Intelligent simulation model framework construction method

Abstract

The invention relates to an intelligent simulation model framework construction method. The method comprises the following steps: S1, completing model label design; s2, constructing a model system architecture specification; s3, based on the model label and the model system architecture specification, establishing a knowledge graph intelligent recommendation model; and S4, completing generation and development of a simulation model by using the knowledge graph intelligent recommendation model. According to the invention, through a label design method, a multi-dimensional and multi-feature model system label library is constructed, a model system knowledge graph is formed, and a label-based model system framework intelligent recommendation capability is provided, so that generation, development and reconstruction of a simulation model are realized; the problems that the model development learning cost is high, the development efficiency is poor, and the intelligent level is low are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of simulation and modeling, and in particular to a method for constructing an intelligent simulation model framework. Background Art

[0002] At present, modeling and simulation technology is one of the key technologies to promote the development of modern high technology and one of the key technologies to build an innovative country. It has great application potential and great application value in the fields of national defense and national economy. The model system framework has important engineering value for the construction of simulation models. At present, in the fields of aviation, aerospace, ships, weapons, etc., the model system framework construction technology has been rapidly promoted, and the system modeling and simulation technology has been widely used in national defense military enterprises and institutions and scientific research and other manufacturing industries.

[0003] In the existing technology, the model system framework and specifications lack uniformity and are highly complex, resulting in high learning costs for model development, long development cycles, low development efficiency, many chimney-style constructions, and low intelligence levels. Models cannot be universalized and reused, and the integrity of the model cannot be reflected. The model system framework is not sound and complete, so the benefits of the model system framework cannot be demonstrated. This adds a lot of repetitive work to model developers and prolongs the development time of the system. Summary of the invention

[0004] In order to solve the technical problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a method for constructing an intelligent simulation model framework, which can solve the problems of high learning cost of model development, poor development efficiency and low intelligence level.

[0005] To achieve the above-mentioned object of the invention, the present invention provides a method for constructing an intelligent simulation model framework, comprising the following steps:

[0006] Step S1, complete model label design;

[0007] Step S2: construct model system architecture specifications;

[0008] Step S3: Establish a knowledge graph intelligent recommendation model based on the model label and the model system architecture specification;

[0009] Step S4: Use the knowledge graph intelligent recommendation model to complete the generation and development of the simulation model.

[0010] According to a technical solution of the present invention, the step S1 specifically includes:

[0011] Use multi-level tree labels to design model category labels;

[0012] Describe the behavioral characteristics of each model in the form of labels, design based on the specific behavior of the model, and complete the model behavior label design;

[0013] Based on the specific parameter attributes of the model, it involves model parameter labels;

[0014] Design model rule labels based on a multi-level rule system;

[0015] Based on the command relationship and / or control relationship between models, design model association relationship labels.

[0016] According to a technical solution of the present invention, in step S2, the model system architecture specification includes at least a basic framework for determining standards and specifications at the software development level and a business framework for determining business aspects of the model system.

[0017] According to a technical solution of the present invention, step S3 specifically includes:

[0018] Step S31: construct a tag library based on the model tag, and construct a tag knowledge graph based on the tag library;

[0019] Step S32: Generate or reconstruct the model system framework based on the model system architecture specification;

[0020] Step S33: Complete componentized modeling and obtain a knowledge graph intelligent recommendation model.

[0021] According to a technical solution of the present invention, in step S31, data of the knowledge graph is formed according to the model type, behavior, parameter, rule, and association relationship labels to construct a label knowledge graph;

[0022] The data is divided into structured data and unstructured data. The structured data at least includes: description of the entity status, actions, and combat effects of the simulation model; the unstructured data at least includes: related tasks and control instructions;

[0023] The label knowledge graph is constructed in a bottom-up manner and stored using the Neo4j graph database.

[0024] According to a technical solution of the present invention, the step S4 specifically includes:

[0025] Step S41, tag retrieval based on semantic understanding;

[0026] Step S42: Model matching based on knowledge graph;

[0027] Step S43: Complete the generation and development of the simulation model according to the matching result.

[0028] According to a technical solution of the present invention, in step S41, the model description information is used as a query to output entities or other labels that are closest to the text in the query statement, specifically including:

[0029] Step S411: perform word segmentation processing on the user's model description information, decompose the sentence into search keywords, and match and convert them with the synonym dictionary;

[0030] Step S412: semantically expand the search keywords and construct new search statements;

[0031] Step S413, performing dependency syntax analysis to convert it into a search statement corresponding to the graph;

[0032] Step S414: sort the obtained search results by relevance, and return the results that best meet the user's actual model building intention.

