A method for implementing an air-space field knowledge graph driven air-space equipment production intelligent question answering machine

Abstract

The method for realizing the air separation equipment production technology intelligent question-answering machine driven by the air separation field knowledge graph mainly includes three stages. Firstly, the HanLP-kf language processing package and the Neo4j database are used to realize the standardized representation and storage of the air separation equipment production technology field knowledge. Secondly, the production technology field knowledge graph is constructed by taking the production technology field literature as the data source. Thirdly, the Multi-Raspberry Pi network acquisition system is used to acquire real-time production data in the distributed air separation equipment production field, so as to construct the dynamic real-time knowledge graph of the air separation production technology field. Finally, the Naive Bayes classification method is used to enable the air separation equipment production technology intelligent question-answering machine to analyze and feed back various professional problems input by the user interface. The method can effectively overcome the shortcomings of the traditional knowledge base and expert system in knowledge updating and iteration, and through the self-learning ability of the question-answering machine, the knowledge graph is managed and more complex and dynamically updated knowledge is applied, so as to provide the technical management personnel participating in the air separation equipment production with the field knowledge advancing with the times.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical management field of air separation equipment production, and particularly relates to an air separation field knowledge graph driven air separation equipment production technology intelligent question answering machine implementation method. TECHNICAL BACKGROUND

[0002] As a national key equipment, air separation equipment has large equipment investment, high value of parts and components, and specific model parameters that need to be customized according to customer needs. Therefore, the air separation equipment production process often presents the characteristics of many models, and the production and manufacturing process of its parts and components is often realized in a decentralized and distributed form, and the production technology involved also has the characteristics of decentralization and complexity.

[0003] Therefore, in view of the problems of professional technical knowledge collection, management and application efficiency exposed in the decentralized production process of air separation equipment, a technical field knowledge intelligent question answering machine is needed to help production technology managers and front-line operators, so as to extract professional air separation equipment production technology field knowledge from massive and professional air separation equipment production process data and form an intelligent solution to air separation equipment production technology field problems based on a user interface. SUMMARY

[0004] In order to solve the problem of efficient management and application of knowledge in the production process of air separation equipment, the purpose of the present application is to provide an air separation field knowledge graph driven air separation equipment production technology intelligent question answering machine implementation method. First, a customized HanLP-kf language processing package and a Neo4j database are used to realize the standardized representation and storage of air separation equipment production technology field knowledge, and real-time production data is acquired in a distributed air separation equipment production site through a Multi-Raspberry Pi acquisition network, so as to build a dynamic real-time knowledge graph of the air separation equipment production technology field. Finally, a Naive Bayes classification method is used to enable the air separation equipment production technology intelligent question answering machine to analyze and feedback various professional problems input by the user interface.

[0005] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0006] 1. An air separation field knowledge graph driven air separation equipment production technology intelligent question answering machine implementation method, characterized in that it comprises the following steps:

[0007] S1: Based on the HanLP processing package, a customized HanLP-kf language processing package for processing air separation production field technology is formed, and based on the Neo4j database, the standardized representation and storage of the technical field knowledge of the air separation equipment production process are realized;

[0008] S2: With the real-time production data collected by Multi-Raspberry Pi, a dynamic real-time knowledge graph of air separation production technology field is constructed.

[0009] S3: According to the stored knowledge of air separation production technology field in S1 and the dynamic knowledge graph of air separation production technology field in S2, a naive Bayes classification method is used to respond to various professional problems, and a user interface of intelligent question and answer machine is developed, so as to realize intelligent analysis and feedback of air separation production technology field problems.

