Data association construction method based on deep learning

Abstract

The invention relates to a data association construction method based on deep learning, and relates to the field of data association relationship construction. According to the method, firstly, a semantic data dictionary based on the business flow is established according to the field business flow; then, according to an actual field system application habit, a data object model based on deep learning is trained, then automatic association relationship construction of the data object model is achieved according to an enterprise production process, and automatic construction of the association relationship of enterprise data information on the premise of context information can be achieved; and the efficiency of using and querying the data by the enterprise is improved.

Description

Technical Field

[0001] The present invention belongs to the field of constructing data association relationships, and proposes a method for constructing data association based on deep learning. Background Art

[0002] Currently, a new round of technological revolution represented by big data, cloud computing, mobile Internet, etc. has swept the globe, and is building a new industrial system of information interconnection, resource sharing, ability collaboration, and open cooperation, greatly expanding the innovation and development space of the manufacturing industry. The development of a new generation of information and communication technologies drives the transformation and upgrading of the manufacturing industry - the arrival of the data-driven era. In this context, the availability, analyzability, and executability of industrial data throughout the entire production process, the entire industrial chain, and the entire product life cycle of the manufacturing industry are important supports for realizing the transformation and upgrading of the manufacturing industry, and are also the inherent requirements for continuously making breakthroughs in the explicitization of implicit knowledge in the manufacturing industry. In this process, data flow is the key. From the perspective of data flow, digitization solves the problem of "having data", networking solves the problem of "being able to flow", and intelligence needs to solve the problem of data "automatically flowing" to cope with and solve problems such as the complexity and uncertainty of the manufacturing process, and continuously improve the allocation efficiency of manufacturing resources in this process.

[0003] Data is the sum of data throughout the entire life cycle in the industrial production process, including design materials during product R & D; monitoring and management data during product production; operation and maintenance data during product sales and service, etc. The automatic flow of data is supported by data streams. In an industrial manufacturing system, the flow of data is carried out in an orderly manner according to business processes. Industrial data has a wide range of sources, and a large amount of data will be continuously generated at each key link in the production process. For example, unstructured design materials in the design link, structured sensor and monitoring data in the production process, customer and transaction data in the management process, and relevant data from external industries, etc. Different data will be applied at different stages. Currently, in a local production and manufacturing scenario, a single system can be used to achieve calculations, controls, optimizations, and collections of local operations. However, to achieve global application across the entire industrial chain, automatically sorting out the intricate data will greatly improve the production efficiency of enterprises, and the establishment of global data association relationships will provide support for each production subdivision link of enterprises. Summary of the Invention

[0004] The object of the present invention is to propose a method for constructing data association based on deep learning, which can solve problems such as data circulation between various systems and devices of an enterprise, and at the same time can automatically construct data association relationships according to the upstream and downstream production processes of the enterprise, improving the efficiency of the enterprise in using and querying data.

[0005] The technical solution adopted by the present invention to achieve the above object is:

[0006] A data association construction method based on deep learning, comprising the following steps:

[0007] 1) Construct a semantic data dictionary based on the business process;

[0008] 2) Identify enterprise data according to the data object model based on deep learning;

[0009] 3) Perform semantic object matching on the recognition result and the semantic data dictionary to find its corresponding position in the semantic data dictionary;

[0010] 4) Connect the sequences with an association relationship in the semantic data dictionary to complete the automatic association construction of industrial data relationships.

[0011] The semantic data dictionary Dic is as follows:

[0012] Dic = <ID, Type, SN, Des, Context, Ver>

[0013] Wherein, ID represents the ID of the data dictionary entry, Type represents the type of the entry, SN represents the ShortName of the entry, Des represents the explanation of the entry, Context represents the ID of the upper-level entry to which the entry belongs, and Ver represents the version of the semantic data dictionary.

[0014] The data object model based on deep learning includes: an embedding representation layer for text representation, a semantic encoding layer for extracting semantic features contained in the vector, and a label decoding layer for predicting the label corresponding to the input sequence.