[0033] According to a technical solution of the present invention, in step S412, the process of semantically expanding the search keyword includes at least entity tag expansion, and the entity tag expansion includes similarity calculation between the keyword and the entity and similarity calculation between entities;

[0034] Among them, the similarity between keywords and entity tags is calculated by combining the literal similarity calculation method and the semantic similarity calculation method. The specific formula is:

[0035] S(K, E) = λS vec (K, E)+(1-λ)S lit (k,E) (1)

[0036] Among them, λ is the weight parameter; K is the keyword; E is the entity noun;

[0037] S vec (K, E) represents the vector space similarity between the retrieval keyword and the entity label, and the calculation formula is:

[0038]

[0039] Among them, k is the word vector corresponding to the search keyword, and e is the word vector corresponding to the entity name. The larger the cosine value, the smaller the angle between the two word vectors, and the higher the similarity between the words;

[0040] S lit (k, E) represents the literal similarity between the search keyword and the entity label, and the calculation formula is:

[0041]

[0042] Where M(K, E) represents the maximum length of the search keyword K and the entity name string, and L(K, E) represents the Levenshtein edit distance between the search keyword and the entity label;

[0043] According to formula (1), the semantic similarity between the search keyword and the entity name is obtained, and the top N entities with the highest similarity are selected as search expansion terms;

[0044] The number of common attributes between the entities is used to determine whether there is a relationship between two entities. altr (E1, E2) represents the attribute similarity between entity E1 and entity E2, and the specific formula is:

[0045]

[0046] Among them, I(E1, E2) represents the number of common attributes of entities E1 and E2; U(E1, E2) represents the number of all attributes of entities E1 and E2;

[0047] Considering the distance between entity nodes and the number of common attributes between two nodes, the semantic similarity between entities can be expressed as:

[0048] S entity (E1, E2) = (1-μ)S altr (E1, E2) + μD(E1, E2) (5)

[0049] Among them, D(E1, E2) represents the shortest path between two entity nodes, and μ is the weight parameter.

[0050] According to a technical solution of the present invention, the step S42 specifically includes:

[0051] Step S421: According to the keyword tags extracted based on semantic understanding, the demand model is queried and screened from the knowledge graph through model matching of Cypher statements;

[0052] Step S422: sort the screened models using a matching algorithm based on semantic features, and recommend models with high matching degrees to users as a basis for simulation model development.

[0053] According to a technical solution of the present invention, in step S422, the matching recommendation algorithm based on semantic features constructs a feature vector according to each feature attribute value of the simulation model, and obtains the matching degree between the models by calculating the similarity of the feature vectors between the two models. The similarity calculation formula based on semantic features is:

[0054]

[0055] SE1 and SE2 represent different simulation models, Lv (SE1, SE2), then it is calculated by the following formula:

[0056]

[0057] n is the number of features in the model; x 1i is the attribute value of the ith feature of SE1; x 2i is the attribute value of the i-th feature of SE2.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] The present invention proposes a method for constructing an intelligent simulation model framework, which extracts knowledge according to model categories, behaviors, parameters, rules, associations, etc., establishes a model system framework knowledge graph, realizes the extraction of demand model information labels and component model matching recommendations, and on this basis, combines the characteristics of the new model to perform component-based flexible configuration, completes the generation and development of the new model, improves the intelligence and reuse of the modeling process, and solves the problems of high learning cost, poor development efficiency, and low intelligence level in model development.

[0060] The present invention introduces multi-dimensional model label design such as construction dimension, management dimension and usage dimension, and implements word segmentation of model information required by users based on intelligent recommendation of knowledge graph, extracts key labels, and realizes precise matching of simulation models through model matching based on Cypher statements and component model recommendation based on semantic features.

[0061] The present invention is based on the model system architecture specification, including interface types, interface requirements, protocol types, etc. It mainly determines the standards and specifications of the model system business, including command rules, field composition, basic protocols, basic ports, etc. When users implement models, expand models, and upgrade models, they implement them based on the basic framework and business framework, thereby realizing label-based model system framework modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0063] Figure 1 A schematic diagram of model category label dimensions in an implementation case according to the present invention is schematically shown;

[0064] Figure 2A flowchart of label disambiguation and fusion in an implementation example according to the present invention is schematically shown;

[0065] Figure 3 A schematic diagram of a basic knowledge graph in an implementation case according to the present invention is schematically shown;

[0066] Figure 4 A schematic diagram of an aircraft tag knowledge graph in an implementation case according to the present invention is schematically shown;

[0067] Figure 5 A schematic diagram of a tag retrieval process based on semantic understanding in an implementation case according to the present invention;

[0068] Figure 6 A diagram schematically showing an example of a Cypher query statement according to an embodiment of the present invention;

[0069] Figure 7 A schematic diagram schematically shows a process flow of constructing an intelligent simulation model framework according to an embodiment of the present invention;

[0070] Figure 8 A flowchart for constructing an intelligent simulation model framework according to an embodiment of the present invention is schematically shown. DETAILED DESCRIPTION

[0071] 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 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.

[0072] like Figures 1 to 8 As shown, a method for constructing an intelligent simulation model framework of the present invention comprises the following steps:

[0073] Step S1, complete model label design;

[0074] Step S2: construct model system architecture specifications;

[0075] Step S3: Establish a knowledge graph intelligent recommendation model based on the model label and the model system architecture specification;

[0076] Step S4: Use the knowledge graph intelligent recommendation model to complete the generation and development of the simulation model.