[0010] 2. The air separation field knowledge graph driven air separation equipment production technology intelligent question and answer machine implementation method according to claim 1, wherein the knowledge normalization in S1 mainly includes the representation of knowledge in the fields of process technology, raw material, supplier, application field, characteristic, product, property, problem, and customer. The knowledge normalization storage means storing in the structure of "subject-predicate-object", representing the knowledge based on RDF, and establishing corresponding knowledge bases for materials, application plates, fault solving methods, fault categories, and process technologies through the Neo4 database. Taking the material characteristics as an example, "material" represents the material, "R" represents the relationship, and "characteristic" represents the material characteristics, which can be formalized as:

[0011] RDF::=={mat.,R,cha}

[0012] The knowledge bulk construction based on the Neo4j database can be divided into four steps:

[0013] (1) First, classify according to nodes, set the table header of each column of nodes, store the data in the form of CSV file, and copy and paste it into the import file under the Neo4j installation directory.

[0014] (2) Then, classify according to relationships, set the table header of each column of relationships, store the data in the form of CSV file, and copy and paste it into the import file under the Neo4j installation directory.

[0015] (3) Next, start the Neo4j server, enter the import command in the instruction box, and the import order is: first import the node CSV file to create entity nodes, and then import the relationship CSV file to create the relationship between entity nodes.

[0016] The Cypher statement for importing the "Problem" and "Category" node relationship in the Neo4j instruction box is composed of the starting node, the ending node, and the relationship.

[0017] The Cypher query statements corresponding to some templates are as follows:

[0018] • Cypher query template for related questions

[0019] match(n:Problem)-[r:is]-(m:Category)where n.problemName = {get_problemName}return m.categoryName

[0020] • Cypher query template for problem-solving

[0021] The `match(n:Problem)` method, where `n.problemName` equals `{get_problemName}`, returns `n.solution`.

[0022] Cypher query template for variable (greater than) type problems

[0023] match(n:application_field)-[:apply]-(m:material)where n.application_field = {get_applicationFieldName}and m.variable >= {get_temperature}return m.material_name

[0024] Cypher query template for variable (less than) type problems

[0025] match(n:application_field)-[:apply]-(m:material)where n.application_field = {get_applicationFieldName}and m.variable < {get_temperature}return m.material_name

[0026] (4) Once entity nodes and node relationships have been created and connected, if there are any additional entities to be added later, new entity nodes can be created and corresponding relationships can be built; if there are any entities to be deleted later, the node relationships must be deleted first, and then the entity nodes can be deleted.

[0027] 3. The method for implementing an intelligent question-and-answer machine for air separation equipment production technology driven by a knowledge graph in the air separation field according to claim 1, characterized in that the Multi-Raspberry Pi acquisition network construction in step S2 includes the following features:

[0028] (1) When there is one configured Raspberry Pi, its Cont component can ensure its communication and interaction with other Raspberry Pis;

[0029] (2) Raspberry Pi will respond to a certain point-to-point-based request through its information I / O component;

[0030] (3) The Len component will provide the Cont component with a specific solution to the response, which needs to consider the unique request involved in step (2) and the real-time state of the Raspberry Pi itself;

[0031] (4) When it is determined that the space division equipment production data needs to be obtained, the Cont component will activate the Ssor component with the help of the Len component;

[0032] (5) The Ssor component will obtain the monitoring data of its configured sensor entity through its physical I / O;

[0033] (6) At the time of action (5), request the Ds component to obtain historical data / knowledge and provide it as input data for the Len component to process algorithms or models;

[0034] (7) The Len component also requests the Ds component and returns how to perform cooperation to complete the task;

[0035] (8) The data / knowledge stored in step S1 and the sensor data together support the question and answer machine in S3 to have better execution effect;

[0036] (9) The learning results of the Len component will also be saved as part of the space division field knowledge in the field knowledge graph of step S1.

[0037] 4. The space division field knowledge graph driven space division equipment production technology intelligent question and answer machine implementation method according to claim 1, characterized in that,

[0038] The response to new questions in step S3 using the Naive Bayes classification method includes the following processes:

[0039] (1) After the user's input question in the question and answer machine is processed by HanLP-kf in S1, it enters the question matching module. The theoretical basis of the question matching module is the word vector based on One-hot encoding, and the key technology is the Naive Bayes classifier.

