers, and superimposes the structured representation with the context representation obtained by identifying each word in the word vector sequence to determine the comprehensive representation of the words and extract the element relationship type.
[0075] The relation sensor comprises:
[0076] Initialize the graph convolutional network according to the preset number of convolutional layers and hidden units;
[0077] Perform convolution operations on each convolution layer separately:
[0078] For each node in the syntactic dependency graph and the semantic graph, determine a node representation, determine a neighbor node corresponding to the node, aggregate the neighbor nodes, and obtain a neighbor representation;
[0079] The neighbor representation is concatenated with the node representation, and a nonlinear activation function is applied to obtain a fused node representation, and a weight matrix is used to perform a linear transformation on the fused node representation to obtain an updated node representation;
[0080] Obtaining structured representations corresponding to the syntactic dependency graph and the semantic graph through a global pooling process;
[0081] Based on LSTM, the context representation of each word in the word vector sequence is identified, the context representation and the structured representation are added to determine the comprehensive representation of the word, the relationship classification is performed on the comprehensive representation of the word, the relationship type between the words is determined, and the element relationship type is obtained.
[0082] Dependency parser is an important tool in the field of natural language processing. It is used to automatically analyze the syntactic structure of sentences and the dependency relationship between words, such as identifying the part of speech of each word in a sentence (such as noun, verb, etc.), determining the main body of the sentence (core word), finding which word each word depends on, and determining the type of dependency relationship between words (such as subject, object, attributive, etc.). Dependency grammar believes that a sentence is composed of a series of dependency relationships. Each word (except the core word of the sentence) depends on another word. The dependency relationship reflects the modifying and modified relationship between words.
[0083] LSTM is a special recurrent neural network that aims to solve the long-term dependency problem of traditional recurrent neural networks. The long-term dependency problem refers to the difficulty of the network to effectively capture the long-distance dependency between time steps in very long sequence data, which causes the gradient of the model to vanish or explode during the learning process. This greatly limits the performance of traditional methods when processing sequence data such as speech and text. LSTM solves the long-term dependency problem well by finely controlling the flow of information. It shows excellent performance when processing long sequence data and is widely used in natural language processing tasks such as machine translation, speech recognition, and text generation.
[0084] Use the dependency parser to perform dependency parsing on the feature text to obtain syntactic dependency relations, and then construct a syntactic dependency graph based on the dependency relations. The nodes represent the words in the feature text, and the edges represent the dependency grammatical relations between the words. Then, a semantic graph is constructed by a semantic role annotator. The semantic role annotator is used to semantically annotate the feature text, identify the semantic relations between the words based on the feature entities, and construct a semantic graph based on the semantic role annotating results after obtaining the semantic role annotating results.
[0085] For example, a graph convolutional network (GCN) is introduced to model text as a syntactic dependency graph or semantic graph, capturing the structured information between entities and improving the accuracy of relationship extraction. To construct a syntactic dependency graph or semantic graph of a text, a dependency parser (such as Stanford CoreNLP, etc.) is used to perform dependency parsing on the text to obtain syntactic dependency relationships, and then a directed graph is constructed based on the syntactic dependency relationships, where nodes represent words in the text and edges represent dependency relationships between words. For a semantic graph, a semantic role labeling (SRL) tool can be used to label the text with semantic roles. After identifying the semantic relationships between words, a directed graph is constructed based on the semantic role labeling results, where nodes represent words in the text and edges represent semantic relationships between words.
[0086] The construction of a graph convolutional network (GCN) first defines the number of layers and hidden units of the GCN, initializes the model parameters, and then for each layer of GCN, executes each node (word) in the graph, aggregates the representations of its neighbor nodes, which can be achieved using adjacency matrices and aggregation functions (such as summation, average, etc.), and then concatenates or adds the aggregated neighbor representations with the representation of the current node, and applies a nonlinear activation function (such as ReLU), and then uses the weight matrix to perform a linear transformation on the updated node representation to obtain a new node representation. In the last layer of GCN, the representation of each node is globally pooled (such as maximum pooling, average pooling) to obtain the representation of the entire graph.