[0077] Step S1 specifically includes: Step S11, using multi-level tree labels to design model category labels;

[0078] For example, the category labels of airplanes include entity labels, action entities, aerial entities, sports categories, etc. At the same time, users can independently select the model type granularity to define labels. The finer the model classification granularity, the higher the accuracy of the category labels, and the more accurate the knowledge graph intelligent recommendation. In the model category label. Model category labels are mainly divided from three dimensions: model construction, management, and use. Figure 1 shown.

[0079] Define category labels according to the model construction dimension. The category labels of the model construction dimension can be divided into:

[0080] 1) Entity model; 2) Decision model; 3) Service model; 4) Action model; 5) Control model; 6) Physical model; 7) Algorithm model; 8) Interface object model.

[0081] Category labels are defined according to the model management dimension, and can be divided into: 1) aviation models; 2) aerospace models; 3) civil models; 4) military models; 5) behavioral rule models; 6) analysis and evaluation models.

[0082] Category labels are defined according to the model usage dimension, and the model usage dimension category labels can be divided into: 1) flight capability model; 2) management capability model; 3) information transmission capability model; 4) communication capability model; 5) confrontation capability model; 6) guarantee capability model.

[0083] Step S12, describe the behavior characteristics of each model in the form of a label, design based on the specific behavior of the model, and complete the model behavior label design; model behavior label, design based on the specific behavior of the model, describe the behavior characteristics of each model in the form of a label, including take-off, landing, cruising, movement, patrol, support, guarantee and other behaviors. Among them, when the behaviors of different types of models are the same, only one behavior label is used, such as: passenger planes and drones have flight behaviors in common, and flight labels are uniformly used. When designing behavior labels, a single model is defined by a combination of multiple labels.

[0084] Step S13, based on the specific parameter attributes of the model, involves model parameter labels; model parameter labels, based on the specific parameter attributes of the model, adopt a multi-label combination design, such as: aircraft, including attributes such as position, attitude, target characteristics, etc., according to the position, it is subdivided into longitude, latitude, and altitude; according to the attitude, it is subdivided into dive attitude, roll attitude, yaw attitude, etc.

[0085] Step S14, design model rule labels based on the multi-level rule system; Model rule labels are designed based on the multi-level rule system, and must be defined under the semantic environment defined by the general grammar and semantics, using specific terms, graphics, symbols and their specific arrangement methods to define the rule label names. Model rule labels can be divided into command planning rules, combat action rules, etc.

[0086] Step S15: Design model association relationship labels based on the command relationship and / or control relationship between models. The behaviors such as command, control, and issuing instructions between models are the association relationships between models. Model association relationship labels are used to describe multiple association relationships in the form of labels, including command association relationships, control association relationships, etc.

[0087] In step S2, a model system architecture specification is constructed.

[0088] The construction of simulation models needs to be based on a certain basic framework and business framework. The basic framework mainly determines the standards and specifications at the software development level, including interface types, interface requirements, protocol types, etc. The business framework mainly determines the standards and specifications of the business aspects of the model system, including command rules, field composition, basic protocols, basic ports, etc. When users implement, expand, and upgrade models, they implement them based on the basic framework and business framework, thereby realizing label-based model system framework modeling.

[0089] Modeling specifications refer to the standard framework and constraints for building models. The model system is based on the IOCE modeling specifications (I is input, O is output, C is calculation, and E is event)

[0090] Configuration port: used to parameterize the key model. The model reads the information of the port to complete the initialization work. The configuration can be passed from one model to another to complete the configuration distribution within the model.

[0091] Data port: The data exchange port between models. The sender actively sends, the receiver receives asynchronously, and calls back the data response function to process the data.

[0092] Service port: a remote method port between models. The initiator initiates a service call, the service provider calls back the service response function, performs the service and returns, and the initiator continues to run after receiving the return. The platform provides system services, such as creating new entities, establishing links, disconnecting links, applying for step length, obtaining simulation entity lists, obtaining port mapping lists, and opening new simulation branches.

[0093] Driver port: The port through which the model drives other or its own main callback functions, realizing workflow-style serial, parallel, and logical judgment branches, and can drive the core computing functions within the model to complete business work.

[0094] State port: The port for obtaining or setting the model state. The data is read actively, that is, "pulling" data. The data publisher does not explicitly publish data. When the data subscriber needs state data, it uses the GET method to obtain the port data without binding the response event. The data updater can use the SET method to update the state data.

[0095] In step S3, based on the model label and the model system architecture specification, a knowledge graph intelligent recommendation model is established, which specifically includes:

[0096] Step S31: construct a tag library based on the model tag, and construct a tag knowledge graph based on the tag library;

[0097] For the simulation model to be constructed, there are corresponding tags, which constitute a tag library. The tag creation is used to create tags based on the tag design and in accordance with the tag format to form a tag library.

