CN113283604A - Multi-element situation self-adaptive reasoning method and application - Google Patents
Multi-element situation self-adaptive reasoning method and application Download PDFInfo
- Publication number
- CN113283604A CN113283604A CN202110572122.8A CN202110572122A CN113283604A CN 113283604 A CN113283604 A CN 113283604A CN 202110572122 A CN202110572122 A CN 202110572122A CN 113283604 A CN113283604 A CN 113283604A
- Authority
- CN
- China
- Prior art keywords
- knowledge
- situation
- elements
- different
- entities
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/288—Entity relationship models
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
In order to solve the defects of the prior art, the invention provides a multi-element situation self-adaptive inference method and application, comprising the following steps: associating knowledge entities in the knowledge map with knowledge to obtain a knowledge iterative reasoning process representing a learning mixing rule; based on the knowledge iterative reasoning process of the expression learning mixing rule, searching elements which are effectively and strongly associated with different industrial chain situations, and establishing a category matching rule of the complex industrial chain situation and the different elements; acquiring weight of multi-element association of a complex industrial chain situation based on an entropy analysis method; analyzing elements corresponding to different situations through the category matching rules; and combining the weight associated with the multiple elements of the complex industry chain situation to obtain the self-adaptive reasoning method of the multiple element situation. The method can determine the category matching rule of the multi-element situation, and determine the influence weight of different production elements on the industrial chain decision through an entropy analysis method, thereby realizing the self-adaptive reasoning of various time-varying complex industrial chain situations.
Description
Technical Field
The invention relates to the technical field of industrial chain intelligent decision, in particular to a multi-element situation self-adaptive inference method applied to an industrial chain.
Background
The new generation artificial intelligence technology continuously attacks and customs around big data intelligence, group intelligence, industrial autonomous intelligent systems and other directions, and constructs the ecological environment of knowledge groups, technology groups and product groups from the aspects of basic theory, support systems, key technologies, innovation application and the like.
There are a number of links in the industrial chain system that require recommendations and decisions, all of which rely on an accurate sense of the industrial chain context. The traditional industry chain situation awareness technology has some applications, however, most of the technology is oriented to a single industry field, and research and application of the reasoning technology aiming at the multi-element situation of the complex industry chain are insufficient.
Disclosure of Invention
In order to solve the defects of the prior art, the invention provides a multi-element situation self-adaptive inference method and application, which aims to construct a multi-element situation self-adaptive inference technology covering the whole industrial chain in the industrial field by means of industrial big data, artificial intelligence and other technologies.
In order to achieve the purpose, the invention adopts the specific scheme that:
a multi-element situation self-adaptive reasoning method comprises the following steps:
associating knowledge entities in the knowledge map with knowledge to obtain a knowledge iterative reasoning process representing a learning mixing rule;
based on the knowledge iterative reasoning process of the expression learning mixing rule, searching elements which are effectively and strongly associated with different industrial chain situations, and establishing a category matching rule of the complex industrial chain situation and the different elements;
acquiring weight of multi-element association of a complex industrial chain situation based on an entropy analysis method;
analyzing elements corresponding to different situations through the category matching rules; and combining the weight associated with the multiple elements of the complex industry chain situation to obtain the self-adaptive reasoning method of the multiple element situation.
The method for obtaining the knowledge iterative inference process expressing the learning mixing rule by associating the knowledge entity with the knowledge in the knowledge graph comprises the following steps:
associating the existing knowledge entities in the knowledge map with knowledge, and learning by utilizing representation to obtain new knowledge entities;
analyzing the support degree, the confidence degree and the promotion degree of the correlation by a correlation analysis method, and extracting an effective strong correlation set;
and performing expression learning through a strong association rule to obtain the expression learning mixed rule knowledge iterative reasoning process.
The method for obtaining the new knowledge entity by means of representation learning through the association of the existing knowledge entity and knowledge in the knowledge graph comprises the following steps:
based on the existing knowledge entities and association rules in the industry chain knowledge graph, embedding the associated knowledge entities and the triples of the corresponding association rules into a low-dimensional continuous vector space for processing;
carrying out normalization processing on the initialized embedded vectors of the knowledge entities and the association rules;
through the TranSparse model, the knowledge entities according to different headsAnd relation ofObtaining new knowledge entities
The knowledge entityThe knowledge graph of the industry chain is supplemented, and the process of the knowledge graph is satisfiedEvaluation was performed by a scoring function:
wherein the content of the first and second substances,θminis a hyper-parameter related to sparsity, satisfies 0 ≦ θmin≤1,NrIs the number of pairs of entities to which the relationship r connects,is the most numerous relationship of connected entities.