[0087] When GCN is integrated into the relation extraction model, GCN is used as an additional feature extractor in the LSTM-based relation extraction model to capture the structured information between entities. The text is input into the LSTM to obtain the contextual representation of each word, and the syntactic dependency graph or semantic graph of the text is input into the GCN to obtain the structured representation of each node (word). The output of the LSTM and the output of the GCN are concatenated or added as the comprehensive representation of the entity, and then the comprehensive representation of the entity is input into the relation classification layer to predict the relationship type between entities.
[0088] For example, when training and optimizing the model, training data with relationship labels are used, and the model parameters are optimized through the back-propagation algorithm and gradient descent. During the training process, the parameters of LSTM and GCN are updated simultaneously to adapt to each other and capture the semantic and structural information between entities. Then, regularization techniques (such as L2 regularization, Dropout, etc.) are used to prevent the model from overfitting, and then suitable optimizers (such as Adam) and learning rate scheduling strategies (such as learning rate decay) are used to control the convergence speed and generalization performance of the model.
[0089] Through the above steps, the graph convolutional network (GCN) can be introduced into the relation extraction model, and the syntactic dependency graph or semantic graph of the text can be used to capture the structured information between entities and improve the accuracy of relation extraction. As a powerful graph neural network, GCN can effectively model and propagate information of graph structure, and show good performance improvement in relation extraction tasks. At the same time, GCN also introduces new perspectives and ideas for relation extraction, opening up the research direction of combining structured information and neural networks.
[0090] S102. Based on the nuclear power construction requirements, an ontology model is constructed using an ontology construction language, and the classes, attributes and relationships corresponding to the ontology model are determined. The types corresponding to the element entities are mapped and aligned with the classes of the ontology model, and the element relationship types are matched and connected with the attributes and relationships of the ontology model, so as to construct a structured knowledge representation and determine the nuclear power construction element map;
[0091] In this embodiment, an ontology model is constructed using an ontology construction language, classes, attributes and relationships corresponding to the ontology model are determined, types corresponding to the element entities are mapped and aligned with classes of the ontology model, and element relationship types are matched and connected with attributes and relationships of the ontology model to construct a structured knowledge representation, and determining a nuclear power construction element map includes:
[0092] Based on the requirements of nuclear power construction, the ontology model is constructed by using the ontology construction language, by determining the class hierarchy, the domain and value range of the attribute, the type and constraint of the relationship;
[0093] Based on the feature entity, the corresponding fine-grained type is parsed to determine the entity type. The similarity between the entity type and the class in the ontology model is calculated by combining the word vector similarity with the string matching degree, the best matching class is determined, and the mapping alignment from entity to class is established;
[0094] Based on the element relationship type, in combination with the semantic relationship and the dependency relationship, the semantic type and semantic direction corresponding to the element relationship type are captured, the similarity with the attributes and relationships in the ontology model is determined, the most matching attribute or the most matching relationship is determined, and relationship-to-attribute and relationship-to-relationship matching connections are established;
[0095] According to the mapping alignment result from entity to class, each entity is converted into an instance of the corresponding class and a unique identifier is generated; according to the matching connection result from relationship to attribute or relationship to relationship, each relationship is converted into an attribute or relationship between the instances, and based on the simple association relationship constructed from the instance, the attribute and the relationship, an ontology element structure corresponding to the simple association relationship is generated;
[0096] If there is a multi-element association relationship or a nested relationship, an intermediate node and a composite attribute are introduced to construct a multi-element composite association relationship, and an ontology element structure corresponding to the multi-element composite association relationship is generated;
[0097] Based on the ontology element structure and in combination with the unique identifier, a structured element representation is formed to construct an element graph.