[0098] 1) Tag editing

[0099] For the target simulation model, users can create various tags for the model. In the tag editing page, users can manually modify the automatically constructed tag attributes. At the same time, the original tag file import time display and attachment download function are provided to facilitate users to manually check and correct the tags.

[0100] 2) New tag disambiguation and fusion based on NLTK similarity calculation

[0101] After the knowledge extraction of tags is completed, the newly added structured tags cannot be directly imported into the tag library. The newly added tags need to be disambiguated and fused to eliminate the conflicts and similarities in the tag information before they can be updated into the library. The disambiguation and fusion process is as follows: Figure 2 shown.

[0102] a) Disambiguation

[0103] Tag disambiguation refers to eliminating the ambiguity of a word based on context information. This process can be achieved through tag matching and similarity calculation. To ensure that each tag in the tag library is unique, disambiguation is required, that is, to find out whether there are identical or similar tags in the existing tag library.

[0104] For similar tags, we can calculate text similarity based on the tag name, keywords or description information, determine the similarity between the new tag and the existing tag, and select the one with the highest similarity as the disambiguation result. For tags involving multiple dimensions, we can use multi-dimensional similarity calculation methods, comprehensively consider the similarity between each dimension of the tag, and use NLTK to disambiguate the new tags.

[0105] NLTK is a Python-based natural language processing toolkit for processing text data. The following is a summary of the steps for using NLTK:

[0106] First, you need to use pip to install NLTK, and then load the labeled text data to be processed into the Python program for preprocessing. This includes removing stop words, stemming, word vectorization, and other processing. These processes can make the text more standardized and comparable, thereby improving the accuracy of text similarity calculation.

[0107] Then, feature extraction is used to convert the preprocessed text into a feature vector, which is one of the key steps in calculating text similarity. Common feature extraction methods include the bag-of-words model, TF-IDF model, Word2Vec model, etc. These models can convert text into numerical vectors for subsequent similarity calculations.

[0108] Then, you can use a weighted method to adjust the weights of different features to more accurately determine the text similarity. Then select the disambiguation result based on the similarity. You can manually set the similarity threshold in the background. When the similarity exceeds the threshold, the two tags are considered similar and need to be disambiguated. In this process, other factors can be considered, such as the importance and historicity of the tag, for comprehensive judgment.

[0109] b) Fusion processing

[0110] After the disambiguation process, the disambiguation results need to be fused, that is, similar tags are merged into one and updated into the database.

[0111] First, the name, description, and keywords of the tag must be retained, and the name and description of the tag with higher similarity must be selected as the fused tag information. For the attribute information of the tag, various methods such as timeline and geographic information are used for fusion, such as time, place, and person, and conflict detection and fusion processing are required to avoid duplicate or inconsistent information. For the content information of the tag, it is necessary to judge and fuse it based on the similarity and importance between the tags, and it can be processed by machine and natural language processing technologies.

[0112] Then all the newly added tags are grouped according to their similarity. For each group of tags, one of the tags is selected as the representative, and other tags with higher similarity are merged to find the ambiguities and inconsistencies in the tags. During the merging process, it is necessary to avoid duplication and conflict of information.

[0113] The merged labels are proofread to ensure their accuracy and completeness. During the proofreading process, attention should be paid to aspects such as the format, wording, and grammar of the labels to ensure that the labels meet specifications and standards.

[0114] c) Update the storage

[0115] After the disambiguation and fusion process, the updated label information needs to be updated into the database to ensure the integrity and accuracy of the label database. The update process can be automated or manually reviewed to ensure the correctness of the label information.

[0116] Step S32: Generate or reconstruct the model system framework based on the model system architecture specification;

[0117] Step S33: Complete componentized modeling and obtain a knowledge graph intelligent recommendation model.

[0118] The basic building blocks of knowledge graph storage are triplets such as "entity-relationship-entity" or "entity-attribute-value pair". Entities are linked to each other through relationships to form a network-like knowledge structure. Figure 3 It is the simplest knowledge graph. M represents an entity or value pair, and L represents a relationship or attribute. Based on the coverage, knowledge graphs can be divided into open knowledge graphs and professional knowledge graphs. Open knowledge graphs focus on breadth, striving to integrate more entities and contain more knowledge.

[0119] Under the existing label design, the data of the knowledge graph is formed according to the model type, behavior, parameter, rule, and association relationship labels, including: massive structured data describing the entity status, action, combat effect, etc. of the simulation model, and a large amount of unstructured data such as related tasks and control instructions. These two parts of data form a data sample set to support the construction of the knowledge graph. The knowledge graph construction process is completed in two steps: fixed point and edge connection.

[0120] (1) Fixed point

[0121] "Fixed points" means determining the nodes in the network structure of the knowledge graph. Each node corresponds to an entity. The more complex the knowledge contained in the graph, the more nodes there are. Due to the large number of entities in the data sample set and the existence of a large amount of redundant data, in order to obtain an accurate knowledge graph structure more quickly, the nodes of the knowledge graph are added with labels, combined with their professional knowledge, and related entities are found to determine.