[0015] The step 2) includes the following steps:

[0016] 2.1) The embedding representation layer divides the enterprise data into a character-level text representation method and a word-level text representation method according to the granularity of text representation;

[0017] 2.2) The semantic encoding layer processes the sequence data output by the embedding representation layer using the RNN method;

[0018] 2.3) The label decoding layer uses the conditional random field method to perform division prediction on the input sequence processed by the semantic encoding layer.

[0019] Use a convolutional network and a pooling layer to extract the vector of the text to obtain a character-level text representation method; use the continuous bag-of-words model and the skip-gram model to represent the vocabulary with low-dimensional vectors, and capture the semantic similarity between words through word training to obtain a word-level text representation method.

[0020] A data association construction system based on deep learning, comprising:

[0021] A semantic data dictionary construction module for constructing a semantic data dictionary based on business processes;

[0022] A data object model recognition module for recognizing enterprise data according to a data object model based on deep learning;

[0023] A semantic matching module for performing semantic object matching between the recognition result and the semantic data dictionary to find its corresponding position in the semantic data dictionary;

[0024] A data association module for associatively connecting sequences having an association relationship in the semantic data dictionary to complete the automatic association construction of industrial data relationships.

[0025] The data object model based on deep learning includes:

[0026] An embedding representation layer for classifying enterprise data into a character-level text representation method and a word-level text representation method according to the granularity of text representation;

[0027] A semantic encoding layer for processing the sequence data output by the embedding representation layer using an RNN method;

[0028] A label decoding layer for performing partition prediction on the input sequence processed by the semantic encoding layer using a conditional random field method.

[0029] A data association construction device based on deep learning includes a memory and a processor; the memory is used for storing a computer program; the processor is used for implementing the described data association construction method based on deep learning when executing the computer program.

[0030] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the described data association construction method based on deep learning is implemented.

[0031] The present invention has the following beneficial effects and advantages:

[0032] 1. The present invention can solve the problem that data of various production systems, devices, sensors, etc. in enterprises cannot flow. Through the present invention, the automatic association relationship construction of data according to the enterprise production process is realized, so that data can flow seamlessly in each system, greatly improving the production efficiency of enterprises.

[0033] 2. The present invention can sort out and preprocess enterprise data, aiming to form a data flow diagram with a global, full-domain or full-supply chain perspective through automatic association analysis of a large amount of production process data, so as to provide support for enterprises to improve production efficiency and provide intelligent decision-making. Description of the Drawings

[0034] Figure 1 The architecture diagram of the automatic association method for data relationships of the present invention;

[0035] Figure 2 The structure diagram of the RNN algorithm;

[0036] Figure 3 The schematic flow diagram of the present invention. Detailed implementation manners

[0037] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0038] The present invention extracts various types of production data information of enterprises, generates a data object model based on deep learning, then performs semantic object matching between the recognition results and the semantic data dictionary, and performs associated connection in the semantic data dictionary in combination with the upstream and downstream relationships of enterprise production, so as to complete the automatic association construction of industrial data relationships.

[0039] As Figure 1 shown is the architecture diagram of the automatic association method for data relationships, and its specific working process is

[0040] 1. Construct a semantic data dictionary based on the business process, fully considering the characteristics of the industrial production process, and expressing various relationships therein separately. Forms of display such as a business association diagram or a knowledge graph in the form of RDF are formed.

[0041] 2. The format of creating the semantic data dictionary is Dic = <ID, Type, SN, Des, Context, Ver>, where ID represents the ID of the data dictionary entry and needs to be unique; Type represents the type of the entry, usually divided into six types: class, attribute, relationship, class-relationship-class, class-attribute-value, unit, etc.; SN represents the Short Name of the entry, which is a simplified expression of the conventional data meaning; Des represents the explanation of the entry; Context represents the ID of the upper-level entry to which the entry belongs; Ver represents the version of the semantic data dictionary.