[0040] (2) Using Naive Bayes classifier to train the sample of the question. First, JavaRDD needs to be created, and then JavaRDD is converted into SaprkRDD, so as to improve the calculation speed of the Naive Bayes classifier. Then, the Naive Bayes classifier is used to train the sample set of the user question, when the user question word vector is input, the Naive Bayes classifier will calculate the probability according to the classification template, and the user question word vector is distributed into the corresponding template.

[0041] (3) Construction of the question query module. The core of the question query module is the query template based on the Cypher statement, and the Cypher is the query language of the Neo4j graph database. In the question query module, the code under each template contains the corresponding Cypher query statement. The principle of the question query module is that, after the processing of the question matching module in (2), the question has been classified, the special word symbols are replaced into the special words, and the variable replacement is realized. Then, according to the Cypher statement set under each template, the search is carried out in the graph database through the query interface of Neo4j, and the answer is obtained.

[0042] The beneficial effects of the present application are:

[0043] 1) Compared with the traditional question and answer machine, the method provided by the present application has the advantages that: the present application dynamically constructs the intelligent question and answer machine of air separation equipment production technology by means of the knowledge graph tool, dynamically updates the data generated in the production process as a part of knowledge, realizes the effective management and control of the related knowledge in the field of air separation equipment production, provides convenience for the production personnel in the related field to query the air separation professional production knowledge, and the results show that the intelligent degree of the air separation knowledge base is significantly improved compared with the prior art.

[0044] 2) The knowledge graph of the field of air separation equipment production constructed by the present application supplements the real-time production data collected by the Multi-Raspberry Pi collection network as the dynamic characteristics of the field knowledge graph, effectively overcomes the shortcomings that the traditional knowledge base and expert system are difficult to update and iterate, and through the self-learning ability of the question and answer machine, more complex and dynamically updated knowledge can be stored in the knowledge graph, so as to provide the field knowledge advancing with the times for the production and technical management personnel. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 It is the execution logic diagram of the present application.

[0046] Figure 2 It is the Cypher statement for importing the "Problem" node

[0047] Figure 3 It is the Cypher statement for importing the relationship between the "Problem" and "Category" nodes

[0048] Figure 4 Principle of the field question matching module

[0049] Figure 5 Professional question processing flow DETAILED DESCRIPTION

[0050] The present application will be described in detail below with reference to the accompanying drawings and examples, which are part of this application and serve to further explain the present application but do not constitute a limitation thereof.

[0051] Referring to Figure 1 The application of the hollow space field knowledge graph driven hollow space equipment production technology intelligent question answering machine implementation method should include the following steps:

[0052] S1: Based on the HanLP processing package, form a customized HanLP-kf language processing package for processing the terminology of the hollow space production field technology, and based on the Neo4j database, realize the standardized representation and storage of the technical field knowledge of the hollow space equipment production process;

[0053] S2: With the real-time production data collected by the Multi-Raspberry Pi collection network as a supplement, a hollow space production technology field knowledge graph with dynamic real-time characteristics is constructed;

[0054] S3: According to the stored hollow space production technology field knowledge in S1 and the dynamic hollow space production technology field knowledge graph in S2, various professional problems are responded to using the Naive Bayes classification method, and the user interaction interface of the intelligent question answering machine is developed, finally realizing the intelligent analysis and feedback of the hollow space production technology field problems.

[0055] The specific implementation process is as follows:

[0056] The standardized representation of knowledge in step S1 mainly includes the representation of process technology, raw materials, suppliers, application fields, characteristics, products, properties, problems, customers, etc. The standardized storage of knowledge refers to the storage in the structure of "subject-predicate-object", based on the RDF method to represent the knowledge, and through the Neo4 database, the corresponding knowledge base is established for materials, application plates, fault solving methods, fault categories and process technology. Taking the material characteristics as an example, "material" represents the material, "R" represents the relationship, and "characteristic" represents the material characteristics.