[0122] (2) Edge connection

[0123] "Edge connection" means determining the relationship between entities in the knowledge graph. In theory, there is a relationship between each node, but the strength of the relationship between entities determines whether the nodes need to be connected. The more entities there are in the graph, the more complex the knowledge, and the more complex the edges between nodes. The relationship between entities is represented by a directed edge, with the direction pointing from the starting point to the end point. When expressing uncertain knowledge, each edge also has a probability value, which indicates the probability value of the actual existence of the edge. In the construction of the knowledge graph, a set of data is first selected from the data sample set to form the edge. The probability value of the edge is obtained by statistical means to form a preliminary knowledge graph structure.

[0124] The label knowledge graph is constructed from the bottom up and stored in the Neo4j graph database. Neo4j is a high-performance NoSQL graph database that uses nodes and relationships to represent data. Nodes and relationships have attributes, and can be labeled through the label library to classify nodes and relationships. Taking the aircraft model as an example, the knowledge graph framework based on Neo4j is shown in the figure below. Figure 4 .

[0125] With the aircraft model label entity as the center, a graph network is established with other label entities according to the model's category, behavior, parameters, rules, associations, etc. The aircraft model is managed by labeling according to the three dimensions of construction, management, and use. From the construction dimension, the aircraft model can be classified step by step according to components and types, such as: entity model-action entity model-air entity model-aircraft entity model. The above aircraft label knowledge graph is only an example framework diagram, and it should be constructed according to the user's granular label design and the label scale of the label library.

[0126] The knowledge graph intelligent recommendation model is based on the label design of model categories, actions, attributes, rules, and associations. It creates new labels, builds a knowledge graph based on the label library, reconstructs the model system framework, and realizes model information matching and intelligent recommendation model labels based on the knowledge graph, which is convenient for the development and design of new models. The knowledge graph intelligent recommendation model uses the Neo4j graph database as a carrier to build a label knowledge graph containing model categories, behaviors, parameters and other types. It formally describes the model categories, behaviors and other features in the knowledge graph, extracts labels from user model information, and then uses model matching based on Cypher statements and model recommendation algorithms based on semantic features to achieve precise matching of simulation models.

[0127] For numerical calculation models, componentization technology is used to develop various simulation models through a unified component development template, which can ensure that different simulation models have a unified code framework and compilation environment, providing a basis for interaction between models. The simulation model adopts object-oriented thinking to split various types of equipment into a series of basic model components with universal significance. Different types of components have their own functional definitions and inheritance relationships. Modelers can flexibly configure components based on the intelligent recommendation model of the knowledge graph to achieve the development and customization of specific models.

[0128] The component model template development function meets the needs of users to develop models independently, automatically generates code frameworks that comply with IOCE standards in multiple languages, simplifies the platform learning cost, and enables developers to focus on the implementation of business codes. The software can generate multi-language, multi-platform model frameworks. Based on the model framework, users can complete the initial construction of the model by filling in the business code. Different model frameworks can be generated for models of different business types, supporting multiple development languages ​​and multiple operating platforms. Users can write business codes in the model framework to form model components.

[0129] In step S4, the generation and development of the simulation model is completed using the knowledge graph intelligent recommendation model, which specifically includes:

[0130] Step S41, tag retrieval based on semantic understanding;

[0131] The tag retrieval service based on semantic understanding is based on users who have the need to develop models. The model description information is used as a query, and the entity or other label that is closest to the text in the query statement is output. The retrieval process is as follows: First, the user's model description information is segmented, the sentence is decomposed into retrieval keywords, and it is matched and converted with a synonym dictionary; secondly, the retrieval keywords are semantically expanded and new retrieval statements are constructed; then dependency syntactic analysis is performed to convert it into a retrieval statement corresponding to the graph; finally, the obtained retrieval results are sorted by relevance, and the results that best meet the user's actual model building intention are returned. The specific process is as follows: Figure 5 shown.

[0132] 1) Search statement processing

[0133] a) Participle

[0134] General dictionaries cannot meet the word segmentation requirements of professional texts. Therefore, we use a Chinese word segmentation tool called HanLP that supports custom dictionaries. We use the labels in the knowledge graph to build a professional dictionary to improve the word segmentation effect.

[0135] b) Synonym conversion

[0136] In order to solve the problem of search failure caused by inconsistency between search terms and knowledge graph label names, the constructed knowledge graph is used to analyze and count the labels in it, and a synonym dictionary is built in combination with a general dictionary. The synonym dictionary can be used to convert search keywords into synonyms, thereby eliminating search failures caused by vocabulary differences.

