A method and system for determining industrial process model
By combining BERT, a bidirectional long-short word memory model, and a conditional random field model, a hierarchical structure tree is generated and node priorities are traversed, which solves the problem of low accuracy of industrial process models in existing technologies and achieves higher recognition accuracy.
Patent Information
- Application Number
- CN202211312876.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-25
- Publication Date
- 2025-10-03
- Estimated Expiration
- 2042-10-25
AI Technical Summary
The existing method of establishing an end-to-end automated mapping and extraction process model through neural networks has low accuracy when processing industrial process text and is difficult to accurately meet actual needs.
The BERT model, bidirectional long-short word memory model and conditional random field model are used to extract activity entity sets from industrial process text documents. An ordered neural network is used to generate a hierarchical structure tree, which is combined with a top-down greedy algorithm and node priority traversal to determine the target industrial process model.
The accuracy of industrial process model determination is improved, and business processes in industrial process text documents can be identified more accurately.
Smart Images
Figure CN115983204B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of model determination, and in particular to a method and system for determining an industrial process model. Background Art
[0002] One of the current hot topics in industrial software research is how to discover the industrial process models underlying industrial software execution. Some researchers have proposed automated process model extraction methods to help process analysts efficiently, cost-effectively, and accurately identify the industrial process models embedded in software execution within large amounts of industrial process text and method documentation in a short period of time. Although the automated extraction of industrial process models from industrial software descriptions is still in its early stages, numerous researchers have conducted a series of studies and achieved considerable success.
[0003] Existing methods for establishing end-to-end automated mapping and extraction process models through neural networks employ deep learning methods to intelligently and automatically determine process models. A representative example is the method proposed by Han et al., which uses business process language modeling to establish the latent structure in industrial process text method documents (A-BPS). This method uses deep learning methods and can retrieve document trees without any manual labeling. The method represents the hierarchical structure of sentences in the document using a tree structure, with each node in the tree containing a sentence. First, the authors extend the encoder using a bidirectional long short-term memory (BiLSTM) model, which embeds sentences and obtains a vector representation. The sentence vector representation is then used as input to an ordered neural network LSTM (ON-LSTM) for unsupervised retrieval of the sentence hierarchy in the text method document. This hierarchy is represented as a tree structure. However, in real industrial process text method documents, a sentence may contain multiple activities or no activities. A-BPS, which treats a sentence as a node and assumes only one activity, struggles to accurately meet real-world needs and has low accuracy. Summary of the Invention
[0004] The purpose of the present invention is to provide a method and system for determining an industrial process model, thereby improving the accuracy of determining the industrial process model.
[0005] To achieve the above object, the present invention provides the following solutions:
[0006] A method and system for determining an industrial process model, the method comprising:
[0007] Obtaining industrial process text documents to be identified;
[0008] Extracting an activity entity set from the industrial process text document to be identified using a BERT model, a bidirectional long-short word memory model, and a conditional random field model; the activity entity set includes multiple activity entities;
[0009] Inputting the activity entity set into an ordered neural network for identification to obtain a hierarchical tree; each node of the hierarchical tree corresponds to one of the activity entities;
[0010] The hierarchical structure tree is traversed according to the priority of each node to obtain a target industrial process model, wherein each node of the hierarchical structure tree corresponds to a business in the target industrial process model; the target industrial process model is used to describe the business process contained in the industrial process text document to be identified.
[0011] Optionally, the extracting of the activity entity set from the industrial process text document to be identified by using the BERT model, the bidirectional long-short word memory model, and the conditional random field model specifically includes:
[0012] Inputting the industrial process text document into the BERT model for encoding to obtain multiple word vectors;
[0013] Inputting each of the word vectors into the bidirectional long-short word memory model to obtain multiple word sequences;
[0014] Each of the word sequences is input into the conditional random field model for decoding to obtain the activity entity set.
[0015] Optionally, inputting the active entity set into an ordered neural network for identification to obtain a hierarchical tree specifically includes:
[0016] Calculating the distance between each active entity in the active entity set;
[0017] The hierarchical structure tree is obtained according to the distances between the activity entity set and the individual activity entities using a top-down greedy algorithm.
