Complex equipment development demand structure information generation method based on knowledge graph

Abstract

The invention discloses a knowledge graph-based complex equipment development demand structure information generation method. Comprising the following steps: firstly, constructing a domain knowledge graph based on BIGRU-CRF and CasRel models, and establishing a product function expression model according to an extension primitive concept; secondly, on the basis of the product function expression model, the domain knowledge graph and the corrected cosine semantic similarity, demand structure solving is carried out on the complex equipment to be researched and developed, and structure information is sought to mine an innovative structure in a product design space; and finally, a node importance sorting method is provided and used for sorting the structural solutions, and leftmost demand structure information is generated. And finally, more accurate guidance is provided for the development process of complex equipment in the field.

Description

Technical Field

[0001] The present invention relates to the technical field of complex equipment development knowledge graph construction, and specifically to a method for generating complex equipment development requirement structure information based on a knowledge graph. Background Art

[0002] In the process of complex equipment development, complex equipment developers usually have low product development efficiency and difficulty due to problems such as non-standard operation, lack of design knowledge, and incomplete consideration of problems. Generally speaking, in order to improve the standardization of complex equipment development, enterprises will rely too much on the experience and knowledge of R&D personnel themselves and previous design plans. With the application of big data and artificial intelligence, there are a large number of innovative product structures on the Internet. If the functional requirements of the products to be designed can be realized in the process of complex equipment development, these innovative structures can be collected and fully utilized through knowledge graph technology, even in the face of complex equipment development plans, it will be beneficial for complex equipment developers to improve their own work efficiency and enhance the innovation vitality of the enterprise.

[0003] Traditional complex equipment development methods mainly include process-based methods, analogy-based methods and knowledge-based methods, but these methods have limitations. The process-based method uses behavior as an intermediary to connect structure and function to achieve the purpose of structure-to-function mapping, and then restores the top-level function of the product through the relationship between functions indicated by meta-functions by function merging operations, and finally achieves the purpose of understanding the development ideas of complex equipment. However, this design method requires a high level of knowledge reserves from designers. The analogy-based complex equipment development method can improve creativity by applying previous successful cases, but the analogy method will produce an over-reliance on similarity, the demand for knowledge is relatively high, and the limited scope of application may lead to design errors. The knowledge-based complex equipment development method is a method that emphasizes the use of professional knowledge and experience to guide and manage all stages of complex equipment development. This method expresses the designer's ideas, criteria and principles with explicit knowledge. Compared with the traditional process-based design process, it can better reflect product characteristics and meet the needs of modern design, but this method may ignore cross-domain innovation and consume more time. Therefore, active steps must be taken to improve the quality of design solutions, enhance objectivity, and improve efficiency. Summary of the invention

[0004] Aiming at the problems of limited objectivity in product function solution and lack of effective auxiliary tools for complex equipment development in existing complex equipment development, the present invention proposes a method for generating demand structure information for complex equipment development based on knowledge graph. Firstly, a domain knowledge graph is constructed based on BIGRU-CRF and CasRel models, and a product function expression model is established according to the concept of extension primitives; secondly, based on the product function expression model, domain knowledge graph and modified cosine semantic similarity, the demand structure of the product to be developed is solved; finally, a node importance ranking method is proposed and used to sort the structure solution and generate the product demand structure information. Finally, more accurate guidance is provided for the complex equipment development process in this field.

[0005] In order to achieve the above object, the present invention proposes a method for generating complex equipment development requirement structure information based on knowledge graph, comprising the following steps:

[0006] (1) Build a dataset and use the BIGRU-CRF and CasRel models to construct a domain knowledge graph.

[0007] 1.1) Select relevant patent texts, define the entity types to be extracted, and use the entity annotation method based on "BMES" to represent the entities;

[0008] 1.2) Define the type of relationship to be extracted and use the wizard annotation assistant to annotate the relationship;

[0009] 1.3) Use the BERT model to generate word vectors that fully integrate the training text information;

[0010] 1.4) Use the BIGRU-CRF model to complete the entity recognition task, the CasRel model to complete the relationship extraction task, and build a domain knowledge graph;

[0011] (2) Use the event model in the extensible primitive theory to represent the behaviors and functional objects in the product function expression model.