[0137] 2) Semantic calculation method of multi-similarity fusion

[0138] a) Entity tag expansion

[0139] In extended retrieval, entity tag extension is a common implementation method, which can identify the potential needs of search users and enrich search results. The present invention divides entity extension into two cases: one is that the search keyword fails to successfully match the knowledge graph entity tag. At this time, it is necessary to calculate the similarity between the keyword and each entity in the knowledge graph to select entities with high similarity for search query; the other is that the search keyword successfully matches the entity in the knowledge graph. At this time, it is necessary to calculate the similarity between the entity and other entities to select entities with high similarity to expand the search results.

[0140] (1) Method for calculating similarity between keywords and entities

[0141] Traditional methods based on literal similarity calculation mainly perform matching calculations based on the structure of the text itself, and it is difficult to obtain the semantic level association between the two. In order to overcome this limitation, the present invention proposes a method that combines literal similarity calculation and semantic similarity calculation to calculate the similarity between keywords and entity tags.

[0142] S(K, E) = λS vec (K, E)+(1-λ)S lit (k,E) (1)

[0143] Where: λ is the weight parameter; K is the keyword; E is the entity noun.

[0144] S vec (K, E) represents the vector space similarity between the retrieval keyword and the entity label, and the calculation formula is:

[0145]

[0146] Among them: k is the word vector corresponding to the search keyword, and e is the word vector corresponding to the entity name. The larger the cosine value, the smaller the angle between the two word vectors, and the higher the similarity between the words.

[0147] S lit (k, E) represents the literal similarity between the search keyword and the entity label, and the calculation formula is:

[0148]

[0149] Where: M(K, E) represents the maximum length of the search keyword K and the entity name string. L(K, E) represents the Levenshtein edit distance between the search keyword and the entity label.

[0150] According to formula (1), the semantic similarity between the search keyword and the entity name is obtained, and the top N entities with the highest similarity are selected as search expansion terms.

[0151] (2) Method for calculating similarity between entities

[0152] The more attributes an entity has, the more information it contains. Whether two entities are related can be determined by the number of their common attributes. The more attributes two entities have in common, the more similar they are. altr (E1, E2) represents the attribute similarity between entities E1 and E2. The specific formula is as follows:

[0153]

[0154] Where: I(E1, E2) represents the number of common attributes of entities E1 and E2; U(E1, E2) represents the number of all attributes of entities E1 and E2.

[0155] Based on the distance between entity nodes and the number of common attributes between two nodes, the following method is proposed to calculate the semantic similarity between entities:

[0156] S entity (E1, E2) = (1-μ)S altr (E1, E2) + μD(E1, E2) (5)

[0157] Where: D(E1, E2) represents the shortest path between two entity nodes, and μ is the weight parameter.

[0158] b) Search statement conversion

[0159] Converting the semantically understood search statements into the Cypher query language in Neo4j is the key to finding search results from the graph. In view of the characteristics of the knowledge graph, the question template matching method is used to convert the search statements.

[0160] For some special query sentences (such as parameter comparison between weapons and equipment, multi-entity query, etc.), basic predicate attribute template matching cannot generate corresponding search statements, so the present invention classifies these sentences and designs specific matching templates. For example, when the identifiers such as "faster than ..." or "heavier than ..." appear in the search sentence, the corresponding Cypher statement will be generated and the WHERE statement will be added for comparative query. When performing the search sentence conversion, firstly, synonym conversion is performed according to the synonym dictionary, then it is classified according to the keywords therein, the query template is determined, and the corresponding Cypher statement is finally generated.

[0161] c) Sorting of search results

[0162] In order to enable users to quickly search for the information they need, it is necessary to sort the results that have been queried and return the results to users in descending order of relevance. The search results are divided into two types: ① Original query results: the query results that directly match the answers in the graph after the original sentence is processed by word segmentation and synonym conversion; ② Extended query results: the search results that are queried after the search sentence is expanded by entities. The algorithm for re-ranking the above two query results is as follows:

[0163] Input: Search text sentence Q

[0164] Output: retrieval result set R

[0165] 1) Perform semantic search on the search statement Q to obtain the original query result set R1 and the extended query result set R2;

[0166] 2) If R1 is not empty, the result of R1 is sorted before the result of R2;

[0167] 3) If R1 is empty, the similarity of each result is calculated according to formula (1); otherwise, the similarity is calculated according to formula (5);

[0168] 4) Calculate the number of other entities associated with each entity to measure the importance of the entity;

[0169] 5) A comprehensive linear weighted formula is used for the results of entity semantic similarity and entity importance to obtain a comprehensive score for each search result;

[0170] 6) Sort in descending order according to the score and return the result R.

[0171] Step S42: Model matching based on knowledge graph;

[0172] After building a model knowledge graph, we can then carry out model matching based on the knowledge graph. First, based on the keyword tags extracted based on semantic understanding, we query and filter out the required models from the knowledge graph through model matching of Cypher statements. Then, we use a matching algorithm based on semantic features to sort the filtered models, and recommend models with high matching degrees to users as the basis for model development.

[0173] Step S421: Model matching based on Cypher statements.