[0018] Optionally, traversing the hierarchical structure tree according to the priority of each node to obtain the target industrial process model specifically includes:
[0019] Determining the priority of each of the nodes;
[0020] The target industrial process model is determined by traversing the hierarchical structure tree starting from the root node according to the priority of each node.
[0021] A system for determining an industrial process model, comprising:
[0022] An industrial process text document acquisition module, used to acquire the industrial process text document to be identified;
[0023] An activity entity extraction module is used to extract an activity entity set from the industrial process text document to be identified using a BERT model, a bidirectional long-short word memory model, and a conditional random field model; the activity entity set includes multiple activity entities;
[0024] A hierarchical tree generation module is used to input the activity entity set into an ordered neural network for identification to obtain a hierarchical tree; each node of the hierarchical tree corresponds to one of the activity entities;
[0025] The target industrial process model generation module is used to traverse the hierarchical structure tree according to the priority of each node to obtain the target industrial process model, where each node of the hierarchical structure tree corresponds to a business in the target industrial process model; the target industrial process model is used to describe the business process contained in the industrial process text document to be identified.
[0026] Optionally, the activity entity extraction module specifically includes:
[0027] a word vector generating unit, configured to input the industrial process text document into the BERT model for encoding to obtain a plurality of word vectors;
[0028] A word sequence generating unit, configured to input each of the word vectors into the bidirectional long-short word memory model to obtain a plurality of word sequences;
[0029] The activity entity extraction unit is used to input each of the word sequences into the conditional random field model for decoding to obtain the activity entity set.
[0030] Optionally, the hierarchical tree generation module specifically includes:
[0031] a distance calculation unit, configured to calculate the distance between each active entity in the active entity set;
[0032] The hierarchical structure tree determining unit is configured to obtain the hierarchical structure tree according to the distances between the activity entity set and the individual activity entities using a top-down greedy algorithm.
[0033] Optionally, the target industrial process model generation module specifically includes:
[0034] a priority determination unit, configured to determine the priority of each of the nodes;
[0035] The target industrial process model determining unit is configured to traverse the hierarchical structure tree starting from the root node according to the priority of each node to determine the target industrial process model.
[0036] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0037] The present invention discloses a method and system for determining an industrial process model. The method includes: obtaining an industrial process text document to be identified; extracting an activity entity set from the industrial process text document to be identified using a BERT model, a bidirectional long-short word memory model, and a conditional random field model; the activity entity set includes multiple activity entities; inputting the activity entity set into an ordered neural network for identification to obtain a hierarchical structure tree; each node of the hierarchical structure tree corresponds to an activity entity; traversing the hierarchical structure tree according to the priority of each node to obtain a target industrial process model, each node of the hierarchical structure tree corresponds to a business in the target industrial process model; the target industrial process model is used to describe the business process contained in the industrial process text document to be identified. The present invention combines the BERT model, the bidirectional long-short word memory model, and the conditional random field model to accurately identify each activity entity in the industrial process text document to be identified, thereby improving the accuracy of industrial process model determination. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0039] Figure 1 A flow chart of a method for determining an industrial process model provided by an embodiment of the present invention;
[0040] Figure 2 A block diagram of a system for determining an industrial process model provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0042] The purpose of the present invention is to provide a method and system for determining an industrial process model, aiming to improve the accuracy of industrial process model determination, and can be applied to the field of model determination technology.
[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0044] Figure 1 Flowchart of the method for determining the industrial process model provided by the embodiment of the present invention. Figure 1 As shown, the method for determining the industrial process model in this embodiment includes:
[0045] Step 101: Obtain an industrial process text document to be identified.
[0046] Step 102: Utilize the BERT model, the bidirectional long-short word memory model, and the conditional random field model to extract an activity entity set from the industrial process text document to be identified; the activity entity set includes multiple activity entities.
[0047] Step 103: Input the activity entity set into the ordered neural network for identification to obtain a hierarchical tree; each node of the hierarchical tree corresponds to an activity entity.