In addition, to solve the problem of unbalanced association relationship, the scoring function employed in the TranSparse model can be further expressed as:
wherein embh,embtIs that different projection matrixes for solving the unbalanced incidence relation in the TranSparse model correspond toNamely different sparsity of the head and tail knowledge entities, the following sparsity is satisfied:
the "analyzing the support degree, the confidence degree and the promotion degree of the association by an association analysis method to extract an effective strong association set" includes:
selecting a rule set with high confidence by combining a correlation analysis method with a knowledge graph, wherein the method comprises the following steps:
wherein, the analysis of the support degree:
the above formula representsWherein I represents the total set of the industry chain knowledge-graphs, and num () represents the number of times of occurrence in a certain set;
wherein, the confidence coefficient analysis:
the above formula represents the knowledge entity from the headDeducing new tail knowledge entity by incidence relation rThe probability of (d);
wherein, the analysis of the lifting degree:
the above formula shows that a part of the knowledge graph containsUnder the conditions of (A) and (B) at the same time containIs a probability of not containingUnder the condition of containingThe ratio of the probabilities of (a);
distinguishing strong association rules from invalid strong association rules by:
the "learning mixing rule knowledge iterative inference process based on the expression, searching for elements having effective and strong association with different industrial chain situations, and establishing a category matching rule of a complex industrial chain situation and different elements" includes:
under different types of industrial chain situations, mining elements relevant to the situations according to different situations through the expression learning mixing rule knowledge iterative reasoning process based on the industrial chain knowledge map;
introducing a membership function, analyzing the association degree of the mined elements and the industrial chain situation, and establishing a category matching rule of the complex industrial chain situation and different elements:
A(x)∈[0,1]
the above formula characterizes how high or low the element x belongs to the context a.
The method for obtaining the weight of the multi-element association of the complex industry chain situation based on the entropy analysis method comprises the following steps:
selecting n scenarios containing the same m elements, then xijThe value of the j element of the i-th scenario (i-1, 2 …, n; j-1, 2 …, m);
normalizing the index to let xij=|xij|:
x'ijThe value of the j element of the i contexts (i ═ 1, 2 …, n; j ═ 1, 2 …, m);
x'ijIs marked as xijAcquiring the proportion of the jth element under the ith situation:
Obtaining information entropy redundancy: dj=1-ej;
has the advantages that: the knowledge graph reasoning method adopts the representation learning method to carry out knowledge graph reasoning, effectively solves the problems of heterogeneity and imbalance existing in the association relation between knowledge entities in the knowledge graph of the industrial chain, and simultaneously adopts the association analysis method to extract the strong association rule set, thereby further improving the knowledge reasoning efficiency and realizing the knowledge iterative reasoning of the representation learning mixed rule; meanwhile, due to the fact that the actual production environment has complex conditions of multiple links, long flow, complex dynamic constraint, time-varying industrial situation and the like, the method determines the category matching rule of the multi-element situation by knowledge reasoning and membership function analysis; and determining the influence weight of different production elements on the industrial chain decision aiming at the same elements under different industrial chain situations through an entropy analysis method, and adjusting the weight ratio of different production elements in real time according to the analysis result and the actual production environment to finally realize the self-adaptive reasoning of various time-varying complex industrial chain situations.
Aiming at the problems of heterogeneity and unbalance existing in the association relation between knowledge entities in the knowledge graph of the industrial chain, the knowledge iterative inference expressing the learning mixing rule is realized by adopting a combination of expression learning and an association analysis method; the method can determine the category matching rule of the multi-element situation by combining the membership function analysis through a knowledge reasoning method according to the complex situation in the actual production environment; moreover, aiming at the problem that the same elements exist in different industrial chain situations and the different situations are difficult to further distinguish, the influence weight of different production elements on industrial chain decision is determined through an entropy analysis method, and the adaptive reasoning of various time-varying complex industrial chain situations is realized.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
As shown in fig. 1, when the multi-element situation adaptive inference method of the present invention is applied to an industrial industry chain, the method specifically includes the following steps:
step 1: establishing a knowledge iterative inference method based on expression learning mixing rules:
step 1.1: knowledge inference method based on representation learning
Based on the existing knowledge entities and association rules in the industry chain knowledge graph, embedding (embedding) the associated knowledge entities and the triples of the corresponding association rules into a low-dimensional continuous vector space, for example:
wherein, the emb () represents the initial embedding method of the knowledge entity and the association rule,representing the head (head) knowledge entity, association rules, tail (tail) knowledge entity, and their initialization embedding vectors, respectively.