[0174] Cypher is the official query language of the Neo4j graph database. To perform model matching based on the knowledge graph, the required knowledge must first be retrieved from Neo4j using the Cypher language. Therefore, queries based on Cypher statements are an essential component of implementing model matching based on the knowledge graph. In order to provide rich model matching functions and meet diverse needs, multiple interfaces that provide different query matching functions are implemented based on Cypher statements, including strict matching based on node attributes, strict matching based on node relationships, fuzzy matching based on keywords in model descriptions, fuzzy matching based on node relationships, and direct matching of all model types. These interfaces receive the description information entered by the user and generate corresponding Cypher query statements. When matching models, different interfaces will be executed based on the model description information entered by the user. If the user selects fuzzy matching and the model description entered only includes the two pieces of information that the behavior type is patrol and the model category is aircraft, the fuzzy matching interface based on node relationships will be executed. This interface will generate the following: Figure 6 The Cypher query statement shown is:

[0175] Step S422: matching recommendation based on semantic features.

[0176] The matching recommendation algorithm based on semantic features constructs a feature vector according to the feature attribute values ​​of the simulation model, and obtains the matching degree between the models by calculating the similarity of the feature vectors between the two models. According to the label type of the knowledge graph, the simulation model has the following characteristics: model type, behavior, parameters, rules, and association relationships.

[0177] The similarity calculation formula based on semantic features is:

[0178]

[0179] SE1 and SE2 represent different simulation models, L v (SE1, SE2), then it is calculated by the following formula:

[0180]

[0181] n is the number of features in the model; x1i is the attribute value of the ith feature of SE1; x 2i is the attribute value of the i-th feature of SE2.

[0182] In terms of characteristic parameter values, since the characteristic parameters of the simulation model are mostly descriptive parameters rather than specific parameter values ​​represented by numbers, each different characteristic parameter needs to be further processed to convert the non-numeric characteristic parameters into numerical types.

[0183] Take the model category feature as an example. Model categories are defined hierarchically according to the coarseness of granularity, and different levels are assigned different scores. For example, in the category label of aircraft entity model, there are 4 different levels: entity model, action entity model, air entity model, and aircraft entity model. Different scores are assigned to the 4 levels: 1 for entity model, 2 for action entity model, 3 for air entity model, and 4 for aircraft entity model. These four scores are the parameter values ​​that can be taken by this feature attribute. The model matching value is calculated based on the parameter values ​​of multiple features. When the feature parameters are the same as the required model information description, it is the highest matching level.

[0184] Depending on the model description information input by the user and the matching method selected, the software executes different matching interfaces and returns matching results.

[0185] Step S43: Complete the generation and development of the simulation model according to the matching result.

[0186] Combined with the knowledge graph database and the model description information of user needs, the matching algorithm based on semantic features is used to calculate the model similarity, and the TOP-K recommendation strategy is adopted to sort the current model from high to low according to the degree of adaptability, and a recommendation list containing K corresponding models is generated and sent to the user for reference. Users can directly view the model's type, behavior, parameters, rules, associations and other labels, and based on this, combined with the characteristics of the new model, perform component-based flexible configuration to complete the development of the new model.

[0187] The present invention discloses a method for constructing an intelligent simulation model framework, the method comprising: step S1, completing model label design; step S2, constructing a model system architecture specification; step S3, establishing a knowledge graph intelligent recommendation model based on the model label and the model system architecture specification; step S4, using the knowledge graph intelligent recommendation model to complete the generation and development of the simulation model. The present invention, through the label design method, constructs a multi-dimensional and multi-feature model system label library, forms a model system knowledge graph, and provides a label-based model system framework intelligent recommendation capability, thereby realizing the generation, development and reconstruction of the simulation model, solving the problems of high model development learning cost, poor development efficiency and low intelligence level.

[0188] Furthermore, the present invention proposes a label retrieval method based on semantic understanding, which solves the problem of intelligent extraction of labels from demand model information; proposes a model matching based on Cypher statements and a model recommendation algorithm based on semantic features, which solves the problem that it is difficult for developers to skillfully and efficiently carry out model design and development work under the framework of a complex model system, and achieves intelligent recommendation, agile development, and reuse of models.

[0189] In addition, it should be noted that the present invention can be provided as a method, an apparatus or a computer program product. Therefore, the embodiments of the present invention can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0190] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0191] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0192] It should also be noted that, in the present invention, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0193] Finally, it should be noted that the above is a preferred embodiment of the present invention. It should be pointed out that although the preferred embodiment of the present invention has been described, for those skilled in the art, once the basic creative concept of the present invention is known, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the attached claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A method for constructing an intelligent simulation model framework, characterized in that: The following steps are involved: Step S1, complete model label design; Step S2: construct model system architecture specifications; Step S3: Establish a knowledge graph intelligent recommendation model based on the model label and the model system architecture specification; Step S4: Use the knowledge graph intelligent recommendation model to complete the generation and development of the simulation model.