[0048] Step 104: Traverse the hierarchical tree according to the priority of each node to obtain the target industrial process model. Each node of the hierarchical tree corresponds to a business in the target industrial process model; the target industrial process model is used to describe the business process contained in the industrial process text document to be identified.
[0049] As an optional implementation, step 102 specifically includes:
[0050] The industrial process text document is input into the BERT model for encoding to obtain multiple word vectors.
[0051] Each word vector is input into the bidirectional long-short word memory model to obtain multiple word sequences.
[0052] Each word sequence is input into the conditional random field model for decoding to obtain the active entity set.
[0053] As an optional implementation, step 103 specifically includes:
[0054] The distance between each activity entity in the activity entity set is calculated; specifically, the distance is the Euclidean distance.
[0055] Using a top-down greedy algorithm, a hierarchical structure tree is obtained based on the activity entity set and the distance between each activity entity.
[0056] As an optional implementation, step 104 specifically includes:
[0057] Determine the priority of each node.
[0058] Starting from the root node according to the priority of each node, the hierarchical tree is traversed to determine the target industrial process model.
[0059] Figure 2 This is a block diagram of a system for determining an industrial process model provided by an embodiment of the present invention. Figure 2 As shown, the industrial process model determination system in this embodiment includes:
[0060] The industrial process text document acquisition module 201 is used to acquire the industrial process text document to be identified.
[0061] The activity entity extraction module 202 is used to extract an activity entity set from the industrial process text document to be identified using the BERT model, the bidirectional long short word memory model and the conditional random field model; the activity entity set includes multiple activity entities.
[0062] The hierarchical tree generation module 203 is used to input the activity entity set into the ordered neural network for identification to obtain a hierarchical tree; each node of the hierarchical tree corresponds to an activity entity.
[0063] The target industrial process model generation module 204 is used to traverse the hierarchical tree according to the priority of each node to obtain the target industrial process model. Each node of the hierarchical tree corresponds to a business in the target industrial process model; the target industrial process model is used to describe the business process contained in the industrial process text document to be identified.
[0064] As an optional implementation, the activity entity extraction module 202 specifically includes:
[0065] The word vector generation unit is used to input industrial process text documents into the BERT model for encoding to obtain multiple word vectors.
[0066] The word sequence generation unit is used to input each word vector into the bidirectional long-short word memory model to obtain multiple word sequences.
[0067] The activity entity extraction unit is used to input each word sequence into the conditional random field model for decoding to obtain an activity entity set.
[0068] As an optional implementation, the hierarchical tree generation module 203 specifically includes:
[0069] The distance calculation unit is used to calculate the distance between each activity entity in the activity entity set.
[0070] The hierarchical structure tree determining unit is used to obtain the hierarchical structure tree according to the activity entity set and the distance between each activity entity by using a top-down greedy algorithm.
[0071] As an optional implementation, the target industrial process model generation module 204 specifically includes:
[0072] The priority determination unit is used to determine the priority of each node.
[0073] The target industrial process model determination unit is used to traverse the hierarchical structure tree starting from the root node according to the priority of each node to determine the target industrial process model. Specific embodiments
[0075] Definition 1: Activity Entity. A sentence in an industrial process description can be represented as a sextuple Se = (SP, O, C, A, D), consisting of a subject, a predicate, an object, an object complement, an attributive, and an adverbial. The steps for defining an activity entity are as follows.
[0076] (1) S is the subject.
[0077] (2)P is the predicate.
[0078] (3) O is the object.
[0079] (4) C is an attributive.
[0080] (5)A is a complement.
[0081] (6) D is an adverbial.
[0082] (7) Se is a sentence in the text document that constitutes the industrial process.
[0083] (8) Define the active entity a = (P, O) or a = (P). Where a∈A, A is the industrial process model
[0084] P = active entity set (A, F, I; S, E).
[0085] As shown in Definition 1, a sentence Se in an industrial process text document is represented by (S, P, O, C, A, D), which contains a subject, a predicate, an object, an object complement, an attributive, and an adverbial. This paper defines an activity entity as a = (P, O) (predicate + object) or a = (P) (predicate).