[0042] 3. The object data of man, machine, material, method, and environment in the enterprise production and manufacturing process are generally stored in software systems, and data entry is realized through automatic or manual reports. The systems for storage generally include EMS, MES, reports, QMS, ERP, etc. These data information is collected

[0043] 4. These data are mainly stored in a relational database. For the enterprise production and manufacturing system, the online working times of such software systems are different and the software vendors are different, so the data is difficult to interoperate and cannot form an industrial big data correlation analysis from a global perspective. Through the deep learning data object model and the data dictionary, these data are uniformly described and recognized, and the specific processing process of the deep learning data object model is as follows:

[0044] First, generate an embedded representation layer of the data object model based on deep learning. The embedded representation layer should provide as much rich and effective prior knowledge to the model as possible. According to the granularity of text representation, it can be divided into character-level text representation methods and word-level text representation methods. The character-level text representation method is obtained by vector convolution; the word-level text representation method is composed of the continuous bag-of-words model and the skip-gram model, and dense low-dimensional vectors are used to represent vocabulary. Capture the semantic similarity between words through word and character training

[0045] Secondly, generate a semantic encoding layer of the data object model based on deep learning. The semantic encoding layer uses the RNN method to process sequence data. RNN is a recursive neural network with sequence data as input, and its structure is as Figure 2 shown, where the structure includes an input layer unit node expression of x = {x 1 , x 2 ,..., x t}. The hidden layer unit node expression is h = {h 1 , h 2 ,..., h t}, and the output layer unit node expression is y = {y 1 , y 2 ,..., y t}. There is a unidirectional information flow in all hidden nodes in the unfolded structure, which represents the transmission of sequence-related information in the model.

[0046] Finally, generate a decoding layer of the data object model based on deep learning, and use the Conditional RandomFields (CRF) method to perform division prediction on the input sequence. CRF is a discriminative probabilistic undirected graph model and also a method for modeling conditional probability distributions.

[0047] 5. Use the database name + table name + column name of the data as the sequence of business data to be associated, and use the data object recognition model based on deep learning to perform entity object recognition.

[0048] 6. Match the recognition result with the semantic data dictionary to find its corresponding position in the semantic data dictionary.

[0049] 7. Perform association connection on the sequences with association relationships in the semantic data dictionary to complete the automatic association construction of industrial data relationships.

[0050] Such as Figure 3As shown in the figure, a method for constructing data association based on deep learning includes a semantic data dictionary based on business processes, a data object model based on deep learning, and a method for automatically associating data relationships based on context. The enterprise data is identified according to the data object model based on deep learning, and then the recognition result is subjected to semantic object matching with the semantic data dictionary to find its corresponding position in the semantic data dictionary, and the sequences having an association relationship in the semantic data dictionary are associated and connected to complete the automatic association construction of industrial data relationships.

[0051] The semantic data dictionary based on business processes is specifically as follows. This semantic data dictionary has a hierarchy, which is defined as follows:

[0052] Dic = <ID, Type, SN, Des, Context, Ver>

[0053] ID represents the ID of the data dictionary entry and needs to be unique;

[0054] Type represents the type of the entry, which can generally be divided into six types: class, attribute, relationship, class-relationship-class, class-attribute-value, and unit;

[0055] SN represents the Short Name of the entry, which is a simplified expression of the conventional data meaning. In particular, synonyms, abbreviations, and the initials of Chinese pinyin need to be clearly expressed. Generally, different IDs will be used to support the naming entity namespace rule;

[0056] Des represents the explanation of the entry;

[0057] Context represents the ID of the upper-level entry to which the entry belongs;

[0058] Ver represents the version of the semantic data dictionary.

[0059] The data object model based on deep learning is divided into three parts: an embedding representation layer, a semantic encoding layer, and a label decoding layer. The embedding representation layer mainly performs text representation, the semantic encoding layer mainly extracts the semantic features contained in the vector, and the function of the label decoding layer is to predict the label corresponding to the input sequence.

