WO2018138205A1

WO2018138205A1 - Model search method and device based on semantic model framework

Info

Publication number: WO2018138205A1
Application number: PCT/EP2018/051839
Authority: WO
Inventors: Qi Wang; Yong Yuan; Ming Kai Dong; Rui Guo ZHANG; Ming Yu; Jing Cao; Zhen Zhang; Ming Zhang
Original assignee: Siemens Aktiengesellschaft
Priority date: 2017-01-25
Filing date: 2018-01-25
Publication date: 2018-08-02
Also published as: CN108345622A

Abstract

The present invention provides a model search method based on a semantic model framework, including: a buffering step of buffering and analyzing model query information of a user and buffering relative knowledge; and a query step of querying a model in a buffer, an index, and a data library, comparing a model queried by the user and the model queried in the buffer, the index, and the data library, ranking relative models, returning a ranking result as a search result, and sending the search result to the user. According to the model search method and device based on a semantic model framework provided in the present invention, search for a relative model may be performed at a high response speed, and particularly this is quite practical in a recommendation process in modeling. According to the present invention, query, analysis, and the search result can be provided to buffer the relative knowledge, so as to ensure rapid search in modeling. According to the present invention, self-extension can be performed under the semantic model framework, that is, new knowledge having no classification information is incorporated into the semantic model framework.

Description

MODEL SEARCH METHOD AND DEVICE BASED ON SEMANTIC

MODEL FRAMEWORK

BACKGROUND OF THE INVENTION

Field of the Invention

The present invention relates to the field of industrial automation technologies, and in particular, to a model search method and device based on a semantic model framework.

Related Art

Semantic models are widely used to describe industrial automation systems. For example, some semantic models are used for system simulation, and some semantic models are used to describe data and relationships. However, an industrial system has an extremely large quantity of devices and complex control logic, and has some problems in aspects of construction and search of a semantic model.

First, most modeling processes require a user to understand an automation system quite deeply, and further require the user to have a capability of "translating" a true system into a model. A search system can be used by a modeling system to provide domain knowledge and recommendation, which may make a modeling process become more convenient.

Moreover, an ordinary semantic model search engine stores and searches for a model in a general manner, for example, a manner of jena of RDF. The ordinary semantic model search engine has no pertinence, for example, does not perform different processing in a domain ontology aspect, for example, does not perform classification, and does not consider query or answer either. A model and data are equally treated without pertinence. Consequently, when a model is quite large, performance of a search function is not good.

Moreover, the ordinary semantic model search engine can be used to search for only a matching model, but cannot be used to search for a relative model.

SUMMARY OF THE INVENTION

A first aspect of the present invention provides a model search method based on a semantic model framework, including the following steps: a buffering step of buffering and analyzing model query information of a user and buffering relative knowledge; and a query step of querying a model in a buffer, an index, and a data library, comparing a model queried by the user and the model queried in the buffer, the index, and the data library, ranking relative models, returning a ranking result as a search result, and sending the search result to the user. According to the model search method and device based on a semantic model framework provided in the present invention, search for a relative model may be performed at a high response speed, and particularly this is quite practical in a recommendation process in modeling. According to the present invention, query, analysis, and the search result can be provided to buffer the relative knowledge, so as to ensure rapid search in modeling. According to the present invention, self-extension can be performed under the semantic model framework, that is, new knowledge having no classification information is incorporated into the semantic model framework.

Further, after the query step, the method further includes the following step: a storage step of storing the query result.

Further, before the buffering step, the method further includes the following step: an analysis step of analyzing a semantic model framework and knowledge data, extracting knowledge, and generating knowledge reference data.

Further, the analysis step further includes the following steps: analyzing the semantic model framework by using classification information, and assigning the classification information to the knowledge data; extracting segment information from the semantic model framework and the knowledge data; and calculating probabilities of the segment information and the classification information.

Further, the buffering step further includes the following steps: buffering the model query information of the user; analyzing the model query information of the user, and comparing classification vocabularies and query vocabularies to classify the model query information of the user; and replicating data that has a high probability and is of a same type. A second aspect of the present invention provides a model search device based on a semantic model framework, including: a buffering device, configured to buffer and analyze model query information of a user and buffer relative knowledge; and a query device, configured to query a model in a buffer, an index, and a data library, compare a model queried by the user and the model queried in the buffer, the index, and the data library, rank relative models, return a ranking result as a search result, and send the search result to the user. According to the model search method and device based on a semantic model framework provided in the present invention, search for a relative model may be performed at a high response speed, and particularly this is quite practical in a recommendation process in modeling. According to the present invention, query, analysis, and the search result can be provided to buffer the relative knowledge, so as to ensure rapid search in modeling. According to the present invention, self-extension can be performed under the semantic model framework, that is, new knowledge having no classification information is incorporated into the semantic model framework.