[0086] After extracting labeled activity entities from industrial process text documents, this paper uses a deep learning method to unsupervisedly retrieve the latent hierarchical structure between activity entities in industrial process text documents. This paper refers to the latent hierarchical structure between activity entities as a hierarchical tree. The definition of the hierarchical tree is shown in Definition 2.
[0087] Definition 2: Hierarchical tree. A hierarchical tree can be represented as a triple q = (A, D, I), containing
[0088] Activity entities, maximum depth, nesting.
[0089] (1)q is a hierarchical structure tree.
[0090] (2) The activity entities a∈A form a hierarchical structure tree, where A is the set of activity entities of the industrial process model P=(A, F, I; S, E).
[0091] (3) D is the maximum depth of the hierarchy tree.
[0092] (4) I is the nested relationship between activity entities.
[0093] After obtaining the hierarchical structure tree, the hierarchical structure tree is automatically converted into an industrial process model. Definition 3 gives the hierarchical depth principle used in this paper to convert the hierarchical structure tree into an industrial process model.
[0094] Definition 3: Hierarchical depth principle. A hierarchical structure tree can be represented as a triple = (A, D, I),
[0095] An industrial process model is represented as a quintuple P = (A, F, I; S, E). The steps for converting the hierarchical structure tree into an industrial process model are as follows.
[0096] (1) The set of active entities of the industrial process model is exactly the same as that of the hierarchical tree.
[0097] (2) a∈A, d∈D, where d is the depth of a in the hierarchical structure tree. The smaller d is, the higher the priority of activity entity a in the industrial process model and the earlier it occurs. The larger d is, the lower the priority of activity entity a in the industrial process model and the later it occurs.
[0098] (3) Let a1, a2, a3…a n (n>1) are at the same depth in the hierarchy tree, then the activity entities a1, a2, a3…a n (n>1) In the industrial process model, the priority is the same, and the activity entities a1, a2, a3...a n (n>1) in the order of precedence.
[0099] (4) Let a1, a2, a3…a n (n>1) have the same parent node a in the hierarchy tree i , then a i Nested a1, a2, a3…a n .
[0100] After obtaining the hierarchical structure tree, the hierarchical structure tree is transformed into an industrial process model according to the hierarchical depth principle.
[0101] The definition of industrial process model is shown in Definition 4.
[0102] Definition 4: Industrial Process Model. An industrial process model describes the business processes inherent in production control software for intelligent manufacturing. An industrial process model is defined as P = (A, F, I; S, E), encompassing activities, flows, nesting, starts, and ends.
[0103] (1) (A, F; S, E) is a network structure.
[0104] (2) A is the active entity set, It is called an active entity.
[0105] (3) F is the flow relationship on the industrial process model, including from activity entity to activity entity, from start to activity entity, or from activity entity to end.
[0106] (4) I is nested, I contains at least two active entities a and
[0107] (5) S is the initial state of p.
[0108] (6) E is the final modality of p.
[0109] Industrial process models are widely used. In actual industrial production, any process that requires electronic information control has an industrial process model flow chart that needs to be strictly implemented. The industrial process model can be automatically generated from the text through the deep learning method (artificial intelligence), which can greatly shorten the time of manually reading the text to generate the process model. The traditional process mining method requires logs to generate the industrial process model, and the method for generating the industrial process model of the present invention can still be used when the log is missing and the process mining method cannot be used. After the method of the present invention improves the method accuracy and model robustness by expanding the amount of data and multi-domain data, it can detect the difference between the actual execution process of industrial equipment and the model initially defined in the design.
[0110] In the industrial sector, the production control software for most core industrial products, including product design, production equipment, product R&D, and sensing devices, is closely tied to process control technology. A significant amount of industrial software itself embodies industrial processes. For example, ASML, a renowned Dutch lithography company, now uses industrial process software (i.e., industrial processes within industrial manufacturing) to control its lithography and chip manufacturing processes. China CNR Corporation's process model library holds over 200,000 process models, and Australia's Suncorp Bank has approximately 6,000 process models. Researchers have also extracted nearly 136,000 business rules from two million COBOL lines of a legacy system for Volkswagen Finance's car rental business, built in the 1980s. In other words, in real-world industrial manufacturing, process models drive numerous application scenarios, including industrial control, product design, production R&D, and sensing devices.