[0060] The embedding representation layer is divided into a character-level (char-level) text representation method and a word-level (word-level) text representation method according to the granularity. The character level is obtained using convolution, and a convolutional network and a pooling layer can be used to extract the vector of the text; the word-level text uses the continuous bag-of-words model and the skip-gram model to capture the semantic similarity between words.

[0061] The semantic encoding layer uses the RNN method to process sequence data. RNN is a recursive neural network with sequence data as the input.

[0062] The described label decoding layer uses the Conditional Random Fields (CRF) method to perform division prediction on the input sequence. CRF is a discriminative probabilistic undirected graph model and also a method for modeling conditional probability distributions.

Claims

1. A method for constructing data association based on deep learning, characterized in that, It includes the following steps: 1) Construct a semantic data dictionary based on the business process; 2) Identify enterprise data according to the data object model based on deep learning; 3) Perform semantic object matching between the recognition result and the semantic data dictionary to find its corresponding position in the semantic data dictionary; 4) Associate and connect the sequences with an association relationship in the semantic data dictionary to complete the automatic association construction of industrial data relationships.

2. The data association construction method based on deep learning according to claim 1, characterized in that The semantic data dictionary Dic is as follows: Dic = <ID, Type, SN, Des, Context, Ver> Among them, ID represents the ID of the data dictionary entry, Type represents the type of the entry, SN represents the Short Name of the entry, Des represents the explanation of the entry, Context represents the ID of the upper-level entry to which the entry belongs, and Ver represents the version of the semantic data dictionary.

3. A data association construction method based on deep learning according to claim 1, characterized in that, The data object model based on deep learning includes: an embedding representation layer for text representation, a semantic encoding layer for extracting semantic features contained in the vector, and a label decoding layer for predicting the label corresponding to the input sequence.

4. A method for constructing data association based on deep learning according to claim 1, characterized in that The step 2) includes the following steps: 2.1) The embedding representation layer divides the enterprise data into a character-level text representation method and a word-level text representation method according to the granularity of text representation; 2.2) The semantic encoding layer uses the RNN method to process the sequence data output by the embedding representation layer; 2.3) The label decoding layer uses the conditional random field method to perform division prediction on the input sequence processed by the semantic encoding layer.

5. A method for constructing data association based on deep learning according to claim 4, characterized in that, Use a convolutional network and a pooling layer to extract the vector of the text to obtain a character-level text representation method; use the continuous bag-of-words model and the skip-gram model to represent the vocabulary with low-dimensional vectors, and capture the semantic similarity between words through word training to obtain a word-level text representation method.

6. A data association construction system based on deep learning, characterized in that, It includes: A semantic data dictionary construction module for constructing a semantic data dictionary based on the business process; A data object model recognition module for identifying enterprise data according to the data object model based on deep learning; A semantic matching module for performing semantic object matching between the recognition result and the semantic data dictionary to find its corresponding position in the semantic data dictionary; A data association module for associating and connecting the sequences with an association relationship in the semantic data dictionary to complete the automatic association construction of industrial data relationships.

7. The data association construction system based on deep learning according to claim 6, wherein, The data object model based on deep learning includes: An embedding representation layer for dividing the enterprise data into a character-level text representation method and a word-level text representation method according to the granularity of text representation; A semantic encoding layer for using the RNN method to process the sequence data output by the embedding representation layer; A label decoding layer for using the conditional random field method to perform division prediction on the input sequence processed by the semantic encoding layer.

8. A data association construction device based on deep learning, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement a data association construction method based on deep learning as described in any one of claims 1-5 when executing the computer program.

9. A computer-readable storage medium, characterized in that, A computer program is stored on the storage medium, and when the computer program is executed by a processor, a data association construction method based on deep learning as described in any one of claims 1-5 is implemented.

Images