[0057] The knowledge bulk construction based on the Neo4j database in step S1 is as follows:

[0058] (1) Classify by node, and set the table header of each column node, store the data in the form of CSV file, and copy and paste it into the import file under the Neo4j installation directory.

[0059] (2) Classify by relationship, and set the table header of each column relationship, store the data in the form of CSV file, and copy and paste it into the import file under the Neo4j installation directory.

[0060] (3) Refer to Figure 2 , input "neo4j.bat console" in the command prompt to start the Neo4j server, and input the import command in the instruction box. The import order is: first import the node CSV file to create entity nodes, and then import the relationship CSV file to create the relationship between entity nodes.

[0061] Refer to Figure 3 , import the Cypher statement of the relationship between "Problem" and "Category" nodes in the Neo4j instruction box. The two ends of the Cypher statement are composed of the starting node, the ending node and the relationship.

[0062] The Multi-Raspberry Pi acquisition network construction in step S2 includes the following processes:

[0063] (1) When there is a configured Raspberry Pi, its Cont component can ensure its communication and interaction with other Raspberry Pis;

[0064] (2) Raspberry Pi will respond to a certain point-to-point based request through its information I / O component;

[0065] (3) The Len component will provide a specific solution for the Cont component to respond, which needs to consider the unique request involved in step (2) and the real-time state of the Raspberry Pi itself;

[0066] (4) When it is determined that the production data of the air separation equipment needs to be obtained, the Cont component will activate the Ssor component with the help of the Len component;

[0067] (5) The Ssor component will obtain the monitoring data of its configured sensor entity through its physical I / O;

[0068] (6) In action (5), request the Ds component to obtain historical data / knowledge, and use it as input data for the Len component to process algorithms or models;

[0069] (7)Len component also requests to Ds component, and returns how to perform cooperation to complete the task;

[0070] (8) The data / knowledge stored in step S1 together with the sensing data support the question-answering machine in S3 to have better execution effect;

[0071] (9) The learning result of Len component is also saved in the domain knowledge graph in step S1 as part of the domain knowledge.

[0072] The response to the new question by using the Naive Bayes classification method in step S3 includes the following processes:

[0073] (1) Reference Figure 4 , in the question-answering machine, after the user input question is processed by HanLP-kf in S1, it enters the question matching module. The theoretical basis of the question matching module is the word vector based on One-hot encoding, and the key technology is the Naive Bayes classifier.

[0074] (2) Training the question sample by using the Naive Bayes classifier. First, JavaRDD needs to be created, and then JavaRDD is converted into SaprkRDD, so as to improve the calculation speed of the Naive Bayes classifier. Then, the Naive Bayes classifier is used to train the user question sample set. When the user question word vector is input, the Naive Bayes classifier will perform probability calculation according to the classification template, and assign the user question word vector to the corresponding template.

[0075] (3) Reference Figure 5 , construction of the question query module. The core of the question query module is the query template based on Cypher statement. Cypher is the query language of Neo4j graph database. In the question query module, the code under each template contains the corresponding Cypher query statement. The principle of the question query module is that after the processing of the question matching module in the previous step, the question has been classified, and the special word symbol is replaced by the special word, so as to realize the replacement of the variable. Then, according to the Cypher statement set in each template, the query interface of Neo4j is used to search in the graph database to obtain the answer.