Further, the model search device further includes: a storage device, configured to store the query result.

Further, the model search device further includes: an analysis device, configured to analyze a semantic model framework and knowledge data, extract knowledge, and generate knowledge reference data.

Further, the analysis device further includes: a classification module, configured to analyze the semantic model framework by using classification information, and assign the classification information to the knowledge data; an extraction module, configured to extract segment information from the semantic model framework and the knowledge data; and a calculation module, configured to calculate probabilities of the segment information and the classification information.

Further, the buffering device further includes: a buffering device, configured to buffer the model query information of the user; an analysis device, configured to analyze the model query information of the user, and compare classification vocabularies and query vocabularies to classify the model query information of the user; and a replication device, configured to replicate data that has a high probability and is of a same type.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a framework diagram of a model search system based on a semantic model according to a specific embodiment of the present invention;

FIG. 2 exemplarily shows a semantic model framework SMF;

FIG. 3 exemplarily shows an assembly line driving template as knowledge data KD;

FIG. 4 is a flowchart of a buffering step SI of a model search method based on a semantic model according to a specific embodiment of the present invention;

FIG. 5 is a flowchart of a query step S2 of a model search method based on a semantic model according to a specific embodiment of the present invention; and

FIG. 6 is a flowchart of an analysis step SO of a model search method based on a semantic model according to a specific embodiment of the present invention.

Description of serial numbers in the accompanying drawings:

100 User query module

200 Search engine

Query interface module

220 Buffering module

230 Ranking module

240 Buffering space

Query buffering module

242 Knowledge buffering module

250 Index space

251 Classification vocabularies

252 Classification index 253 Query buffering module 300 Analysis module

400 Data space

410 Model data

420 Segment data

SMF Semantic model framework

KD Knowledge data

MQ Model query information

254 Saving queue module

SR Search result

310 Classification module

312 Classification data

314 Model of a particular type

320 Extraction module

321 Vocabularies in knowledge data KD

322 Segment information having classification information

330 Calculation module

331 Segment information having classification information and frequency thereof

DETAILED DESCRIPTION OF THE INVENTION

Specific implementations of the present invention are described below with reference to accompanying drawings.

The present invention provides a model search mechanism based on a semantic model framework, and particularly search based on a relative model. A semantic model framework and knowledge data can be analyzed and stored in a search engine, and the search engine may be used to search for a relative model. In a modeling system, by means of the present invention, a relatively fast response may be achieved in search for a relative model, and particularly the present invention is quite effective for a recommendation step in a modeling process.

A semantic model framework (SMF) is a structured knowledge resource. The knowledge includes a semantic model standard ISA-95, a Semantic Sensor Ontology (SSN), and the like that are considered as different classification information. The semantic model framework SMF includes two main parts: a core and a knowledge package. Semantic knowledge in the core includes semantic ontology standards that describe general common sense about an industrial automation system. Knowledge in the knowledge package includes a model having particular classification information. FIG. 2 exemplarily shows a semantic model framework SMF. As shown in the figure, "control system", "process plant", "vehicle", and "assembly line" are classes, and relative semantic standards (for example, "ISA-95" and "ISO- 15926") and templates (for example, "engine template" and "driving template") are assigned to these classes. Moreover, all knowledge data KD (for example, "templates", "libraries", and "sample models") is assigned to one or more classes, and added to a framework. A semantic model framework having classification information is used to identify query classification and a search target.

Model templates, libraries, and sample models generated from a past project are all considered as knowledge data KD. The templates and the libraries are semantic models derived from experiences or standards, for example, Modelica libraries. FIG. 3 exemplarily shows an assembly line driving template as knowledge data KD, and the assembly line driving template is a sample model generated from a past project, for example, a sample model that is set specially from an assembly line of a Ford automobile. As shown in FIG. 3, "motor", "gearbox", "roller", "vibration sensor", and "displacement sensor" are building blocks, and lines between the building blocks indicate connections between each other. For example, a connection relationship between "motor" and "vibration sensor" and a connection relationship between "roller" and "displacement sensor" are connections, and a connection relationship between "gearbox" and each of "motor" and "roller" is driving. Knowledge data KD is described by using a uniform format, for example, an RDF or a Modelica language.