[0098] The ontology model refers to the structured and standardized expression and modeling of conceptual knowledge in the field of nuclear power construction, including: determining the class hierarchy, that is, the domain concept classification system; defining the attributes of the class, including the domain and value range of the attribute; determining the various relationship types between concepts and their constraints; and organically organizing elements such as classes, attributes, instances, and relationships.
[0099] Feature entities refer to key entity words or phrases such as task entities, resource entities, organizational entities, time entities, and location entities that are automatically identified from text corpora describing the nuclear power construction process, tasks, resources, etc. through natural language processing technologies such as named entity recognition, term extraction, and relationship extraction.
[0100] Element relationship refers to the automatic identification and extraction of key relationship information such as task dependency, resource consumption, organizational relationship, time relationship, and spatial location relationship from texts describing nuclear power construction processes, task dependencies, resource consumption, etc. through natural language processing technologies such as relationship extraction, event extraction, and semantic role labeling.
[0101] For example, when constructing an ontology model, you can use tools such as Protege, TopBraid Composer, WebProtege, etc. to build the ontology from scratch, or expand the domain ontology based on the existing general ontology (such as DOLCE, SUMO). For example, starting from the abstract concept of DOLCE, you can introduce specific entity classes for nuclear power plant construction, such as:
[0102] Physical Object ->Construction ->Nuclear Plant;
[0103] Role ->Construction Role ->Design Role, Construction Worker;
[0104] Process ->Nuclear Plant Construction Process;
[0105] We can search and reuse the already built ontology resources in related fields such as construction engineering and project management, and integrate them with the scene characteristics of nuclear power engineering. For example, we can use the Industry Foundation Classes (IFC) standard organized by buildingSMART, which has defined a large number of concepts and relationships related to buildings and facilities, and can serve as the basis for ontology construction.
[0106] By using natural language processing and ontology learning technology, we can automatically extract conceptual terms, entities and relationship information from corpora such as standard architectural design documents and engineering construction guidelines to assist in ontology construction. For example, we can use term extraction to identify the nuclear power engineering concept "reactor containment", identify the positional relationship of "is-inside" from "rebar is located inside the reactor containment", and use pattern learning to discover the "has-part" relationship between "buffer pool" and "containment".
[0107] This automatic extraction combined with manual review and correction can accelerate the construction of ontology. In the process of ontology construction, it is necessary to follow some common design specifications and best practices, such as structured and modular design, naming conventions, and maintaining sufficient semantic richness, to ensure the quality of the ontology model. For example, when defining the "reactor" class of a nuclear power plant, the necessary attributes and constraints should be clearly given, such as core temperature range, cooling mode, operating status, etc., to make it more in line with actual business scenarios.
[0108] By comprehensively utilizing these methods and practices, combined with the participation of domain experts, a high-quality ontology model can be gradually constructed, which will help provide a knowledge-driven application foundation for nuclear power plant construction projects.
[0109] Mapping alignment refers to the process of mapping the unstructured element entities and element relationships extracted from the original text to the corresponding concepts and relationships in the pre-built structured ontology model. The purpose is to symbolize the unstructured text element knowledge and associate it with standardized ontology concepts so that it can be converted into a structured knowledge representation to provide support for subsequent queries, reasoning, applications, etc.
[0110] When mapping and aligning, the extracted feature entities will be mapped to the ontology class. The key is to design an efficient similarity calculation method. Here, a comprehensive strategy of word vector similarity and string edit distance is used. Word vectors are used to capture semantic information, and string information is combined for comprehensive scoring to achieve accurate matching. For the mapping of relationships to attributes / relationships, it is necessary to match according to the semantic type and direction of the relationship. The prediction results of the relationship classifier can be used, or pattern matching can be performed based on rules.
[0111] Based on the mapping alignment results, the feature entities are instantiated into ontology individuals, and the relationships are instantiated into attribute associations or relationship associations, thereby generating a simple binary association structure. For multi-dimensional or nested relationships, it is necessary to introduce technologies such as intermediate nodes and composite attributes to build a more complex ontology structure to fully express the relationship semantics in the original text. Finally, the standardized ontology structure data is persistently stored to form a structured feature knowledge base based on the knowledge graph. This graph database representation retains rich semantic information such as entities, concepts, attributes, and relationships, and is also convenient for subsequent retrieval, reasoning, and intelligent applications.
[0112] Exemplarily, a text fragment mention is extracted from the original text to represent an entity or relationship. First, the text string of the mention and the list of ontology class concept names are obtained, and then the semantic similarity between the mention and each class concept is calculated. The pre-trained word vector model such as Word2Vec is used to convert the mention and the class name into a word vector representation, and the cosine similarity between the mention word vector and the class noun vector is calculated as the semantic similarity score. The string edit distance between the mention and the class concept name is calculated, and then the semantic similarity score and the edit distance score are weighted and averaged as the comprehensive similarity score, and the class concept with the highest comprehensive score is selected as the mapping target.
[0113] The mapping of relations to attributes / relationships uses relation classifiers to predict the semantic type of relation mention phrases, such as sequence annotations or classifiers based on semantic models such as BERT. According to the relation type, matching attribute / relationship sets are searched in the ontology, such as "use" corresponds to usedIn / usedFor and other attributes, the syntactic structure of the mention is analyzed, the direction of the relationship (subject-object) is determined, and for each candidate attribute / relationship, the semantic similarity score with the mention is calculated based on its definition, and the attribute / relationship with the highest similarity and matching direction is selected as the mapping target. The pattern rules summarized manually can also be used for direct matching.
[0114] When instantiated into an ontology structure, the entities mapped to the class will be created as ontology individuals (OWL individuals), and the relationships mapped to the attributes will be instantiated as OWL data attribute associations, and the relationships mapped to the relationships will be instantiated as OWL object attribute associations. For n-ary relationships, intermediate nodes are introduced to model the relationship, and composite attributes are used to connect. For nested relationships, intermediate concept classes are created to represent semantics in a hierarchical manner.
[0115] Ontology data persistence: The constructed ontology individuals, class concepts, attributes, relationships and other OWL structured data are stored persistently in a relational database or RDF graph database using triple storage solutions such as Apache Jena and AllegroGraph, thus forming a computable, knowledge graph-based structured knowledge base.
[0116] In the actual system, it is also necessary to consider: incremental update and version management of ontology models, synonym mapping and normalization conflict handling, data quality issues such as errors, conflicts, sparsity, etc. The process of this embodiment converts unstructured text knowledge into a structured ontology model form, which is a key link in realizing knowledge base construction and intelligent applications. It requires comprehensive semantic analysis, ontology modeling, data processing and other technical means. This method converts natural language element knowledge into structured and formalized knowledge representation, which is the basis for supporting subsequent knowledge-driven intelligent applications.
[0117] S103. Based on the nuclear power construction project, determine the project data, import the project data into the nuclear power construction element map, extract task dependencies and resource occupancy status, determine decision goals and constraints, and obtain the optimal scheduling plan that meets the constraints based on a pre-built target constraint solving model. According to the optimal scheduling plan, execute the construction scheduling of the nuclear power construction project.
[0118] In this embodiment, based on the nuclear power construction project, obtaining the optimal scheduling solution that meets the constraint conditions includes:
[0119] Acquire task data, resource data and dependency data of a nuclear power construction project, construct a resource description triple, and import the resource description triple into the nuclear power construction element map;
[0120] For each of the tasks, create a task instance and fill in corresponding attribute values;
[0121] For each resource, create a resource instance and fill in corresponding attribute values;
[0122] For the dependency relationship between tasks, create a dependency instance, associate the corresponding predecessor task with the corresponding successor task, determine the end time of the predecessor task and the start time of the successor task, and determine the first constraint condition that the start time of the successor task is greater than or equal to the end time of the predecessor task;
[0123] For the resources occupied by the task, an occupation relationship between the task instance and the resource instance is created to determine whether the task occupies the resource and the available quantity of the resource, and to determine a second constraint condition that the sum of the resource occupation quantities of all tasks occupying the same resource is less than or equal to the available quantity of the resource;
[0124] Based on the task data, the resource data and the dependency data, the decision objectives and constraints are determined, and a constraint satisfaction model is constructed by combining the decision variables and the objective function. The first constraint and the second constraint are integrated into the constraint satisfaction model, and the constraint satisfaction model is solved based on a preset solver to obtain an optimal scheduling solution that satisfies the constraints.
[0125] Exemplarily, task data is extracted from the project management system / document, including task name, estimated duration, required resources, etc., resource data is collected, such as the name, type, available quantity, etc. of various resources, task dependency data is obtained, predecessor and successor tasks of each task are determined, resource triples are constructed, and for each resource, its key attributes, such as type, subtype, unit, etc., are determined, and the resource and its attributes are constructed as triples in the form of (subject, relation, object), for example: (rebar, type, building material); (rebar, unit, ton); (cement, type, building material).
[0126] For tasks, you can also build triples that describe their attributes, such as: (concrete pouring, construction period, 5 days). For dependencies, you can build triples, such as: (concrete pouring, predecessor task, steel bar binding). After the construction is completed, use the knowledge graph tool to import the triples. You can use knowledge graph tools and libraries such as Apache Jena and Neo4j to import triples of resources, tasks, and dependencies into the knowledge graph. The graph will form a structure of nodes (resources, tasks) and relationships (attributes, dependencies).
[0127] When processing data cleaning and entity linking, it is necessary to clean and normalize the triple data, complete the entity linking work of the same entity mention and eliminate the redundant representation of the same entity. Finally, the constructed nuclear power construction knowledge graph is persistently stored, which can be stored as RDF triple data or imported into the graph database to prepare for subsequent query, reasoning and application.
[0128] Through these steps, we can build a formalized knowledge graph based on the original nuclear power construction project data, laying the foundation for subsequent scheduling decision analysis based on the knowledge graph.
[0129] To create a task instance and fill in attributes, first traverse the task data, create an OWL individual instance for each task, and create an OWL individual instance for each resource. Use tool libraries such as Apache Jena to operate ontology / knowledge graph data, set rdf:type for the task instance to the constructed "Task" type, and set the following data attribute values for the instance according to the task data:
[0130] Task name: rdfs:label;
[0131] Estimated duration: :estimatedDuration (customizable data attribute);
[0132] Other attributes: such as person in charge, remarks, etc.
[0133] Set rdf:type to "Resource" for the resource instance;
[0134] Set the following data attribute values for the instance:
[0135] Resource name: rdfs:label;
[0136] Resource type: :resourceType (customizable object property);
[0137] Available quantity: :availableQuantity;
[0138] Traverse the task dependency data, create an OWL individual instance for each dependency, and set the instance type to "Dependency".
[0139] Create two object attribute associations:
[0140] hasPredecessor is associated with the predecessor task instance;
[0141] hasSuccessor is associated with the post-task instance;
[0142] Set the data attributes:
[0143] constrainedStartTimeDelay (decision variable);
[0144] When creating a task-resource occupation relationship, the occupation data of tasks and resources are also traversed. Using the created task instances and resource instances, an instance of the OWL object attribute: occupy is created. The task instance and the resource instance are associated with the occupy attribute, and the occupation quantity data attribute: occupyAmount is set for the occupation instance.
[0145] Import all Dependency instances and the constrainedStartTimeDelay decision variable. For each Dependency instance, add a linear constraint: successor.startTime - predecessor.endTime>=constrainedStartTimeDelay. Import occupancy instances and resource availability. For each resource, add a linear constraint: sum(occupancy of all tasks occupying the resource)<= resource availability. Set the decision goal: min(max(endTime of allTasks)) to minimize the total project duration to build a constraint satisfaction model.