Then, the initialized embedded vector of the knowledge entity and the association rule is normalized, such as:
wherein the content of the first and second substances,euclidean distances representing knowledge entities and their initialized embedded vectors.
There are many representations of learning models used to learn the representation of entities and relationships in the knowledge graph, including TransE, TransH, TransR, TransSparse, and FT models, among which the TransSparse model can deal well with the heterogeneity and imbalance in the knowledge graph of the industry chain, where heterogeneity refers to the difference in the number of entities connected by relationships in the knowledge graph, some relationships may be connected by a large number of entities, and some relationships may be connected by only a small number of entities. Imbalance means that the types and numbers of head and tail entities connected by relationships are different and may be very different.
Through the TranSparse model, the knowledge entities according to different headsAnd relation ofObtaining new knowledge entitiesAnd the knowledge graph of the industry chain is supplemented, and the process of the knowledge graph of the industry chain is satisfiedThe evaluation can be done by a scoring function:
wherein the content of the first and second substances,θminis a hyper-parameter related to sparsity, satisfies 0 ≦ θmin≤1,NrIs the number of pairs of entities to which the relationship r connects,is the most numerous relationship of connected entities.
In addition, to solve the problem of unbalanced association relationship, the scoring function employed in the TranSparse model can be further expressed as:
wherein embh,embtIs that different projection matrixes for solving the unbalanced incidence relation in the TranSparse model correspond toNamely different sparsity of the head and tail knowledge entities, the following sparsity is satisfied:
In order to improve the efficiency of knowledge inference and knowledge graph supplementation, on the basis of the step 1.1, association rules need to be mined, and association rules capable of learning and inferring new knowledge with higher efficiency are extracted, and in order to achieve the purpose, an association analysis method is adopted to select a rule set with high confidence degree in combination with a knowledge graph, and the method comprises the following steps:
support (Support) analysis:
the above formula representsWherein I represents the total set of industry chain knowledge-graphs, num () represents the number of occurrences in a certain set,
confidence (Confidence) analysis:
the above formula represents the knowledge entity from the headDeducing new tail knowledge entity by incidence relation rThe probability of (c).
Lift (Lift) analysis:
the above formula shows that a part of the knowledge graph containsUnder the conditions of (A) and (B) at the same time containIs a probability of not containingUnder the condition of containingThe ratio of the probabilities of (a) to (b).
The rule which satisfies the minimum support degree and the minimum confidence coefficient is called as a strong association rule through analysis based on the method. However, in the strong association rule, the valid strong association rule and the invalid strong association rule are also distinguished by the following method:
in summary, the step aims to obtain a new knowledge entity through representation learning by associating the existing knowledge entity with knowledge in the knowledge map, and meanwhile, in order to improve the efficiency of knowledge representation learning and avoid excessive calculation power on irrelevant association rules, the support degree, confidence degree and promotion degree of association are analyzed through an association analysis method, so that an effective strong association set is extracted, and then representation learning is performed through strong association rules, so that the knowledge iterative inference method based on the representation learning mixing rules of the knowledge map is realized.
Step 2: knowledge iterative reasoning-based industrial chain situation multi-element category matching rule:
and (3) based on the expression learning mixed rule knowledge iterative reasoning in the step (1), searching different elements which are effectively and strongly associated with different industrial chain situations, and establishing a category matching rule of the complex industrial chain situation and the different elements. The method comprises the following specific steps:
under different types of industrial chain situations, based on the industrial chain knowledge graph, by the knowledge iterative reasoning method in the step 1, elements relevant to the situations, such as places, time, people, products, production data and the like, are fully mined aiming at different situations.
Then, introducing a membership function, and analyzing the association degree of the mined elements and the industrial chain situation:
A(x)∈[0,1]
the above formula characterizes how high or low the element x belongs to the context a.
Through the membership function, elements with high correlation degree in different industrial chain situations can be distinguished, and therefore the multi-element class matching rule for repeating the complex industrial chain situations is established.
And step 3: establishing multi-element association weight of a complex industrial chain situation based on an entropy analysis method:
according to task requirements, firstly, in a complex industry chain situation, different situations may match the same category element, and in the case of the matching of the same element, an entropy analysis method is introduced, and under the condition that the matched elements are the same, the self-adaptive association weight of the industry chain situation element is further established according to different functions of the same element in different situations.
First, n scenarios containing the same m elements are selected, and then x is obtainedijIs the value of the j-th element of the first context (i-1, 2 …, n; j-1, 2 …, m).