[0012] (3) Based on this behavior, the modified cosine semantic similarity is used to match similar relational words in the knowledge graph, obtain all tail entities of the relational words, and form a matrix.

[0013] (4) Extract the functional objects in the event model and compare them with all elements in the candidate matrix. Use the modified cosine semantic similarity to match them. Collect the elements that exceed the similarity threshold. Infer all head entities through the elements and the corresponding relational words to generate the structural solution set corresponding to the functional object. Circulate all behaviors and functional objects in the product functional expression model to infer the overall structural solution set.

[0014] (5) The knowledge graph and the set of demand structure solutions corresponding to product functions are used as input data and passed into the node importance evaluation method.

[0015] (6) Using the node importance evaluation calculation method based on network analysis, all elements in the structural solution set are sorted according to the knowledge graph and node importance.

[0016] (7) Starting from the first functional object in the overall functional set, select the element with the first ranking in the corresponding structural solution set, that is, the element with the highest node importance, and judge whether it can serve as the final demand structure information of the first functional object from multiple perspectives such as economy and convenience. If it can serve as the final demand structure information, put it into the final result set. If not feasible, continue to take out the second ranked element in the structural solution set and judge again, and so on.

[0017] (8) Obtain the final demand structure information of all functional objects in the overall functional set and comprehensively form a complex equipment development plan.

[0018] In the step (2), the product function expression model is built in the form of gradually splitting the structure from top to bottom, and each structure has different functional representations. The "behavior" in the product function expression model is a description of how the structure realizes its function. This framework of "structure-behavior-function" or "system-behavior-function" simplifies the description of complex equipment development models, and this structured approach is also convenient for computers to process. In the present invention, the splitting of the overall structure into subsystems and the further splitting of subsystems into structures are both represented by the composite metamodel in the extension theory; and the structure itself and the functions it exhibits are represented by the event metamodel in the extension theory, thereby completing the construction process of the product function expression model. In the product function expression model, the most important are the "behavior" and "function" objects. Only when these two objects are defined accurately can the system seek the structural information it needs based on the knowledge graph.

[0019] The present invention uses Γ(D) to represent the name of the complex equipment or the subsystem of the complex equipment, verb i represents the ith "behavior" of the complex equipment or the subsystem of the complex equipment, and noun i represents the ith functional object of the complex equipment or the subsystem of the complex equipment, then

[0020]

[0021] In step (3), a fixed set of verbs is defined. i and its corresponding noun i is a set of product functional units to be solved F i , denoted as Fi =[verb i ,noun i ], then the overall function set is M = {F1, F2, ..., F n}. In the domain knowledge graph, search and verb i The relationship names with high modified cosine semantic similarity are selected, and the tail entities containing these relationships are summarized to form the matrix B of the “behavior” and “function” objects. i , then B i =[b1,b2,...,b n ]. Among them, b i Indicates the i-th verb i Similar relationship names and corresponding matrices Represents all tail entities corresponding to the relation word.

[0022] Then for the i-th group of product function units to be solved F i , whose matrix B i for:

[0023]

[0024] The calculation process of the modified cosine semantic similarity in step (3) is:

[0025]

[0026] Among them, Sim(D1,D2) represents the adjusted cosine semantic similarity of two entities D1 and D2. Represents the semantic vector of entity D1, Represents the semantic vector of entity D2. represents the mean value of the semantic vector of entity D1, Represents the mean semantic vector of entity D2.

[0027] In step (4), the set B i Elements with a corrected cosine semantic similarity less than 0.8 After all are removed, search for the corresponding triple head entity in the corrected candidate matrix Composition of product functional unit F i The candidate solution matrix C i . Then the matrix is

[0028]

[0029] If C i If it is an empty set, the user can define the default implementation function unit F i structure.

[0030] Repeat the above steps until all functions in Γ(D) can find the candidate solution matrix C i , and then the overall structural solution set is obtained.

[0031] In step (6), a node importance evaluation calculation method based on network analysis is proposed, and all elements in the structural solution set are sorted according to the calculation results. The specific steps are as follows.

[0032] 6.1) Take a certain structural solution in the demand structural solution set corresponding to the product function as node v i , use Dijstra algorithm to calculate the shortest path dis(v between the current node and the remaining nodes i ,v j ).