FIG. 1 is a framework diagram of a model search system based on a semantic model according to a specific embodiment of the present invention, and the model search system includes a user query module 100, a search engine 200, an analysis module 300, and a data space 400. The user query module 100 is configured to send query information to the search engine, and accept a query result. Typically, the query information includes a model matching a query target, and optionally includes classification information of some target models. After completing some column operations such as buffering, indexing, and storing, the search engine 200 returns a search result to the user query module 100. The analysis module 300 is configured to analyze a semantic model framework SMF and knowledge data KD, extract knowledge, and generate knowledge reference data to be used by the search engine 200 in a search process. The data space 400 stores segment information and model data, the foregoing two types of data both include corresponding classification information and probabilities, and data in the data space 400 may be searched for by using position information stored in an index. A query interface module 210 as a bridge between the user query module 100 and a core module of the search engine 200 receives the query information and sends the query information to a buffer module 220 and a ranking module 230; and receives a search result SR and sends the search result SR to the user query module 100.

Referring to FIG. 1 , a model search method based on a semantic model framework SMF provided in the present invention includes the following steps.

First, a buffering step S I of buffering and analyzing model query information MQ of a user and buffering relative knowledge is performed. Further, as shown in FIG. 1 and FIG. 4, the buffering step further includes the following step: buffering the model query information MQ of the user in a query buffering module 243; analyzing, in the query buffering module 241, the model query information MQ of the user, and comparing classification vocabularies and query vocabularies to classify the model query information MQ of the user; and replicating data that has a high probability and is of a same type from a hot knowledge index 253 to a knowledge buffering module 242. For example, when a piece of model query information MQ "generator driving gearbox" of the user is received, classification information is locked to "assembly line". As shown in the following table, segment information "assembly line" and a model that are ranked first in the list and have high probabilities are buffered in the knowledge buffering module 242.

Table 1: Segment information and a model having high probabilities

As shown in the foregoing table, "from" and "to" indicate directions of building blocks and connections in a segment, and the directions exemplarily show a connection relationship between building blocks that is represented by an arrow direction shown in FIG. 3, where "connection" indicates a connection relationship, "classification" indicates performing classification, and "count" indicates a frequency at which the segment occurs. "FromProb", "ToProb", "ConnProb", and "ClassProb" are calculation of conditional probabilities of all elements.

A buffering space 240 shown in FIG. 1 includes the query buffering module 241 and the knowledge buffering module 242, where query and buffering in a particular time range are considered as a queue in the query buffering module 241 , and are analyzed by the buffering module 220, and a query class is determined by comparing query vocabularies and classification vocabularies 251. For the knowledge buffering module 242, a model and segment data 420 that have high probabilities are queried in a same class, and are replicated from the hot knowledge index 253 to the knowledge buffering module 242.

An index space 250 shown in FIG. 1 includes classification vocabularies 251 , a classification index 252, and the hot knowledge index 253 that are stored in the analysis module 300 and read by the ranking module 230 and the buffering module 220. The classification vocabularies 251 are configured to determine a class of a query or model. The classification index 252 is a storage position of the segment data 420 and model data 410 in classification, and is configured to allocating classification in particular data. The hot knowledge index 253 is replication of segment information and a model that have high probabilities, is organized by means of classification, and is configured to reduce a search range.

The data space 400 shown in FIG. 1 is configured to store the segment data 420 and the model data 410, the two types of data both include corresponding classes and probabilities, and the foregoing data may be searched for by using position information in an index.

The segment data 420 includes building blocks, connections between the building blocks, and a probability that a building block is connected to another building block. The segment data 420 further includes building blocks in different classes and connection probabilities. All including the semantic model framework SMF and the knowledge data KD is considered in extraction of the segment data, and some mathematical statistic algorithms are used to calculate the semantic model framework SMF and the knowledge data KD and perform expression in a simple form. For example, as shown in FIG. 3, in a piece of "assembly line driving template", all of "motor", "gearbox", "roller", "vibration sensor", and "displacement sensor" are building blocks, and a connection relationship between any two building blocks is represented by an arrow shown in the figure and "connection" or "driving" on the arrow. A model of a minimum unit includes two building blocks and a connection relationship between the two building blocks that are extracted from an input template shown in FIG. 3 and are considered as a piece of segment information.

Table 2: Segment information example

The foregoing table is an example of segment information. As shown in the foregoing table, "from" and "to" indicate directions of building blocks and connections in a segment, and the directions exemplarily show a connection relationship between building blocks that is represented by an arrow direction shown in FIG. 3, where "connection" indicates a connection relationship, "classification" indicates performing classification, and "count" indicates a frequency at which the segment occurs. "FromProb", "ToProb", "ConnProb", and "ClassProb" are calculation of conditional probabilities of all elements.