[0111] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.
[0112] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the device and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.
Claims
1. A method for determining an industrial process model, characterized in that: The method comprises: Obtaining industrial process text documents to be identified; Extracting an activity entity set from the industrial process text document to be identified using a BERT model, a bidirectional long-short word memory model, and a conditional random field model; the activity entity set includes multiple activity entities; Inputting the activity entity set into an ordered neural network for identification to obtain a hierarchical tree; each node of the hierarchical tree corresponds to one of the activity entities; The hierarchical structure tree is traversed according to the priority of each node to obtain a target industrial process model, wherein each node of the hierarchical structure tree corresponds to a business in the target industrial process model; the target industrial process model is used to describe the business process contained in the industrial process text document to be identified.
2. The method for determining an industrial process model according to claim 1, characterized in that: The method of extracting an activity entity set from the industrial process text document to be identified by using the BERT model, the bidirectional long-short word memory model, and the conditional random field model specifically includes: Inputting the industrial process text document into the BERT model for encoding to obtain multiple word vectors; Inputting each of the word vectors into the bidirectional long-short word memory model to obtain multiple word sequences; Each of the word sequences is input into the conditional random field model for decoding to obtain the activity entity set.
3. The method for determining an industrial process model according to claim 1, wherein: The step of inputting the active entity set into an ordered neural network for identification to obtain a hierarchical structure tree specifically includes: Calculating the distance between each active entity in the active entity set; The hierarchical structure tree is obtained according to the distances between the activity entity set and the individual activity entities using a top-down greedy algorithm.
4. The method for determining an industrial process model according to claim 1, wherein: Traversing the hierarchical structure tree according to the priority of each node to obtain the target industrial process model specifically includes: Determining the priority of each of the nodes; The target industrial process model is determined by traversing the hierarchical structure tree starting from the root node according to the priority of each node.
5. A system for determining an industrial process model, characterized in that: include: An industrial process text document acquisition module, used to acquire the industrial process text document to be identified; An activity entity extraction module is used to extract an activity entity set from the industrial process text document to be identified using a BERT model, a bidirectional long-short word memory model, and a conditional random field model; the activity entity set includes multiple activity entities; A hierarchical tree generation module is used to input the activity entity set into an ordered neural network for identification to obtain a hierarchical tree; each node of the hierarchical tree corresponds to one of the activity entities; The target industrial process model generation module is used to traverse the hierarchical structure tree according to the priority of each node to obtain the target industrial process model, where each node of the hierarchical structure tree corresponds to a business in the target industrial process model; the target industrial process model is used to describe the business process contained in the industrial process text document to be identified.
6. The industrial process model determination system according to claim 5, characterized in that: The activity entity extraction module specifically includes: a word vector generating unit, configured to input the industrial process text document into the BERT model for encoding to obtain a plurality of word vectors; A word sequence generating unit, configured to input each of the word vectors into the bidirectional long-short word memory model to obtain a plurality of word sequences; The activity entity extraction unit is used to input each of the word sequences into the conditional random field model for decoding to obtain the activity entity set.
7. The industrial process model determination system according to claim 5, characterized in that: The hierarchical tree generation module specifically includes: a distance calculation unit, configured to calculate the distance between each active entity in the active entity set; The hierarchical structure tree determining unit is configured to obtain the hierarchical structure tree according to the distances between the activity entity set and the individual activity entities using a top-down greedy algorithm.
8. The industrial process model determination system according to claim 5, characterized in that: The target industrial process model generation module specifically includes: a priority determination unit, configured to determine the priority of each of the nodes; The target industrial process model determining unit is configured to traverse the hierarchical structure tree starting from the root node according to the priority of each node to determine the target industrial process model.