[0033] 6.2) Calculate the current node v i The closeness center C(i) of .

[0034] Define dis(v i ,v j ) is the node v i To node v j The shorter the length, the greater the closeness to the center. i The closeness centrality can be expressed as C(i).

[0035]

[0036] 6.3) Based on the constructed domain knowledge graph, solve the current node v i Neighborhood δ ki With node key domain O i .

[0037] Define node v i The neighborhood of is the set of all nodes connected to it by edges, and the node neighborhood is δ ki . Node v s Neighborhood Indicates that v s The set of all nodes associated with edges; for node v t Neighborhood Indicates that v t The set of all nodes that are connected by edges. Then node v i The node key domain is expressed as

[0038]

[0039] 6.4) Traverse the neighborhood δ ki Any two nodes v p and v q , solve P(vp ,v q ).

[0040] P(v p ,v q )={{v p ,v q}∪{v p ,v i ,v q}∪{v p ,v′ i ,v q |v′ i ∈F i ∩v′ i ≠v i}}.

[0041] P(v p ,v q ) indicates that at node v i Neighborhood δ ki Any two nodes v p and v q The set of shortest paths between .

[0042] 6.5) Calculate the number of nodes passing through v i The number of shortest paths M(i) and the number of paths that do not pass through node v i The number of shortest paths N(i).

[0043] Define a node v with degree k i Neighborhood δ ki In the example, any pair of nodes passes through node v i The number of shortest paths is M(i), which does not pass through node v i The number of shortest paths is N(i). Assume that node v s and v t The shortest path between sj Combined with the node key domain, M(i) and N(i) can be expressed as

[0044] in

[0045] in

[0046] 6.6) Calculate the node v i The criticality K(i).

[0047]

[0048] 6.7) In the process of network analysis, due to the different network structures, the number and position of nodes and edges will affect the importance of each node in the entire network. In a complex network, the importance of central nodes and non-central nodes is different; whether the nodes are connected will also affect the importance. The larger the number of shortest paths of the node, the more important the node is. Therefore, we can define and calculate the node v i The importance D(i).

[0049]

[0050] Provided is a computer storage medium, which contains a computer program for executing a method for generating structural information of complex equipment development requirements based on a knowledge graph. When the program is executed, the program controls the device where the computer storage medium is located to implement the method for generating structural information of complex equipment development requirements based on a knowledge graph;

[0051] A processor is proposed. The processor is used to run a program, wherein when the program is run, the above-mentioned technical method for generating complex equipment development requirement structure information based on knowledge graph is executed;

[0052] The beneficial effects of the present invention are:

[0053] A method for generating structural information of complex equipment development requirements based on knowledge graph is proposed. The method first focuses on building a product function expression model to effectively process, store and express the deep knowledge of the designed product. After the model is built, the design knowledge stored in the knowledge graph is used to map the product function expression model with the functional objects in the knowledge graph to comprehensively explore the set of structural solutions that meet the functional requirements. This process essentially matches the structural framework in the product function expression model with the functional objects in the knowledge graph to mine the innovative structure in the complex equipment development space. After sorting, selecting and evaluating the mined structural information, the design scheme that best meets the requirements is finally formed. The method for generating structural information of complex equipment development requirements based on knowledge graph enables designers to process key knowledge in design tasks more flexibly and accurately, thereby effectively improving design efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Overall flow chart of the complex equipment development requirement structure information generation method based on knowledge graph.

[0055] Figure 2 Partial annotation of entity dataset.

[0056] Figure 3 Flowchart of the node importance evaluation method based on network analysis.

[0057] Figure 4Automated guided vehicle product solution diagram. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions 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. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. 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 use of terms in the present invention is intended to describe specific embodiments rather than to set limitations thereon. In the present invention and the related claims, singular terms such as "a", "the" and "the" also include their plural forms, unless the context clearly indicates otherwise. Such wording is intended to ensure that the form of the term does not restrict the understanding of the present invention and that it encompasses a variety of possible implementations.