A conditional probability P (B|A) indicates a probability of B when A occurs, where A and B indicate different building blocks. Therefore, the following may be obtained:

FromProb=P (motor driving gearbox|motor)=0.85;

ToProb=P (motor driving gearbox |gearbox)=0.75;

ConnProb=P (motor driving gearbox |driving)=0.41 ; and

ClassProb=P (assembly line (motor driving gearbox)=1.0.

The model data 410 includes a model and a query rate, and each model has an independent storage space. An ordinary semantic model search engine may be used to store, for example, Jena. A position of a model is recorded in the classification index 252.

Then, a query step S2 of querying a model in a buffer, an index, and a data library, comparing a model queried by the user and the model in the buffer, the index, and the data library, ranking relative models, returning a ranking result as a search result, and sending the search result to the user is performed.

Specifically, as shown in FIG. 1 and FIG. 5, S21 is first performed, the model query information MQ is received, measured, and ranked in the knowledge buffering module 242, and a relative model list is generated. If a quantity of relative models in the relative model list is greater than a predetermined threshold (that is, a result is sufficiently good), step S24 is performed; otherwise step S22 continues to be performed. In step S22, measurement and ranking continue to be performed in the hot knowledge index 253, a relative model list is also generated, and if a quantity of relative models in the relative model list is greater than the predetermined threshold (that is, a result is sufficiently good), step S24 is performed; otherwise, step S23 continues to be performed. In step 23, measurement and ranking continue to be performed in the classification index 252 and the data space 400, and a relative model list is also generated. Finally, step S24 is performed, a ranking result is placed in both a saving queue module 254 and the knowledge buffering module 242, and the model and a query rate of segment information are stored again in the hot knowledge index 253 and the data space 400. The hot knowledge index 253 is arranged again by the ranking result. For example, if the hot knowledge index 253 does not have the model and the segment information, but the query rate of the segment information is greater than the predetermined threshold, the model and the segment information are replicated to the hot knowledge index 253. For example, as shown in FIG. 3, the query information Q includes a model that has some building blocks and connections, and the building blocks include "generator", "gearbox", and "roller". Specifically, "generator" is connected to and drives "gearbox", "gearbox" is connected to and drives "roller", and a query result includes relative segment information and models shown in the following table.

Table 3: Relative model and segment information

A rule and a process of measuring model query information MQ are specifically described below. For segment information measurement, if combing probabilities of query models and segment information are greater than a predetermined threshold, the segment information is considered as a search result SR. For model measurement, if query models are included in a piece of model data 410, and a quantitative proportion of building blocks in the query models is greater than that in the model data 410 and greater than a predetermined threshold, the model data 410 is considered as a search result SR. If query models are not included in a piece of model data 410, and a ratio of a quantity of types of building blocks to a quantity of query models in a piece of model data 410 is greater than a predetermined threshold, the model data 410 is considered as a search result SR.

Optionally, before the buffering step S 1 , the method further includes an analysis step SO of analyzing a semantic model framework SMF and knowledge data KD, extracting knowledge, and generating knowledge reference data. Specifically, the analysis step SO further includes the following steps: analyzing the semantic model framework SMF by using classification information, and assigning the classification information to the knowledge data KD; extracting segment information from the semantic model framework SMF and the knowledge data KD, and storing the segment information in a storage space; and calculating probabilities of the segment information and the classification information, and storing the probabilities in the storage space. The data generated by means of the analysis step SO is stored in the storage space, and the storage space returns the reference data to the analysis step SO.

Specifically, as shown in FIG. 6, when the semantic model framework SMF is embedded into the analysis module 300, a classification module 310 in the analysis module stores classification data 312 in the classification index 252, and a model 314 of a particular type is stored in the model data 410. When the knowledge data KD is embedded into the analysis module 300, the classification module 310 determines, by checking vocabularies 321 in the knowledge data KD having the classification vocabularies 251 , a class to which the knowledge data KD belongs. An extraction module 320 is configured to extract vocabularies from an input model, store the vocabularies in the classification vocabularies 251, and generate segment information 322 having classification information. A calculation module 330 is configured to calculate a frequency of segment information, and store segment information having classification information and a frequency thereof (331) in the segment data 420 and the hot knowledge index 253. For example, when input knowledge data is "assembly line driving template", output information is a knowledge index shown in FIG. 3, including segment information such as "motor driving gearbox" and a model such as "driving template".

Optionally, after the query step S2, the method further includes a storage step S3, that is, storing the query result.

A second aspect of the present invention further provides a model search device based on a semantic model framework, including: a buffering device, configured to buffer and analyze model query information of a user and buffer relative knowledge; and a query device, configured to query a model in a buffer, an index, and a data library, compare a model queried by the user and the model queried in the buffer, the index, and the data library, rank relative models, return a ranking result as a search result, and send the search result to the user.