Claims

1. A method for implementing an intelligent question-and-answer machine for air separation equipment production technology driven by a knowledge graph in the air separation field, characterized in that: Includes the following steps: S1: Based on the HanLP processing package, a customized HanLP-kf language processing package is formed to process the terminology of air separation production technology. Based on the Neo4j database, the technical knowledge of the air separation equipment production process is standardized and stored. S2: Using real-time production data collected by the Multi-Raspberry Pi acquisition network as a supplement, construct a knowledge graph in the field of air separation production technology with dynamic real-time characteristics; S3: Based on the knowledge of air separation production technology stored in S1 and the dynamic knowledge graph of air separation production technology in S2, the Naive Bayes classification method is used to respond to various professional questions, and the user interface of the intelligent question-and-answer machine is developed to ultimately realize intelligent analysis and feedback of questions in the field of air separation production technology. The knowledge standardization representation and Neo4j-based mass knowledge construction in step S1 include: (1) classifying knowledge based on process technology, raw materials, suppliers, application fields, characteristics, products, properties, problems, and customers, representing each type of knowledge in the form of a "subject-verb-object" triple structure using RDF, and establishing knowledge bases for materials, application modules, troubleshooting methods, fault categories, and process technology in the Neo4j graph database respectively; (2) classifying by node, setting a header for each type of node, storing the node data in CSV file format, and storing it in the import directory under the Neo4j installation directory; (3) classifying by relation, setting a header for each type of relation, and storing the relation data in the import directory under the Neo4j installation directory. The system data is stored in the form of CSV files and stored in the import directory under the Neo4j installation directory; (4) Start the Neo4j server, enter the import command in the command box, import in the order of first importing the node CSV file to create entity nodes, and then importing the relationship CSV file to create the relationship between entity nodes, and import the Cypher statement of the relationship between the Problem node and the Category node in the Neo4j command box, so that the two ends of the Cypher statement correspond to the start node and the end node respectively; (5) Set the corresponding Cypher query template for the category problem and the solution problem, which is used to generate the query statement according to the question classification result in step S3.

2. The method for implementing an intelligent question-and-answer machine for air separation equipment production technology driven by a knowledge graph in the air separation field according to claim 1, characterized in that, The construction of the Multi-Raspberry Pi acquisition network in step S2 includes the following features: (1) When there is a configured Raspberry Pi, the Cont component of the Raspberry Pi is used to communicate and interact with other Raspberry Pis; (2) The Raspberry Pi responds to point-to-point requests through its information I / O component; (3) The Len component provides a response scheme for the Cont component, the response scheme taking into account the point-to-point request in step (2) and the Raspberry Pi's own information I / O component. (4) When it is determined that air separation equipment production data needs to be obtained, the Cont component activates the Ssor component under the control of the Len component; (5) The Ssor component obtains the monitoring data of the configured sensors through its physical I / O; (6) During the execution of step (5), the Ds component receives the request and provides historical data or knowledge, and uses the historical data or knowledge as input data for the Len component to perform algorithm processing or model processing; (7) The Len component also requests the Ds component and returns how to perform cooperation to complete the task; (8) The data / knowledge stored in step S1 and the sensor data together support the execution of the question-and-answer machine in S3; (9) The learning results of the Len component are stored as part of the knowledge in the air separation production technology field described in step S1 in the knowledge graph.

3. The method for implementing an intelligent question-and-answer machine for air separation equipment production technology driven by a knowledge graph in the air separation field according to claim 1, characterized in that, The process of responding to new questions using the Naive Bayes classification method described in step S3 includes: (1) In the question-answering machine, the user-inputted question is processed by the HanLP-kf language processing package described in step S1 and then enters the question matching module. The theoretical basis of the question matching module is word vectors based on One-hot encoding, and the key technology is the Naive Bayes classifier; (2) The Naive Bayes classifier is used to train the question sample: First, a JavaRDD needs to be created, and then the JavaRDD is converted into a SaprkRDD to improve the calculation speed of the Naive Bayes classifier. Then, the Naive Bayes classifier is used to train the user question sample set. When the user question word vector is input, the Naive Bayes classifier... The classifier will perform probability calculations based on the classification template and assign the word vectors of the user's question to the corresponding template; (3) Construct a question query module. The core of the question query module is a query template based on Cypher statements. Cypher is the query language of Neo4j graph database. In the question query module, the code under each template contains the corresponding Cypher query statement. The principle of the question query module is that after the processing of the question matching module in (2), the question has been classified and the proper part-of-speech symbols have been replaced with proper vocabulary to realize the replacement of variables. Then, according to the Cypher statement set under each template, the query interface of Neo4j is used to enter the graph database to search and obtain the answer.

Images