2. The intelligent simulation model framework construction method according to claim 1, characterized in that: The step S1 specifically includes: Use multi-level tree labels to design model category labels; Describe the behavioral characteristics of each model in the form of labels, design based on the specific behavior of the model, and complete the model behavior label design; Based on the specific parameter attributes of the model, it involves model parameter labels; Design model rule labels based on a multi-level rule system; Based on the command relationship and / or control relationship between models, design model association relationship labels.

3. The intelligent simulation model framework construction method according to claim 2, characterized in that: In step S2, the model system architecture specification includes at least a basic framework for determining standards and specifications at the software development level and a business framework for determining business aspects of the model system.

4. The intelligent simulation model framework construction method according to claim 3 is characterized in that: The step S3 specifically includes: Step S31: construct a tag library based on the model tag, and construct a tag knowledge graph based on the tag library; Step S32: Generate or reconstruct the model system framework based on the model system architecture specification; Step S33: Complete componentized modeling and obtain a knowledge graph intelligent recommendation model.

5. The intelligent simulation model framework construction method according to claim 4 is characterized in that: In the step S31, data of the knowledge graph is formed according to the model type, behavior, parameter, rule, and association relationship labels to construct a label knowledge graph; The data is divided into structured data and unstructured data. The structured data at least includes: description of the entity state, action, and combat effect of the simulation model; the unstructured data at least includes: related tasks and control instructions; The label knowledge graph is constructed in a bottom-up manner and stored using the Neo4j graph database.

6. The intelligent simulation model framework construction method according to claim 1, characterized in that: The step S4 specifically includes: Step S41, tag retrieval based on semantic understanding; Step S42: Model matching based on knowledge graph; Step S43: Complete the generation and development of the simulation model according to the matching result.

7. The intelligent simulation model framework construction method according to claim 6, characterized in that: In step S41, the model description information is used as a query to output entities or other labels that are closest to the text in the query statement, specifically including: Step S411: perform word segmentation processing on the user's model description information, decompose the sentence into search keywords, and match and convert them with the synonym dictionary; Step S412: semantically expand the search keywords and construct new search statements; Step S413, performing dependency syntax analysis to convert it into a search statement corresponding to the graph; Step S414: sort the obtained search results by relevance, and return the results that best meet the user's actual model building intention.

8. The intelligent simulation model framework construction method according to claim 7, characterized in that: In the step S412, the process of semantically expanding the search keyword includes at least entity tag expansion, and the entity tag expansion includes similarity calculation between the keyword and the entity and similarity calculation between entities; Among them, the similarity between keywords and entity tags is calculated by combining the literal similarity calculation method and the semantic similarity calculation method. The specific formula is: S(K,E)=λS vec (K,E)+(1-λ)S lit (k,E) (1) Among them, λ is the weight parameter; K is the keyword; E is the entity noun; S vec (K, E) represents the vector space similarity between the search keyword and the entity label, which is calculated as: Among them, k is the word vector corresponding to the search keyword, and e is the word vector corresponding to the entity name. The larger the cosine value, the smaller the angle between the two word vectors, and the higher the similarity between the words; S lit (k,E) represents the literal similarity between the search keyword and the entity label, and the calculation formula is: Where M(K,E) represents the maximum length of the search keyword K and the entity name string, and L(K,E) represents the Levenshtein edit distance between the search keyword and the entity label; According to formula (1), the semantic similarity between the search keyword and the entity name is obtained, and the top N entities with the highest similarity are selected as search expansion terms; The number of common attributes between the entities is used to determine whether there is a relationship between two entities. altr (E1, E2) represents the attribute similarity between entity E1 and entity E2, and the specific formula is: Among them, I(E1,E2) represents the number of common attributes of entities E1 and E2; U(E1,E2) represents the number of all attributes of entities E1 and E2; Considering the distance between entity nodes and the number of common attributes between two nodes, the semantic similarity between entities can be expressed as: S entity (E1,E2)=(1-μ)S altr (E1,E2)+μD(E1,E2) (5) Among them, D(E1, E2) represents the shortest path between two entity nodes, and μ is the weight parameter.

9. The intelligent simulation model framework construction method according to claim 6, characterized in that: The step S42 specifically includes: Step S421: According to the keyword tags extracted based on semantic understanding, the demand model is queried and screened from the knowledge graph through model matching of Cypher statements; Step S422: sort the screened models using a matching algorithm based on semantic features, and recommend models with high matching degrees to users as a basis for simulation model development.

10. The intelligent simulation model framework construction method according to claim 9, characterized in that: In step S422, the matching recommendation algorithm based on semantic features constructs a feature vector according to the feature attribute values ​​of the simulation model, and obtains the matching degree between the models by calculating the similarity of the feature vectors between the two models. The similarity calculation formula based on semantic features is: Among them, SE1 and SE2 represent different simulation models, L v (SE1, SE2), then it is calculated by the following formula: n is the number of features in the model; x 1i is the attribute value of the ith feature of SE1; x 2i is the attribute value of the i-th feature of SE2.

Images