Further, the model search device based on a semantic model framework further includes a storage device, configured to store the query result.

Further, the model search device based on a semantic model framework further includes an analysis device, configured to analyze a semantic model framework and knowledge data, extract knowledge, and generate knowledge reference data.

Further, the analysis device further includes:

a classification module 310, configured to analyze the semantic model framework by using classification information, and assign the classification information to the knowledge data;

an extraction module 320, configured to extract segment information from the semantic model framework and the knowledge data; and

a calculation module 330, configured to calculate probabilities of the segment information and the classification information.

Further, the buffering device further includes:

a buffering device, configured to buffer the model query information of the user; an analysis device, configured to analyze the model query information of the user, and compare classification vocabularies and query vocabularies to classify the model query information of the user; and

a replication device, configured to replicate data that has a high probability and is of a same type.

Functions of the foregoing devices and modules are described in detail in the model search method based on a semantic model framework described above, and for clarity, details are not described herein again.

According to the model search method and device based on a semantic model framework provided in the present invention, search for a relative model may be performed at a high response speed, and particularly this is quite practical in a recommendation process in modeling. According to the present invention, query, analysis, and the search result can be provided to buffer the relative knowledge, so as to ensure rapid search in modeling. According to the present invention, self-extension can be performed under the semantic model framework, that is, new knowledge having no classification information is incorporated into the semantic model framework.

Although content of the present invention is described in detail by using the foregoing preferred embodiments, it should be understood that the foregoing description should not be considered as a limitation on the present invention. After a person skilled in the art reads the foregoing content, various modifications and substitutions for the present invention are obvious. Therefore, the protection scope of the present invention shall be subject to the appended claims. Moreover, any accompanying drawing tag in a claim should not be considered as a limition on the related claim. A term "include" does not exclude a device or a step not listed in another claim or the specification. A term such as "first" or "second" is used to only indicate a name, but does not indicate any particular sequence.

Claims

1. A model search method based on a semantic model framework, comprising the following steps:

a buffering step of buffering and analyzing model query information of a user and buffering relative knowledge; and

a query step of querying a model in a buffer, an index, and a data library, comparing a model queried by the user and the model queried in the buffer, the index, and the data library, ranking relative models, returning a ranking result as a search result, and sending the search result to the user.

2. The model search method based on a semantic model framework according to claim 1, wherein after the query step, the method further comprises the following step: a storage step of storing the query result.

3. The model search method based on a semantic model framework according to claim 1, wherein before the buffering step, the method further comprises the following step: an analysis step of analyzing a semantic model framework and knowledge data, extracting knowledge, and generating knowledge reference data.

4. The model search method based on a semantic model framework according to claim 3, wherein the analysis step further comprises the following steps:

analyzing the semantic model framework by using classification information, and assigning the classification information to the knowledge data;

extracting segment information from the semantic model framework and the knowledge data; and

calculating probabilities of the segment information and the classification information.

5. The model search method based on a semantic model framework according to claim 1, wherein the buffering step further comprises the following steps: buffering the model query information of the user;

analyzing the model query information of the user, and comparing classification vocabularies and query vocabularies to classify the model query information of the user; and

replicating data that has a high probability and is of a same type.

6. A model search device based on a semantic model framework, comprising:

a buffering device, configured to buffer and analyze model query information of a user and buffer relative knowledge; and

a query device, configured to query a model in a buffer, an index, and a data library, compare a model queried by the user and the model queried in the buffer, the index, and the data library, rank relative models, return a ranking result as a search result, and send the search result to the user.

7. The model search device based on a semantic model framework according to claim 6, further comprising:

a storage device, configured to store the query result.

8. The model search device based on a semantic model framework according to claim 6, further comprising:

an analysis device, configured to analyze a semantic model framework and knowledge data, extract knowledge, and generate knowledge reference data.

9. The model search device based on a semantic model framework according to claim 8, wherein the analysis device further comprises:

a classification module (310), configured to analyze the semantic model framework by using classification information, and assign the classification information to the knowledge data;

an extraction module (320), configured to extract segment information from the semantic model framework and the knowledge data; and

a calculation module (330), configured to calculate probabilities of the segment information and the classification information.

10. The model search device based on a semantic model framework according to claim 1, wherein the buffering device further comprises:

a buffering device, configured to buffer the model query information of the user; an analysis device, configured to analyze the model query information of the user, and compare classification vocabularies and query vocabularies to classify the model query information of the user; and a replication device, configured to replicate data that has a high probability and is of a same type.