Then normalizing the indexes, i.e. homogenizing heterogeneous indexes, because the measurement units of all indexes are not uniform, before calculating the comprehensive index by using the indexes, normalizing the indexes, i.e. converting the absolute value of the index into a relative value and making x be a relative valueij=|xijTherefore, the homogenization problem of various heterogeneous index values is solved. Moreover, since the positive index and the negative index have different meanings (the higher the positive index value is, the better the negative index value is), the data normalization processing is performed on the high and low indexes by using different algorithms. The specific method comprises the following steps:
x'ijIs the value of the j-th element of the i contexts (i ═ 1, 2 …, n; j ═ 1, 2 …, m). For convenience, x will remainijIs marked as xij。
Then, the specific gravity of the j element in the i context is calculated:
Calculating the information entropy redundancy: dj=1-ej。
therefore, by the method of entropy analysis, the adaptive association weight of the context elements of the industry chain can be further established according to different functions of the same element under different contexts.
And 4, step 4: the self-adaptive reasoning method of the multi-element situation comprises the following steps:
the step 2 and the step 3 are combined, elements corresponding to different situations are analyzed through the situation category matching rules established in the step 2, then the multi-element association weight obtained through entropy analysis is combined, the self-adaptive reasoning method of the multi-element situation is further obtained, and accurate reasoning of the complex industrial chain situation with various elements and time varying is achieved.
The above description is only an embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily change or replace the present invention within the technical scope of the present invention. Therefore, the protection scope of the present invention is subject to the protection scope of the claims.
Claims (8)
1. A multi-element situation self-adaptive reasoning method is characterized by comprising the following steps:
associating knowledge entities in the knowledge map with knowledge to obtain a knowledge iterative reasoning process representing a learning mixing rule;
based on the knowledge iterative reasoning process of the expression learning mixing rule, searching elements which are effectively and strongly associated with different industrial chain situations, and establishing a category matching rule of the complex industrial chain situation and the different elements;
acquiring weight of multi-element association of a complex industrial chain situation based on an entropy analysis method;
analyzing elements corresponding to different situations through the category matching rules; and combining the weight associated with the multiple elements of the complex industry chain situation to obtain the self-adaptive reasoning method of the multiple element situation.
2. The multi-element context adaptive inference method according to claim 1, wherein said "associating knowledge entities in a knowledge graph with knowledge to obtain a knowledge iterative inference process representing a learning mixing rule", comprises:
associating the existing knowledge entities in the knowledge map with knowledge, and learning by utilizing representation to obtain new knowledge entities;
analyzing the support degree, the confidence degree and the promotion degree of the correlation by a correlation analysis method, and extracting an effective strong correlation set;
and performing expression learning through a strong association rule to obtain the expression learning mixed rule knowledge iterative reasoning process.
3. The method of claim 2, wherein the "learning to obtain new knowledge entity by representation through association of knowledge entity with knowledge existing in knowledge map" comprises:
based on the existing knowledge entities and association rules in the industry chain knowledge graph, embedding the associated knowledge entities and the triples of the corresponding association rules into a low-dimensional continuous vector space for processing;
carrying out normalization processing on the initialized embedded vectors of the knowledge entities and the association rules;
4. The multi-element context adaptive inference method according to claim 3, characterized by:
the knowledge entityThe knowledge graph of the industry chain is supplemented, and the process of the knowledge graph is satisfiedEvaluation was performed by a scoring function:
wherein the content of the first and second substances,θminis a hyper-parameter related to sparsity, satisfies 0 ≦ θmin≤1,NrIs the number of pairs of entities to which the relationship r connects,is the most numerous relationship of connected entities;
in addition, the scoring function employed in the TranSparse model can be further expressed as:
5. the multi-element context adaptive inference method according to claim 2, wherein said "analyzing support, confidence and promotion of said association by association analysis method to extract effective strong association set" comprises:
selecting a rule set with high confidence by combining a correlation analysis method with a knowledge graph, wherein the method comprises the following steps:
wherein, the analysis of the support degree:
the above formula representsWherein I represents the total set of the industry chain knowledge-graphs, and num () represents the number of times of occurrence in a certain set;
wherein, the confidence coefficient analysis:
the above formula represents the knowledge entity from the headDeducing new tail knowledge entity by incidence relation rThe probability of (d);
wherein, the analysis of the lifting degree:
the above formula shows that a part of the knowledge graph containsUnder the conditions of (A) and (B) at the same time containIs a probability of not containingUnder the condition of containingThe ratio of the probabilities of (a);
distinguishing strong association rules from invalid strong association rules by:
6. the method according to claim 1, wherein said "finding elements that are strongly and effectively associated with different industry chain contexts based on said representation learning blending rule knowledge iterative inference process, and establishing a category matching rule between a complex industry chain context and different elements" comprises:
under different types of industrial chain situations, mining elements relevant to the situations according to different situations through the expression learning mixing rule knowledge iterative reasoning process based on the industrial chain knowledge map;
introducing a membership function, analyzing the association degree of the mined elements and the industrial chain situation, and establishing a category matching rule of the complex industrial chain situation and different elements:
A(x)∈[0,1]
the above formula characterizes how high or low the element x belongs to the context a.