[0060] Next, the specific implementation of the present invention will be described in detail and clearly with reference to the accompanying drawings. Figure 1 A method for generating complex equipment development requirement structure information based on knowledge graph is presented, which includes the following steps:

[0061] (1) Obtain relevant patents in the field of complex equipment, perform data annotation and processing, build a data set in the field of complex equipment, and then obtain the corresponding word vectors and term vectors, specifically:

[0062] 1.1) Taking the field of automated guided vehicles as an example, relevant patents in this field in 2020 and later are obtained. The relevant entity types are divided into five categories according to the complex equipment designers: automated guided vehicle name, functional components, components, implemented functions and subsystems; the annotation description of entity recognition task data is shown in Table 1:

[0063] Table 1 Data annotation for entity recognition task

[0064] Taking "The patent mentions an automated guided vehicle, which has a laser sensor for sensing obstacles in the environment. In addition, there is a navigation system to determine the position of the vehicle" as an example, use the Wizard Assistant for sequence labeling.

[0065] The result after annotation is as follows Figure 2As shown in the figure. For the functional component entity of "laser sensor", the following annotation method is adopted: the character "激" is represented by B-FC, indicating that "激" is the first character of the entity of this type of functional component; the characters "光", "传", "感", etc. are represented by M-FC, indicating that they are the middle characters of the entity of this type of functional component; the character "器" is represented by E-FC, indicating that "器" is the last character of the entity of this type of functional component. For the characters that do not want to be extracted and the punctuation marks that have no effect, the "S" notation is used. After annotation, the annotated dataset is finally obtained.

[0066] In the annotated dataset, the first column is each specific character that appears in the corpus, and the second column is one of "B", "M", "E", "S", and "O". When all the patent text data is annotated, more than 17,000 lines of entity recognition datasets in the field of automated guided vehicles are obtained;

[0067] 1.2) The construction of the text relationship dataset is consistent with the source of the entity dataset. To meet the requirements of knowledge mining and utilization in the process of complex equipment development, combined with the characteristics of entity annotation, the relationships are defined and the data is annotated for the collected data.

[0068] The relationships between entities are defined as three categories: composition, realization, and unknown relationship. The composition relationship means that the head entity can be composed of the tail entity, the realization relationship means that the head entity can realize the function of the tail entity, and the unknown relationship means that the relationship between the two entities is not clear and有待挖掘. The relationship annotation instructions are shown in Table 2:

[0069] Table 2 Relationship Annotation Instructions

[0070] The data format of the annotation examples is shown in Table 3, including the head entity, the tail entity, the relationship between the entities, and the sentence text. Similar to the division method of the entity recognition dataset, after sorting and summarizing, it is divided into two parts. One is for training the model, and the other is for verifying the relationship recognition effect of the model. Similarly, to enable the model to be fully trained, the ratio of the training model dataset to the verification model dataset is 9:1.

[0071] Table 3 Partial Annotation of the Relationship Dataset

[0072] 1.3) After the dataset construction is completed, the BERT model is used to obtain word vectors and character vectors. The specific process is as follows;

[0073] a) It means that a knowledge text sequence to be extracted is segmented by character, and each independent character is used as input and passed into BERT

[0074] b) After the text characters to be extracted are encoded by BERT, they can be converted into a 768-dimensional feature vector, i.e., h i ∈R 1 ×768 After linear transformation, the output of BERT can be mapped into a low-dimensional semantic feature vector.

[0075] 1.4) The semantic feature vector is passed into the BIGRU-CRF model and CasRel model for entity recognition and relationship extraction. The knowledge graph of the automated guided vehicle domain is obtained.

[0076] (2) Use the model in the extension primitive theory to express the product function expression object. Taking the field of automatic guided vehicles as an example, the specific construction process is as follows:

[0077] Note that during the model building process, the first column requires the user to provide a default structure so that when the final demand structure fails to be solved, the result is not NULL.

[0078] a)

[0079] b)

[0080] c)

[0081] d)

[0082] e)J 0-DHMK =[Magnetic sensor realizes positioning]

[0083] f)

[0084] g)

[0085] h)J 0-TXMK =[Wireless network realizes LAN transmission]

[0086] i)

[0087] j)

[0088] (3) Based on the "behavior-function" object in the product function expression model, in the driving module of the constructed automatic guided vehicle product function expression model, "provide" is the first action, and "rotational motion power" is the first functional object corresponding to it. This example is used for explanation.

[0089] In the knowledge graph, we search for relationship names with similar meanings to "provide". After querying, we find that there are three types of relationships with modified cosine semantic similarity greater than or equal to 0.8, namely, provide, realize, and pass. We obtain all the tail entities of the three types of relationships and form a matrix.

[0090] (4) Taking the first two relations "provide" and "realize" as examples, query all tail entities whose modified cosine semantic similarity with the tail entity named "rotational motion power" is greater than or equal to 0.8, and search for the corresponding triple head entity to generate the partial requirement structure solution set corresponding to the product functional object, and complete the functional requirement structure solution process of the automatic guided vehicle drive module. As shown in Table 4, the partial requirement structure solution set of the target functional object "rotational motion power" in the drive module is displayed. After repeated cycles, the overall structure solution set can be inferred.

[0091] Table 4 Demand structure solution set (partial) of the target function object of "rotational motion power" in the drive module

[0092] (5) The domain knowledge graph and the set of demand structure solutions corresponding to product functions are used as input data and passed into the node importance evaluation method, such as Figure 3 shown.

[0093] (6) Taking the requirement structure solution set of the "rotational motion power" target functional object of the automatic guided vehicle drive module as an example, the node importance is sorted and the top 10 items are selected for display. The following results can be obtained, which is convenient for finalizing the requirement structure information of the final complex equipment development plan, as shown in Table 5.

[0094] Table 5 Importance ranking of the solution set of the target functional requirement structure of “rotational motion power” in the driving module (top 10 items)

[0095] (7) Select the structure name ranked first in importance and judge whether it can serve as the required structure of the final functional object. If it can, it is directly put into the final result set; if it cannot, continue to take out the second one in the sorted list and judge again, and so on.

[0096] (8) Finally, the final demand structure information of all functional objects in the overall functional set is obtained. See Table 6.

[0097] Table 6 Automated guided vehicle complex equipment development requirements information

[0098] By comparing the final complex equipment development function results with the initial function results, in the drive module, the initial structure of the rotation and direction-changing function object has changed from a wheel to a Magnum wheel, indicating that the patent text has been more interested in the Magnum wheel structure in the past three years. Compared with traditional wheels, the Mecanum wheel can achieve omnidirectional movement and achieve self-rotation while translating in any direction in the plane; in the navigation module, the initial structure of the positioning function object has changed from radar to SLAM laser navigation. It can be seen that this laser technology has become the mainstream in the field of AGV navigation at this stage. Compared with traditional magnetic sensors, SLAM has a high map accuracy and no cumulative error, which can be directly Used for positioning and navigation; in the load-bearing module, based on actual needs, you can choose a specific structure that meets the business scenario to transport and clamp goods, and replace the fixed pallet with the load bracket to improve the efficiency of transporting goods in specific scenarios; in the safety control module, use a buffer structure instead of a rubber strip to provide more solid protection for the automatic guided vehicle and its loads; in the communication module, use visible light wireless communication instead of traditional wireless networks to achieve signal stability; in the energy module, the use of lithium batteries is conducive to extending the cycle life, environmental protection, and better safety; in the auxiliary module, the ammeter is replaced by a touch screen, which is more convenient for users to read. By observing Table 4-6, it can be found that among the 15 functional objects, 8 innovative demand structures were replaced based on the knowledge graph, and the proportion of improved structures reached 53.33%; after calculation, it was found that the total node importance of the scheme increased by 0.164, and the average node importance of each structure increased by 0.021. Finally, based on Table 4-6, the product structure selection can be further carried out to draw a product solution diagram for the automatic guided vehicle, see Figure 4 The circle in the figure represents the design structure, and the square represents the implemented functional object.

[0099] The above embodiments are only used to illustrate the design concepts and features described herein, and are intended to enable technical professionals in the same field to understand the connotation of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above specific examples, and technicians can make appropriate modifications and changes as needed to adapt to different applications and implementation scenarios, and these changes and modifications are still included in the legal protection of the present invention.

Claims

1. A method for generating complex equipment development requirement structure information based on knowledge graph, comprising the following steps: (1) Build a data set and use the BIGRU-CRF and CasRel models to build a domain knowledge graph. 1.1) Select relevant patent texts, define the entity types to be extracted, and use the entity annotation method based on "BMES" to represent the entities; 1.2) Define the type of relationship to be extracted and use the wizard annotation assistant to annotate the relationship; 1.3) Use the BERT model to generate word vectors that fully integrate the training text information; 1.4) Use the BIGRU-CRF model to complete the entity recognition task, the CasRel model to complete the relationship extraction task, and build a domain knowledge graph; (2) Use the event model in the extensible primitive theory to represent the behaviors and functional objects in the product function expression model; (3) Based on this behavior, use the modified cosine semantic similarity to match similar relation words in the knowledge graph, obtain all tail entities of the relation words, and form a matrix; (4) Extract the functional objects in the event model and compare them with all elements in the candidate matrix. Use the modified cosine semantic similarity to match them. Collect the elements that exceed the similarity threshold. Infer all head entities through the elements and the corresponding relational words to generate the structural solution set corresponding to the functional object. Circulate all behaviors and functional objects in the product functional expression model to infer the overall structural solution set; (5) The knowledge graph and the set of demand structure solutions corresponding to the product functions are used as input data and passed into the node importance evaluation method; (6) Using the node importance evaluation calculation method based on network analysis, all elements in the structure solution set are sorted according to the knowledge graph and node importance; (7) Starting from the first functional object in the overall functional set, select the first ranked element in the corresponding structural solution set, that is, the element with the highest node importance, and judge whether it can serve as the final demand structure information of the first functional object from multiple perspectives such as economy and convenience. If it can serve as the final demand structure information, put it into the final result set. If it is not feasible, continue to take out the second ranked element in the structural solution set and judge it again, and so on; (8) Obtain the final demand structure information of all functional objects in the overall functional set and comprehensively form a complex equipment development plan.

2. According to the method for generating complex equipment development requirement structure information based on knowledge graph in claim 1, the marking and processing process in step 1.1) is specifically as follows: Obtain relevant patents in 2020 and later in the field of complex equipment to be studied, analyze the literature content and the field of complex equipment to be studied, and combine the suggestions of designers in this field to formulate the entity types to be identified in the knowledge graph; Use the Wizard Annotation Assistant and the BMES data annotation method to annotate the relevant text; BMES is a common data annotation method, in which "B" represents the annotation mark at the beginning of the entity, "M" represents the annotation mark in the middle of the entity, and "E" represents the annotation mark at the end of the entity; "S" represents a non-entity mark, and "B", "M" and "E" must be followed by the entity's abbreviation code; finally, the annotation results are saved as a txt file.

3. According to the method for generating complex equipment development demand structure information based on knowledge graph in claim 1, the step 1.2) defines the relationship type to be extracted, and uses the wizard annotation assistant to annotate the relationship specifically as follows: First, the relationship type between the head entity and the tail entity in the text sentence is defined based on the text content and technical experts in the relevant complex equipment field; secondly, the annotation format of {{"h":{"name":"Here is the name of the head entity"},"t":{"name":"Here is the name of the tail entity"},"relation":"Here is the relationship type","text":"Here is the original text sentence content"} is adopted; finally, the wizard annotation assistant is used to perform relationship annotation.

4. According to the method for generating complex equipment development requirement structure information based on knowledge graph in claim 1, the word vector generated by using BERT model to fully integrate training text information is specifically: a) means that a sequence of knowledge text to be extracted is segmented into characters, and each independent character is used as input and passed into BERT; b) After the text characters to be extracted are encoded by BERT, they can be converted into a 768-dimensional feature vector, i.e., h i ∈R 1×768 ; After linear transformation, the output of BERT can be mapped into a low-dimensional semantic feature vector.

5. According to the method for generating complex equipment development requirement structure information based on knowledge graph in claim 1, the specific process of using the event model in the extension primitive theory to characterize the behavior and function objects in the product function expression model in step (2) is as follows: Use Γ(D) to represent the name of the complex equipment or the subsystem of the complex equipment, verb i represents the ith "behavior" of the complex equipment or the subsystem of the complex equipment, and noun i represents the ith functional object of the complex equipment or the subsystem of the complex equipment, then 6. According to the method for generating complex equipment development requirement structure information based on knowledge graph in claim 1, in step (3), the modified cosine semantic similarity is used to match the similar relation words in the knowledge graph, and all the tail entities of the relation words are obtained to form a matrix. The specific process is as follows: Define a fixed set of verbs i and its corresponding noun i is a set of product functional units to be solved F i , denoted as F i =[verb i ,noun i ], then the overall function set is M = {F1, F2, ..., F n }; In the domain knowledge graph, search and verb i The relationship names with high modified cosine semantic similarity are summarized, and the tail entities containing these relationships are summarized to form the matrix B of the "behavior" and "function" objects. i , then B i =[b1,b2,...,b n ];in, b i Indicates the i-th verb i Similar relationship names and corresponding matrices Indicates all tail entities corresponding to the relation word; Then for the i-th group of product function units to be solved F i , whose matrix B i for: In step (3), the calculation process of modified cosine semantic similarity is: Among them, Sim(D1,D2) represents the adjusted cosine semantic similarity of two entities D1 and D2; Represents the semantic vector of entity D1, The semantic vector representing entity D2; represents the mean semantic vector of entity D1, Represents the mean semantic vector of entity D2.

7. According to the method for generating complex equipment development requirement structure information based on knowledge graph in claim 1, the specific process of generating the structure solution set corresponding to the functional object in step (4) is as follows: Define set B i Elements with a corrected cosine semantic similarity less than 0.8 After all are removed, search for the corresponding triple head entity in the corrected candidate matrix Composition of product functional unit F i The candidate solution matrix C i ; then the matrix is If C i If it is an empty set, the user can define the default implementation function unit F i structure; Repeat the above steps until all functions in Γ(D) can find the candidate solution matrix C i , and then the overall structural solution set is obtained.

8. According to the method for generating complex equipment development demand structure information based on knowledge graph in claim 1, the specific process of using the node importance evaluation calculation method based on network analysis in step (6) to sort all elements in the structure solution set according to the knowledge graph and node importance is as follows: 6.1) Take a certain structural solution in the demand structural solution set corresponding to the product function as node v i , use Dijstra algorithm to calculate the shortest path dis(v between the current node and the remaining nodes i ,v j ); 6.2) Calculate the current node v i The closeness center C(i); Define dis(v i ,v j ) is the node v i To node v j The shortest path length of node v is shorter, the closer it is to the center. i The closeness centrality can be expressed as C(i); 6.3) Based on the constructed domain knowledge graph, solve the current node v i Neighborhood δ ki With node key domain O i ; Define node v i The neighborhood of is the set of all nodes connected to it by edges, and the node neighborhood is δ ki ; Node v s Neighborhood δ kss Indicates that v s The set of all nodes associated with edges; for node v t Neighborhood Indicates that v t The set of all nodes with edges connected to them; then node v i The node key domain is expressed as 6.4) Traverse the neighborhood δ ki Any two nodes v p and v q , solve P(v p ,v q ); P(v p ,v q ) indicates that at node v i Neighborhood δ ki Any two nodes v p and v q The set of shortest paths between ; 6.5) Calculate the number of nodes passing through v i The number of shortest paths M(i) and the number of paths that do not pass through node v i The number of shortest paths N(i); Define a node v with degree k i Neighborhood δ ki In the example, any pair of nodes passes through node v i The number of shortest paths is M(i), which does not pass through node v i The number of shortest paths is N(i); assuming that node v s and v t The shortest path between sj Combined with the node key domain, M(i) and N(i) can be expressed as in in 6.6) Calculate the node v i The criticality K(i); 6.7) In the process of network analysis, due to the different network structures, the number and position of nodes and edges will affect the importance of each node in the entire network. In complex networks, the importance of central nodes and non-central nodes is different. Whether nodes are connected will also affect their importance. The larger the number of shortest paths of a node, the more important the node is. Therefore, we can define and calculate the node v i The importance of D(i); Finally, all elements in the set of structural solutions are sorted according to the knowledge graph and node importance.

9. A computer storage medium, characterized in that The storage medium contains a program which, when running, controls a device to execute a method for generating complex equipment development requirement structure information based on a knowledge graph as described in any one of claims 1 to 8.

10. A processor, characterized in that: The processor is used to execute a program, wherein the program, when running, executes the method for generating complex equipment development requirement structure information based on a knowledge graph as described in any one of claims 1 to 8.

Images