[0146] To solve the constraint satisfaction model, an optimization solver SDK, such as Python's CPLEX library, is used to import the variables, objective function, and linear constraints in the model, call the solver, and execute the goal planning algorithm to obtain the solution result, including the values of all decision variables. Based on the values of the decision variables, the optimal start time of each task is determined and output as the final optimal scheduling solution.
[0147] This embodiment formalizes various hard constraints in project management (such as task dependency, resource availability, etc.) into model constraints, and then combines them with decision-making goals (such as shortening the construction period, etc.) to build a constraint optimization model, and uses a solver to calculate the optimal solution that satisfies all constraints. This method has strong optimization capabilities and interpretability compared to manual scheduling.
[0148] Figure 2 FIG. 1 is a schematic diagram of a full life cycle management system based on nuclear power construction element data according to an embodiment of the present invention. Figure 2 As shown, the system comprises:
[0149] The first unit is used to collect nuclear power construction element data, determine element text, construct a word vector sequence based on words, parts of speech, and word context in the element text, input the word vector sequence into an entity recognizer, and determine the element entity; based on the element text, determine the syntactic dependency relationship through a dependency parser, construct a syntactic dependency graph, and at the same time, determine the semantic relationship through a semantic role tagger in combination with the element entity, construct a semantic graph, input the syntactic dependency graph and the semantic graph into a trained relational perceptron, and determine the element relationship type;
[0150] The second unit is used to construct an ontology model based on the nuclear power construction requirements using an ontology construction language, determine the classes, attributes and relationships corresponding to the ontology model, map and align the types corresponding to the element entities with the classes of the ontology model, and match and connect the element relationship types with the attributes and relationships of the ontology model, construct a structured knowledge representation, and determine the nuclear power construction element map;
[0151] The third unit is used to determine project data based on the nuclear power construction project, import the project data into the nuclear power construction element map, extract task dependencies and resource occupancy status, determine decision goals and constraints, obtain the optimal scheduling plan that meets the constraints based on a pre-built target constraint solving model, and execute the construction scheduling of the nuclear power construction project according to the optimal scheduling plan.
[0152] According to a third aspect of the embodiments of the present invention,
[0153] An electronic device is provided, comprising:
[0154] processor;
[0155] a memory for storing processor-executable instructions;
[0156] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0157] According to a fourth aspect of the embodiments of the present invention,
[0158] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.
[0159] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.
[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A full life cycle management method based on nuclear power construction element data, characterized in that: include: Collect nuclear power construction element data, determine element text, construct a word vector sequence based on words, parts of speech, and word context in the element text, input the word vector sequence into an entity recognizer, and determine element entities; based on the element text, determine syntactic dependency relationships through a dependency parser, construct a syntactic dependency graph, and simultaneously determine semantic relationships through a semantic role tagger in combination with the element entities, construct a semantic graph, input the syntactic dependency graph and the semantic graph into a trained relational perceptron, and determine element relationship types; Based on the requirements of nuclear power construction, an ontology model is constructed using an ontology construction language, the classes, attributes and relationships corresponding to the ontology model are determined, the types corresponding to the element entities are mapped and aligned with the classes of the ontology model, and the element relationship types are matched and connected with the attributes and relationships of the ontology model, a structured knowledge representation is constructed, and a nuclear power construction element map is determined; Based on the nuclear power construction project, determine the project data, import the project data into the nuclear power construction element map, extract the task dependency and resource occupancy status, determine the decision-making goals and constraints, obtain the optimal scheduling plan that meets the constraints based on the pre-built target constraint solving model, and execute the construction scheduling of the nuclear power construction project according to the optimal scheduling plan; The entity recognizer comprises: Define a state probability matrix and a path pointer matrix, determine that each state probability element in the state probability matrix represents the maximum probability of reaching the target element entity category and outputting a word sequence at a corresponding time, and at the same time determine that each path pointer element in the path pointer matrix records the previous state of the optimal path to reach the target element entity category at the corresponding time; Based on the initial state probability vector, determining the first column of the state probability matrix and the first column of the path pointer matrix, and simultaneously initializing the bundle; At each moment, calculate all possible transition states and corresponding scores of each state in the bundle, select the K states with the highest scores as the new bundle, and use the state in the bundle as the optimal state at the current moment, and update the path pointer matrix at the same time; In the bundle corresponding to the last moment, the state with the highest probability is selected as the terminal state of the approximate optimal path; Step by step, the approximate optimal path is traced back from the terminal state to obtain a feature entity annotation sequence, the feature entity annotation sequence is post-processed to remove illegal annotations, and adjacent feature entities of the same category are merged to determine a final feature entity sequence, thereby completing the construction of an entity identifier; Based on the element text, a syntactic dependency relationship is determined by a dependency parser, and a syntactic dependency graph is constructed. Meanwhile, a semantic role annotator is used to determine a semantic relationship in combination with the element entity, and a semantic graph is constructed. The syntactic dependency graph and the semantic graph are input into a trained relational perceptron. Determining the element relationship type includes: Using the dependency parser to perform dependency parsing on the element text to obtain syntactic dependency relations, and constructing a syntactic dependency graph according to the syntactic dependency relations, wherein the nodes of the syntactic dependency graph represent words in the element text, and the edges of the syntactic dependency graph represent dependency relations between words; Using the semantic role annotator to semantically annotate the element text, based on the element entity, identifying the semantic relationship between words, obtaining a semantic role annotating result, and constructing a semantic graph according to the semantic role annotating result, wherein the nodes of the semantic graph represent the words in the element text, and the edges of the semantic graph represent the semantic relationship between words; The syntactic dependency graph and the semantic graph are input into a relational perceptron, and the relational perceptron obtains a structured representation of each node in the syntactic dependency graph and the semantic graph by constructing a graph convolutional network and performing convolution operations of multiple convolutional layers, and superimposes the structured representation with the context representation obtained by identifying each word in the word vector sequence to determine a comprehensive representation of the words and extract the element relationship type; The relation sensor comprises: Initialize the graph convolutional network according to the preset number of convolutional layers and hidden units; Perform convolution operations on each convolution layer separately: For each node in the syntactic dependency graph and the semantic graph, determine a node representation, determine a neighbor node corresponding to the node, aggregate the neighbor nodes, and obtain a neighbor representation; The neighbor representation is concatenated with the node representation, and a nonlinear activation function is applied to obtain a fused node representation, and a weight matrix is used to perform a linear transformation on the fused node representation to obtain an updated node representation; Obtaining structured representations corresponding to the syntactic dependency graph and the semantic graph through a global pooling process; Based on LSTM, the context representation of each word in the word vector sequence is identified, the context representation and the structured representation are added to determine the comprehensive representation of the word, the comprehensive representation of the word is classified into relationships, the relationship type between the words is determined, and the element relationship type is obtained; Based on the requirements of nuclear power construction, an ontology model is constructed using an ontology construction language, and the classes, attributes and relationships corresponding to the ontology model are determined. The types corresponding to the element entities are mapped and aligned with the classes of the ontology model, and the element relationship types are matched and connected with the attributes and relationships of the ontology model to construct a structured knowledge representation. The nuclear power construction element map is determined to include: Based on the requirements of nuclear power construction, the ontology model is constructed by using the ontology construction language, by determining the class hierarchy, the domain and value range of the attribute, the type and constraint of the relationship; Based on the feature entity, the corresponding fine-grained type is parsed to determine the entity type. The similarity between the entity type and the class in the ontology model is calculated by combining the word vector similarity with the string matching degree, the best matching class is determined, and the mapping alignment from entity to class is established; Based on the element relationship type, in combination with the semantic relationship and the dependency relationship, the semantic type and semantic direction corresponding to the element relationship type are captured, the similarity with the attributes and relationships in the ontology model is determined, the most matching attribute or the most matching relationship is determined, and relationship-to-attribute and relationship-to-relationship matching connections are established; According to the mapping alignment result from entity to class, each entity is converted into an instance of the corresponding class and a unique identifier is generated; according to the matching connection result from relationship to attribute or relationship to relationship, each relationship is converted into an attribute or relationship between the instances, and based on the simple association relationship constructed from the instance, the attribute and the relationship, an ontology element structure corresponding to the simple association relationship is generated; If there is a multi-element association relationship or a nested relationship, an intermediate node and a composite attribute are introduced to construct a multi-element composite association relationship, and an ontology element structure corresponding to the multi-element composite association relationship is generated; Based on the ontology element structure and in combination with the unique identifier, a structured element representation is formed to construct an element graph.
2. The method according to claim 1, characterized in that Based on the nuclear power construction project, the project data is determined, the project data is imported into the nuclear power construction element map, the task dependency and resource occupancy status are extracted, the decision target and constraint conditions are determined, and based on the pre-built target constraint solving model, the optimal scheduling solution that meets the constraint conditions is obtained, including: Acquire task data, resource data and dependency data of a nuclear power construction project, construct a resource description triple, and import the resource description triple into the nuclear power construction element map; For each of the tasks, create a task instance and fill in corresponding attribute values; For each resource, create a resource instance and fill in corresponding attribute values; For the dependency relationship between tasks, create a dependency instance, associate the corresponding predecessor task with the corresponding successor task, determine the end time of the predecessor task and the start time of the successor task, and determine the first constraint condition that the start time of the successor task is greater than or equal to the end time of the predecessor task; For the resources occupied by the task, an occupation relationship between the task instance and the resource instance is created to determine whether the task occupies the resource and the available quantity of the resource, and to determine a second constraint condition that the sum of the resource occupation quantities of all tasks occupying the same resource is less than or equal to the available quantity of the resource; Based on the task data, the resource data and the dependency data, the decision objectives and constraints are determined, and a constraint satisfaction model is constructed by combining the decision variables and the objective function. The first constraint and the second constraint are integrated into the constraint satisfaction model, and the constraint satisfaction model is solved based on a preset solver to obtain an optimal scheduling solution that satisfies the constraints.
3. A full life cycle management system based on nuclear power construction element data, used to implement the full life cycle management method based on nuclear power construction element data described in any one of claims 1-2, characterized in that: include: The first unit is used to collect nuclear power construction element data, determine element text, construct a word vector sequence based on words, parts of speech, and word context in the element text, input the word vector sequence into an entity recognizer, and determine the element entity; based on the element text, determine the syntactic dependency relationship through a dependency parser, construct a syntactic dependency graph, and at the same time, determine the semantic relationship through a semantic role tagger in combination with the element entity, construct a semantic graph, input the syntactic dependency graph and the semantic graph into a trained relational perceptron, and determine the element relationship type; The second unit is used to construct an ontology model based on the nuclear power construction requirements using an ontology construction language, determine the classes, attributes and relationships corresponding to the ontology model, map and align the types corresponding to the element entities with the classes of the ontology model, and match and connect the element relationship types with the attributes and relationships of the ontology model, construct a structured knowledge representation, and determine the nuclear power construction element map; The third unit is used to determine project data based on the nuclear power construction project, import the project data into the nuclear power construction element map, extract task dependencies and resource occupancy status, determine decision goals and constraints, obtain the optimal scheduling plan that meets the constraints based on a pre-built target constraint solving model, and execute the construction scheduling of the nuclear power construction project according to the optimal scheduling plan.
4. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 2.
5. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 2 is implemented.