7. The multi-element situation adaptive inference method according to claim 1, wherein said "obtaining weights of multi-element associations of complex industry chain scenarios based on entropy analysis" comprises:
selecting n scenarios containing the same m elements, then xijThe value of the j element of the i-th situation (i 1, 2., n; j 1, 2., m);
normalizing the index to let xij=|xij|:
x'ijThe value of the j element of the i contexts (i 1, 2.., n; j 1, 2.., m);
Obtaining information entropy redundancy: dj=1-ej;
8. an application of the multi-element situation adaptive inference method according to any one of claims 1-7 in the direction of industrial industry chain.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110572122.8A CN113283604A (en) | 2021-05-25 | 2021-05-25 | Multi-element situation self-adaptive reasoning method and application |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110572122.8A CN113283604A (en) | 2021-05-25 | 2021-05-25 | Multi-element situation self-adaptive reasoning method and application |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113283604A true CN113283604A (en) | 2021-08-20 |
Family
ID=77281466
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110572122.8A Pending CN113283604A (en) | 2021-05-25 | 2021-05-25 | Multi-element situation self-adaptive reasoning method and application |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113283604A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116029604A (en) * | 2023-02-03 | 2023-04-28 | 华南农业大学 | Cage-raised meat duck breeding environment regulation and control method based on health comfort level |
-
2021
- 2021-05-25 CN CN202110572122.8A patent/CN113283604A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116029604A (en) * | 2023-02-03 | 2023-04-28 | 华南农业大学 | Cage-raised meat duck breeding environment regulation and control method based on health comfort level |
CN116029604B (en) * | 2023-02-03 | 2024-01-23 | 华南农业大学 | Cage-raised meat duck breeding environment regulation and control method based on health comfort level |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111126482B (en) | Remote sensing image automatic classification method based on multi-classifier cascade model | |
Peng et al. | A new approach for imbalanced data classification based on data gravitation | |
CN108717568A (en) | A kind of image characteristics extraction and training method based on Three dimensional convolution neural network | |
CN110309331A (en) | A kind of cross-module state depth Hash search method based on self-supervisory | |
CN107766850A (en) | Based on the face identification method for combining face character information | |
CN113076994B (en) | Open-set domain self-adaptive image classification method and system | |
CN107194524B (en) | RBF neural network-based coal and gas outburst prediction method | |
CN110210515A (en) | A kind of image data multi-tag classification method | |
CN111127385A (en) | Medical information cross-modal Hash coding learning method based on generative countermeasure network | |
CN106845528A (en) | A kind of image classification algorithms based on K means Yu deep learning | |
Nowak et al. | Multi-class nearest neighbour classifier for incomplete data handling | |
CN113259331A (en) | Unknown abnormal flow online detection method and system based on incremental learning | |
CN102706573A (en) | Fault classification diagnosis method of equipment | |
CN111401156A (en) | Image identification method based on Gabor convolution neural network | |
CN108877947A (en) | Depth sample learning method based on iteration mean cluster | |
CN114841151B (en) | Medical text entity relation joint extraction method based on decomposition-recombination strategy | |
CN112183652A (en) | Edge end bias detection method under federated machine learning environment | |
CN112560948A (en) | Eye fundus map classification method and imaging method under data deviation | |
CN108520201A (en) | Robust face recognition method based on weighted mixed norm regression | |
CN113283604A (en) | Multi-element situation self-adaptive reasoning method and application | |
CN109617864A (en) | A kind of website identification method and website identifying system | |
CN112149556B (en) | Face attribute identification method based on deep mutual learning and knowledge transfer | |
CN110414407B (en) | Ship propeller air entrainment identification method based on evidence reasoning rule and self-adaptive lifting | |
CN112926739A (en) | Network countermeasure effectiveness evaluation method based on neural network model | |
Zhang et al. | A method of deep belief network image classification based on probability measure rough set theory |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |