Ethylene plant anomaly handling system based on knowledge graph, and device
The knowledge graph-based anomaly handling system for ethylene plants utilizes embedded learning and inference prediction modules to quickly and accurately detect faults and provide solutions, thus solving the problem of low efficiency in anomaly handling for ethylene plants and improving both processing efficiency and safety.
Patent Information
- Application Number
- PCT/CN2025/125627
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-15
- Filing Date
- 2025-09-30
- Publication Date
- 2026-11-19
AI Technical Summary
The operation manuals for ethylene plants are updated slowly, and traditional methods of handling anomalies rely on manual research or consultation with experts, resulting in low efficiency, poor accuracy, and a high error rate in handling anomalies.
An anomaly handling system for an ethylene plant based on a knowledge graph is adopted. The system performs representation learning on the knowledge graph through an embedded learning module and various embedding strategies, combines it with an inference and prediction module to perform anomaly inference, provides anomaly prediction results and solutions, and realizes intelligent question answering function through an interactive question answering module.
It enables rapid and accurate detection of ethylene unit faults, locates the source of anomalies, and provides efficient fault solutions, thereby improving the efficiency and safety of ethylene unit anomaly handling and enhancing the level of intelligent operation.
Smart Images

Figure CN2025125627_19112026_PF_FP_ABST
Abstract
Description
A knowledge graph-based anomaly handling system and equipment for an ethylene plant Technical Field
[0001] This application relates to the field of chemical production, and in particular to an anomaly handling system and equipment for an ethylene plant based on a knowledge graph. Background Technology
[0002] In modern industrial production, ethylene, as an important chemical raw material, is widely used in plastics, synthetic fibers, rubber, and other fields. As a crucial piece of equipment in the chemical industry, ethylene plants have become increasingly complex in their operating procedures and related knowledge due to the continuous development of ethylene production processes.
[0003] However, the operation manuals and related operating procedures of ethylene plants are usually updated slowly, and traditional methods of handling ethylene plant anomalies often rely on manual research or consultation with experts. This is not only inefficient but also easily limited by experience, resulting in a high error rate in handling anomalies and seriously reducing the efficiency and accuracy of handling anomalies in ethylene plants. Summary of the Invention
[0004] This invention provides a knowledge graph-based anomaly handling system and equipment for ethylene plants, which improves the efficiency and accuracy of anomaly handling in ethylene plants, thereby improving ethylene production efficiency and safety.
[0005] In a first aspect, this application provides an anomaly handling system for an ethylene plant based on a knowledge graph, the system comprising:
[0006] An embedding learning module is used to perform representation learning on the knowledge graph based on a variety of preset embedding strategies to obtain the vector representation corresponding to the knowledge graph; the knowledge graph contains multiple entities of the ethylene plant and the relationships between the entities, and each entity corresponds to the equipment, sensors and control logic of the ethylene plant respectively; the vector representation represents the feature vector of each entity and relationship in the knowledge graph.
[0007] The reasoning and prediction module is used to perform anomaly reasoning based on the vector representation to obtain anomaly prediction results for the ethylene unit and corresponding solutions.
[0008] Optionally, the system further includes an interactive question-and-answer module, which, in response to an input question about the ethylene plant from the target object, outputs corresponding question-and-answer results based on the knowledge graph.
[0009] Optionally, the system further includes a knowledge graph construction module for acquiring equipment data of the ethylene plant and constructing the knowledge graph based on the equipment data.
[0010] Optionally, the map construction module is specifically used for:
[0011] Based on the ethylene plant data, an ontology file corresponding to the knowledge graph is determined. The ontology file includes custom classes, relational attributes, and data attributes. The classes represent the physical entities, operational logic, and abnormal phenomena involved in the operation of the ethylene plant. The relational attributes represent the relationships between the classes. The data attributes represent the attribute information corresponding to each class.
[0012] The unstructured data in the ethylene plant data is transformed into structured data to obtain the triple data of the knowledge graph, and the knowledge graph is then visualized.
[0013] Optionally, the various embedding strategies include geometric model embedding strategies, language model embedding strategies, and graph neural network embedding strategies; wherein,
[0014] The geometric model embedding strategy represents the triplet data of the knowledge graph through a geometric model.
[0015] The language model embedding strategy is characterized by embedding text information in the ethylene unit data based on natural language processing technology and combined with a pre-trained language model to obtain the corresponding semantic representation.
[0016] The graph neural network embedding strategy is characterized by: processing the entities and relationships of the knowledge graph based on the graph neural network to determine the high-order relationships between entities in the knowledge graph; and generating vector representations of each entity through a graph convolutional network.
[0017] Optionally, the inference prediction module is specifically used for:
[0018] Based on a preset anomaly detection training set and the vector representation, anomaly reasoning is performed on the ethylene unit; the anomaly detection training set includes ethylene unit data labeled with normal and abnormal states.
[0019] When an abnormal situation occurs in the ethylene unit, the abnormality prediction result and the solution are output based on the knowledge graph.
[0020] Optionally, the inference prediction module is further configured to:
[0021] When the knowledge graph does not include the solution, a solution corresponding to the anomaly prediction result is generated based on a preset decision algorithm;
[0022] Based on the current operating status of the ethylene plant, a personalized recommendation for the solution is made.
[0023] Optionally, the ethylene unit data includes at least equipment information, sensor data, process parameters, and control loop information of the ethylene unit.
[0024] Optionally, the interactive question-and-answer module is specifically used for:
[0025] Keyword extraction was performed on the problems of the ethylene unit to determine the corresponding anomaly description information;
[0026] Based on the knowledge graph and the reasoning and prediction module, the question-and-answer results corresponding to the anomaly description information are obtained and fed back to the target object; the question-and-answer results include the corresponding anomaly prediction results and solutions.
[0027] Optionally, the system further includes a visualization module for displaying the question-and-answer results and the equipment status of the ethylene plant to the target object based on a visual interface.
[0028] Secondly, this application provides a method for constructing an anomaly handling system for an ethylene plant based on a knowledge graph, the method comprising:
[0029] Obtain the requirements analysis information of the ethylene unit's abnormal handling system, including user requirements information and system requirements information;
[0030] Based on the aforementioned requirements analysis information, a system development environment is constructed, and based on the system development environment, the system architecture of the ethylene plant anomaly handling system is constructed; the system architecture includes a data layer, an application service layer, and a user layer;
[0031] Based on a pre-defined modular development strategy, the functional modules of the ethylene unit's anomaly handling system are constructed sequentially.
[0032] Thirdly, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory is used to store a knowledge graph of an ethylene plant; and the processor executes the computer program to implement the steps of any of the ethylene plant anomaly handling systems described in the first aspect above.
[0033] The beneficial effects of this invention are as follows:
[0034] This application provides an anomaly handling system and device for an ethylene plant based on a knowledge graph. The system performs representation learning on the knowledge graph through an embedding learning module and various preset embedding strategies to obtain the vector representation corresponding to the knowledge graph. The inference and prediction module performs anomaly inference based on the vector representation to obtain the anomaly prediction result and corresponding solution for the ethylene plant. By combining knowledge graph, embedding learning, inference and prediction algorithms, the system associates information such as equipment, faults, and process parameters of the ethylene plant in a structured manner, thereby achieving rapid and accurate detection of equipment faults in the ethylene plant, locating the source of the anomaly, and providing efficient fault solutions. Attached Figure Description
[0035] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0036] Figure 1 is a schematic diagram of the structure of an ethylene plant anomaly handling system based on knowledge graph provided in an embodiment of this application;
[0037] Figure 2 is a schematic diagram of the processing flow of a knowledge graph construction module provided in this application;
[0038] Figure 3 is a schematic diagram of an ontology file provided in an embodiment of this application;
[0039] Figures 4(a) and 4(b) are schematic diagrams of an entity of a knowledge graph provided in an embodiment of this application;
[0040] Figure 5 is a schematic diagram of a partial representation of a knowledge graph provided in an embodiment of this application;
[0041] Figure 6 is a schematic diagram of the processing flow of an embedded learning module provided in an embodiment of this application;
[0042] Figure 7 is a schematic diagram of the architecture of an interactive question-and-answer module provided in an embodiment of this application;
[0043] Figure 8 is a schematic diagram of an abnormal reasoning process provided in an embodiment of this application;
[0044] Figure 9 is a schematic diagram of the system architecture of an ethylene plant anomaly handling system provided in an embodiment of this application;
[0045] Figure 10 is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0046] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.
[0047] The terms "first" and "second" in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising" and any variations thereof are intended to cover non-exclusive protection. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. The term "multiple" in this application can mean at least two, for example, two, three, or more, and this application does not impose limitations.
[0048] The term "and / or" in the embodiments of this application is merely a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.
[0049] It is understood that the following specific embodiments of this application involve data related to the chemical production process. When the various embodiments of this application are applied to specific products or technologies, relevant licenses or consents are required, and the collection, use, and processing of the relevant data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, relevant volunteers can be recruited and agreements can be signed to authorize their data, thereby enabling the implementation using the data of these volunteers; alternatively, implementation can be carried out within an authorized organization, using data from members of the organization to implement the following implementation methods for data management; or, in specific implementations, the relevant data used are all simulated data, such as simulated data generated in a virtual scenario.
[0050] The design concept of the embodiments of this application will be briefly introduced below.
[0051] In modern industrial production, ethylene, as an important chemical raw material, is widely used in plastics, synthetic fibers, rubber, and other fields. As a crucial piece of equipment in the chemical industry, ethylene plants have become increasingly complex in their operating procedures and related knowledge due to the continuous development of ethylene production processes.
[0052] However, the operation manuals and related operating procedures of ethylene plants are usually updated slowly, and traditional methods of handling ethylene plant anomalies often rely on manual research or consultation with experts, which is not only inefficient but also easily limited by experience, resulting in a high error rate in handling anomalies and seriously reducing the efficiency of handling anomalies in ethylene plants.
[0053] In view of the above problems, this application provides an anomaly handling system for ethylene plants based on knowledge graphs. This system performs representation learning on the knowledge graph through an embedding learning module and multiple preset embedding strategies to obtain the vector representation corresponding to the knowledge graph. The inference and prediction module performs anomaly inference based on the vector representation to obtain the anomaly prediction results and corresponding solutions for the ethylene plant. By combining knowledge graphs, embedding learning, inference and prediction algorithms, the system associates information such as equipment, faults, and process parameters of the ethylene plant in a structured manner, thereby achieving rapid and accurate detection of equipment faults in the ethylene plant, locating the source of the anomaly, and providing efficient fault solutions.
[0054] Furthermore, this system also includes an interactive question-and-answer module and a visualization module. Through natural language processing technology, it realizes intelligent question-and-answer functions, thereby improving the level of intelligence in the operation of the ethylene plant.
[0055] The system provided by exemplary embodiments of this application will now be described with reference to the accompanying drawings. It should be noted that the above application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way.
[0056] Referring to Figure 1, which is a schematic diagram of a knowledge graph-based anomaly handling system for an ethylene plant according to an embodiment of this application, the system includes the following unit modules:
[0057] The embedding learning module is used to perform representation learning on the knowledge graph based on a variety of preset embedding strategies, and obtain the vector representations of each entity and relationship in the knowledge graph.
[0058] The reasoning and prediction module is used to perform knowledge reasoning based on vector representation to determine the anomaly prediction results and corresponding solutions for the ethylene unit.
[0059] The interactive question-and-answer module is used to respond to the question input operation of the target object, obtain the abnormal questions input by the target object for the ethylene unit, and output the corresponding question-and-answer results based on the knowledge graph.
[0060] In one possible implementation, the system may further include a graph construction module for acquiring equipment data of the ethylene plant and constructing a knowledge graph of the ethylene plant based on the equipment data.
[0061] Specifically, the knowledge graph construction module can collect ethylene plant data, including equipment information, sensor data, process parameters, and control loop information, to construct a knowledge graph containing entities such as equipment, sensors, and control logic. This knowledge graph is then stored and managed through a graph database, managing these entities and their relationships, such as causal relationships of equipment interlocking actions, topological relationships of material flow, and sensor-control loop mapping relationships. The knowledge graph also defines equipment operating status, sensor reading ranges, and anomaly criteria, providing foundational data support for subsequent reasoning and question-answering modules. Thus, through ontology construction, classes, relational attributes, and data attributes can be defined based on the fundamental concepts of the ethylene plant regulations, providing semantic-level definitions and specifications for nodes and edges (i.e., entities and relationships) in the knowledge graph.
[0062] In one possible implementation, referring to Figure 2, which is a schematic diagram of the processing flow of a knowledge graph construction module provided in this application, the graph construction module can construct an ontology file corresponding to the knowledge graph using ethylene plant data. This ontology file includes custom classes, relational attributes, and data attributes. Specifically, classes represent the physical entities, operational logic, and abnormal phenomena involved in the operation of the ethylene plant; relational attributes represent the relationships between different classes, i.e., the relationships between two entities in the same knowledge representation, reflecting the abnormal propagation path and processing logic in the ethylene plant; and data attributes represent the attribute information corresponding to each class, i.e., the attribute values and / or state description information of an entity data in the same knowledge representation.
[0063] Specifically, the categories of the body documents in this application may include process parameters, process streams, equipment, and instruments of the ethylene plant. For example, this application may categorize the categories into equipment, contaminants, mixtures, process parameters, process streams, operating units, abnormal phenomena, actuators, controllers, control loops, and interlocking action results based on process specifications and operating guidelines. The relational attributes of the ethylene plant may include interlocking actions and their causes, stream flow directions describing the stream topology, and merging relationships describing equipment combinations, such as interlocking actions, originating from, including unit equipment, merging, flow direction, diversion, and contaminants. Data attributes may include equipment tag numbers, sensor range values, and operating pressure units for process parameters, such as temperature, tag number, pressure, inner diameter, name, specifications and set pressure, description, alarm values, values, blind flange positions, operating temperature, operating pressure, operating specifications, and phase state.
[0064] Specifically, Figure 3 is a schematic diagram of an ontology file provided in this application. Each node in Figure 3 covers the equipment entities (actuators, controllers, etc.) of the ethylene plant, process parameters (NH3, SO2, NOx, etc.), control loops (layered control loops, three-impulse control, etc.), abnormal phenomena (runaway phenomena, external phenomena, etc.), correlation relationships (material flow direction, merging, etc.) and data attributes, etc.
[0065] Next, the graph construction module in this application can perform structured transformation on the unstructured data in the ethylene plant data to obtain the triple data corresponding to the knowledge graph, realize knowledge extraction, and store the knowledge graph in different forms. Based on the above knowledge extraction, the knowledge graph is visualized using the Neo4j graph database.
[0066] Specifically, as shown in Figure 2 above, for the chart data in ethylene plant data such as ethylene operating procedures, this application can extract knowledge using a self-developed ethylene operating procedure extraction program and knowledge graph extraction program written in Python, and store it in the format of Resource Description Framework (RDF) to conform to the knowledge ontology specification. For unstructured text statements, this application can use the Begin-Inside-Outside (BIO) annotation method and use Doccano software to obtain the corresponding triple data.
[0067] In one possible implementation, based on the constructed knowledge ontology, this application can extract data, primarily from engineering and technical documents, including the "Technical Specifications for Ethylene Plants" and the "Process Flow Diagram (PFD) for Ethylene Plants." The "Technical Specifications for Ethylene Plants" contains basic data such as raw material and chemical raw material quality indicators, equipment lists, interlock actions and their causes, and control parameter alarm values, mostly presented in tabular form. The data in the "PFD for Ethylene Plants" mainly connects the relationships between equipment based on the flow of materials, presented graphically. Table 1 below shows the boundary blind plate table data in tabular form. For this type of tabular data, this application can use relevant Python table documentation libraries (such as Pandas) for extraction. Graphical data is extracted manually and saved in RDF format.
[0068] Table 1 Boundary Blind Plate Table
[0069] Furthermore, referring to Figures 4(a) and 4(b), which are entity diagrams of a knowledge graph provided in this application embodiment, Figures 4(a) and 4(b) respectively show entities of blind flange type and pipe type, which are visualized by Neo4j in this application. Referring to Figure 5, which is a schematic diagram of a partial representation of a knowledge graph provided in this application, burgundy represents "system unit class" and blue represents "instrument class", and the lines connecting the two represent the association relationship. Figure 5 also shows what measuring instruments are in the "cracking gas compressor system".
[0070] In one possible implementation, the embedding learning module of this application performs representation learning on the knowledge graph through various embedding strategies, thereby extracting the semantic information of entities and relationships in the graph and representing them in vector form. The various embedding strategies of this application may include geometric model embedding strategies, language model embedding strategies, and graph neural network embedding strategies. Specifically, the geometric model embedding strategy represents representing the triplet data of the knowledge graph using a geometric model; the language model embedding strategy represents embedding textual information in the ethylene plant data based on natural language processing technology and a pre-trained language model to obtain the corresponding semantic representation; and the graph neural network embedding strategy represents processing the entities and relationships of the knowledge graph using a graph neural network to determine the high-order relationships between entities in the knowledge graph and generating vector representations of each entity through a graph convolutional network.
[0071] Specifically, the geometric model embedding strategy in this application can embed the relationships between nodes and edges through spatial geometry methods to handle more complex spatial relationships between entities. The language model embedding strategy in this application can combine natural language processing technology, utilizing a pre-trained language model with Bidirectional Encoder Representations from Transformers (BERT) to embed textual information such as process parameters and equipment descriptions appearing in ethylene plant data, obtaining more semantic representations. This helps the system understand complex relationships and semantics in the knowledge graph, providing high-quality input data for subsequent reasoning and prediction. The graph neural network (GNN) embedding strategy in this application can process nodes and edges (i.e., entities and relationships) in the knowledge graph through graph neural networks to learn higher-order relationships between entities in the graph, and generate a vector representation of each node (entity) through a graph convolutional network (GCN). In summary, the embedded learning module in this application can transform entities such as equipment, sensors, and process parameters in an ethylene plant and their interrelationships into low-dimensional vector representations through embedded learning techniques such as geometric models, language models, and graph neural networks. This allows the system to capture the complex relationships and dependencies between devices. Through multi-dimensional embedded learning, the system can more accurately model the normal and abnormal states of the equipment.
[0072] In one possible implementation, the reasoning and prediction module of this application can perform anomaly reasoning on the ethylene unit using a preset anomaly detection training set and vector representation, wherein the anomaly detection training set includes ethylene unit data labeled with normal and abnormal states. When an anomaly occurs in the ethylene unit, anomaly prediction results and solutions are output through a knowledge graph.
[0073] Specifically, the inference and prediction module in this application can construct an anomaly detection training set using ethylene plant data such as historical equipment operation data, sensor data, and fault records, and label normal and abnormal states. Next, using an embedding learning module, the module is trained on the equipment operation data to learn the differences between normal and abnormal states. Combining the embedding results of graph neural networks, geometric models, and language models, it infers possible anomalies. For example, if a sensor reading deviates significantly from historical data, it infers a possible equipment failure. Furthermore, by analyzing the equipment relationships in the inference graph, the source of the anomaly is quickly located. For instance, when the pressure of a valve is abnormal, the graph structure can be used to infer whether the equipment connected to that valve is affected. After obtaining the inferred anomaly prediction result, the module queries the knowledge graph for relevant repair measures and solutions related to equipment failures and outputs corresponding solutions.
[0074] In one possible implementation, the reasoning prediction module in this application is further used to: generate a solution corresponding to the abnormal prediction result through a preset decision algorithm when the knowledge graph does not include a solution, and make personalized recommendations for the solution based on the current operating status of the ethylene plant.
[0075] Specifically, if the knowledge graph lacks relevant solutions for a particular field source, this application can also use a decision algorithm to generate corresponding solutions and make personalized recommendations based on the current operating status of the ethylene plant. For example, if the system detects a fault in device A, it recommends replacement or adjustment.
[0076] In one possible implementation, referring to Figure 6, which is a schematic diagram of the processing flow of an embedding learning module provided in an embodiment of this application, after data preprocessing and knowledge graph construction, this application can select three embedding methods: distance-based Translating Embeddings (TransE) model, language-based Knowledge Graph-BERT (KG-Bert) model, and graph neural network model. The three models are trained using triplet data to generate embedding vectors. Based on the node representations generated by the graph neural network, inference is performed according to the user-submitted query request, returning relevant entities and relationships. Among these, graphs are one of the most expressive data structures and have been used to model various problems. To systematically utilize the entity-relationship combination operation in knowledge graph embedding technology, this application can achieve better embedding performance by jointly learning the vector representations of nodes and relationships in the graph.
[0077] Specifically, this application can learn a low-dimensional embedding vector for each node and edge in the graph. The node embedding represents the feature information of the node, and the edge embedding represents the relational information represented by the edge. These embedding vectors are randomly initialized and updated during training based on the optimization objective.
[0078] For example, this application may define a multi-relationship graph G = (V, R, E, X, Z).
[0079] Where V represents the node set, R represents the relation set, and E represents the edge set;
[0080] This represents the d0-dimensional input features of each node;
[0081] This indicates the initial relational characteristics.
[0082] In traditional knowledge graphs, edges are directional; for example, (u,r,v) represents entity u pointing to entity v through relation r. This application leverages the symmetry of information propagation to explicitly add a reverse edge (v,r) to each edge (u,r,v). -1 ,u), and expand the relation set to R′=R∪{r -1}
[0083] For example, in the dataset of abnormal operation of ethylene plants, there exists a pair of inverse relationships: "caused by" and "result in". By introducing reverse edges, the neighborhood N(v) of node v includes not only the nodes pointing to it (such as u→v), but also the nodes it points to (such as the reverse edge w→v of v→w), thus capturing graph structure information more comprehensively.
[0084] Furthermore, in order to incorporate relation embedding into the GCN formula, this application can utilize the entity-relation composition operation used in knowledge graph embedding methods, which takes the form of:
[0085] e0=φ(e s ,e r )
[0086] Wherein, φ:R d ×R d →R d It is a combination operator, where s, r, and o represent the subject, relation, and object in the knowledge graph, and e... (.) ∈R d This represents their respective embeddings. For each edge (u,v,r)∈E, there is a reverse edge (v,u,r). -1 The representation obtained after k layers of directed GCN is:
[0087] Here, set N(v) is defined as the set of direct neighbors of node v connected by its outgoing edges, and W... r Specific parameters for the model's relationships are defined. To avoid the over-parameterization problem of traditional methods, this application also introduces a combination function φ to fuse neighbor nodes u with their corresponding relationships r, enabling the model to perceive relational semantics while maintaining linear complexity (O(|R|d)) in the feature dimension. Furthermore, to distinguish between original edges, reverse edges, and self-looping edges, independent filters can be defined for each edge type, with the update equation as follows:
[0088] Where, x u ,z r Let h be the initial features of node u and relation r, respectively. v This represents the update representation for node v. These are relation type-specific parameters. In this application, the GCN model uses direction-specific weights, i.e., λ(r) = dir(r), as shown below:
[0089] Furthermore, after the node embedding update, the relation embedding in this application is also transformed as follows: h r =W rel z r
[0090] in, It is a learnable transformation matrix that can project all relations as nodes into the same embedding space and allow them to be used in the next layer of GCN.
[0091] In one possible implementation, the interactive question-and-answer module in this application can extract keywords from ethylene plant issues to determine the corresponding abnormal description information, and use a knowledge graph and reasoning prediction module to obtain the abnormal prediction results and solutions corresponding to the abnormal description information, and then feed the question-and-answer results back to the target object. In this way, the interactive question-and-answer module can efficiently handle users' abnormal questions about ethylene plants and provide accurate answers based on the reasoning results of the knowledge graph.
[0092] Specifically, the interactive question-and-answer module in this application performs semantic analysis on questions posed by users or other target objects, and uses regular expressions and keywords to extract and identify key entities (such as "temperature sensor") and operational behaviors (such as "fault handling"). Based on the identified anomaly descriptions, relevant information or inference results are determined from the knowledge graph through knowledge graph query algorithms (such as graph search, pathfinding, etc.). In this way, combining the information in the knowledge graph with the inference results yields a more accurate answer. For example, if the question involves "sensor fault," the knowledge graph is queried and combined with the inference module to generate possible fault analyses and solutions, which are then returned to the user through the question-and-answer module.
[0093] In one possible implementation, referring to Figure 7, which is a schematic diagram of the architecture of an interactive question-answering module provided in an embodiment of this application, the interactive question-answering module consists of a user input layer, a data preprocessing layer, a knowledge graph inference layer, and an output result layer from bottom to top. In the user input layer, users can interact with the system through the interface, inputting fault descriptions, equipment information, or process parameters for ethylene plant issues, and initiating query requests to the system. The input method can be free text or historical input records. In the data preprocessing layer, the interactive question-answering module is mainly used to clean, format, and structure the raw data input by the user, and to identify the user's query intent (e.g., querying equipment status, requesting fault diagnosis, retrieving historical records, etc.) through Natural Language Processing (NLP) technology, providing structured data for subsequent inference. The knowledge graph inference layer encapsulates the KG-BERT and GCN trained models into two backend interfaces based on Flask, so that after the user inputs content, the module will call the pre-trained model interface of the corresponding inference method in the backend to infer the user's input content. Finally, the output layer is responsible for presenting the question-and-answer results returned by the reasoning layer to the user through a graphical interface or text. For example, the output content will visualize the most likely answers in the form of a knowledge graph.
[0094] Specifically, referring to Figure 8, which is a schematic diagram of an abnormal reasoning process provided in an embodiment of this application, by selecting the reasoning type as KG-BERT and inputting the abnormal phenomenon "partial ice blockage, causing low-load production of the device" and the reasoning content "the cause is", after clicking reasoning, 10 sets of answers will appear after a while, and the answers will be sorted according to the score.
[0095] For ease of description, the above sections are divided into functional units (or modules) and described separately. Of course, in implementing this application, the functions of each unit (or module) can be implemented in one or more software or hardware components. Those skilled in the art will understand that various aspects of this application can be implemented as systems, methods, or program products. Therefore, various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, collectively referred to herein as "circuit," "module," or "system."
[0096] In one possible implementation, this application also provides a method for constructing an ethylene plant anomaly handling system, the specific implementation steps of which are as follows:
[0097] Step 901: Obtain the demand analysis information of the ethylene unit anomaly handling system, the demand analysis information including user demand information and system demand information.
[0098] In this embodiment of the application, the user requirements information of the ethylene plant anomaly handling system may include: the ability to raise questions related to the abnormal operation of the ethylene plant based on abnormal phenomena and obtain corresponding reasoning results; the system requirements information may include: the system's processing speed, stability, reliability, and scalability, etc.
[0099] Specifically, regarding user needs, users expect the system to understand the relationship between device malfunctions and solutions, quickly return reliable query results, and provide an intuitive interface for interaction and result display. The user-side goal is to implement three functions: query input, selection of inference algorithms, and visualization of inference results. Due to the vast amount of knowledge graph data, users require selectable and searchable entities and relationships, and the ability to visually construct inference statements. Similarly, users can independently select the inference algorithm on the page. Visualizing inference results includes graph visualization and a table display of the Top K most reliable inference entities. To meet system requirements, this system needs to improve the speed of geographic logic queries by employing pre-trained models and calling the inference interface. In terms of stability, it supports input validation and data verification, ensuring the reliability and validity of user input through front-end and back-end rule settings. Regarding reliability, it possesses high availability and accessibility, and can run stably on multiple devices. In terms of scalability, the system needs to efficiently manage and maintain the massive knowledge graph database and possess scalability to support future knowledge updates and feature enhancements.
[0100] Step 902: Based on the requirements analysis information, build the system development environment, and based on the system development environment, build the system architecture of the ethylene plant anomaly handling system.
[0101] In this embodiment of the application, the system architecture of the ethylene plant anomaly handling system includes a data layer, an application service layer, and a user layer.
[0102] In one possible implementation, this application can create a system development environment based on a browser-server (BS) architecture, including front-end interface design, back-end logic processing, and data storage and management.
[0103] Specifically, the front-end interface design in this application can utilize the Vue2 progressive front-end development framework to build the user interface. This framework is not only ready to use and easy to develop, but also provides a rich interactive experience. Furthermore, it utilizes desktop component libraries based on Vue 2.0, such as Element, to provide a wealth of UI components, enabling the rapid construction of aesthetically pleasing and minimalist front-end pages. D3.js, as a powerful JavaScript library, can be used to create dynamic and interactive data visualizations on web pages. After obtaining data from Neo4j, this application can use D3.js to render graphs and add interactions and styles to nodes and relationships, enabling the front-end page to display a complete knowledge graph. It is worth noting that the aforementioned software tools are merely illustrative examples; this application can select other suitable software tools according to actual needs, and this application does not impose any specific limitations on them.
[0104] Specifically, the backend logic processing of this application can be performed using Python. Python has extensive library support in the fields of data processing and machine learning, and is one of the most widely used languages in the field of artificial intelligence. Python can be used to build and deploy pre-trained deep learning models, and to parse and process geographic knowledge graph query and question answering tasks. The data storage and management of this application can utilize the Neo4j graph database. Neo4j, as the most popular high-performance NoSQL database, is used to store structured data on the network and employs the Cypher language for efficient querying.
[0105] Specifically, referring to Figure 9, which is a schematic diagram of the system architecture of an ethylene plant anomaly handling system provided in this application embodiment, the data layer is used for data storage and response of the ethylene plant anomaly handling system. It uses the Neo4j graph database to store the knowledge graph data of the ethylene plant, supporting basic knowledge graph queries. Simultaneously, it employs distance-based, language-based, and graph neural network-based algorithms to support plant anomaly reasoning and query functions. The application service layer is used to parse the query statements returned from the front end and implement plant operation anomaly logic queries. For general knowledge graph queries input by the user, after obtaining the query statement on the front end page, it is parsed into a Cypher query statement, and the query content required by the user is retrieved from the Neo4j database. For operation anomaly queries input by the user, after obtaining the query statement on the front end page, the operation anomaly logic query statement is parsed, and pre-trained models such as GCN, KG-Bert, and TransE are used to reason about the user's query request, generating reasoning results and returning them to the user. The user layer is a web-based front end interface built using the Vue2 front end framework, used for user input of reasoning query statements and displaying graph visualization and attribute tables of reasoning results.
[0106] Step 903: Based on the preset modular development strategy, construct each functional module of the ethylene unit anomaly handling system in sequence.
[0107] In this embodiment of the application, the functional design of the ethylene plant anomaly handling system needs to revolve around core requirements, including algorithm selection, query parsing, knowledge graph matching, logical query, and result display. It mainly includes parsing the user-input query by setting specified input rules and commands, converting it into corresponding reasoning statements and query statements, searching for relevant information in the knowledge graph through the knowledge graph query engine, executing logical queries in the knowledge graph through reasoning algorithms, and displaying query results and reasoning paths through the user interface.
[0108] In one possible implementation, this application may adopt a modular development strategy, sequentially designing the front-end page design and the back-end inference algorithm. This allows for the sequential construction of the various functional modules of the ethylene plant anomaly handling system.
[0109] Specifically, firstly, regarding the user interface, the system ensures the interactivity of the query process and the intuitiveness of the results. The server receives inference commands from the front-end user, parses and converts them into corresponding queries or geographic inference tasks, calls the inference algorithm to generate abnormal inference results and the Top K most reliable entities, and displays the graph and attributes through the front-end. When a user accesses the front-end page of the geographic knowledge graph inference system, the system loads relevant CSS stylesheets, JavaScript scripts, and graphics visualization libraries. The front-end page provides interactive functions such as user input and inference algorithm selection. The user interface mainly includes two areas: an input area and an inference result display area.
[0110] Specifically, users can input the triple information (head entity, relation, tail entity) they want to infer through input boxes on the front-end interface. Users can also select different inference algorithms and whether to enable feature aggregation. The inference algorithm selection box includes GCN, KG-Bert, and TransE; users select the corresponding inference algorithm via dropdown menus. The triple generation area consists of three search selection boxes: head entity search, relation edge search, and tail entity search. Due to the large number of entities in the knowledge graph, the system also supports users manually selecting entities via dropdown menus and searching for entities via search boxes. After selecting the head entity, relation edge, and tail entity, clicking "Generate Edge" allows users to preview the generated inference triple. Clicking "Generate Inference Path" adds the generated triple to the inference content input box, and multiple additions are allowed. The inference content input box supports manual addition and modification of rules. After adding the content to be inferred, clicking "Start Inference" transmits the inference content to the backend, which returns the inference results displayed in the inference results display area.
[0111] Specifically, the reasoning results display area comprises two sections: the upper section displays a visualized knowledge graph and reasoning path, while the lower section displays the attribute table of the reasoning results. The reasoning results returned from the backend are sent to the frontend via HTTP response. The D3.js library is used to visualize the query results, displaying the geographic logical query path of the knowledge graph and listing the attribute values of the relevant query results in the attribute table. Furthermore, this system supports users adjusting their input to perform multiple queries using different algorithms.
[0112] In one possible implementation, during the design of the backend inference algorithm of this application, after the user clicks the "Start Inference" button, the frontend can listen for the button click event via JavaScript. The user's input data is acquired and sent to the application layer on the server side via an Ajax request. The application layer on the server side receives the inference query and the user's settings options from the frontend. The application layer parses the data containing the user input into specific query commands, divided into basic queries of the knowledge graph and geo-logical queries. For basic queries of the knowledge graph such as single triples, the system generates Cypher statements. The application layer sends the query commands to the data layer, and the data layer uses the Neo4j graph database to execute these commands, retrieve data, and generate query results. For inference of knowledge graphs such as multiple triples, the corresponding pre-trained model is called for logical query processing. The data layer returns the inference results to the application layer, and the application layer organizes and formats the results.
[0113] In one possible implementation, the ethylene plant anomaly handling system of this application further includes a general question-and-answer module and an operational anomaly question-and-answer module. After receiving a user's query request, the server can parse the query statement into a Cypher language form for general query-and-answer and a reasoning language form for operational anomaly query-and-answer, depending on the query rules. General query-and-answer refers to questions and answers that do not involve operational anomaly logic reasoning in the knowledge graph query-and-answer task, but only perform simple retrieval. General query-and-answer uses Neo4j as the database and uses Cypher commands to retrieve the corresponding answers from the database for output. The operational anomaly question-and-answer module allows users to perform reasoning using a pre-trained model. When a user needs to perform a geographic logic query, they can select the corresponding algorithm, generate a reasoning path through reasoning triples, and after the server obtains the reasoning requirement, it calls the pre-trained model to generate the reasoning results and attribute table.
[0114] Please refer to Figure 10. Based on the same technical concept, this application also provides a computer device 100. In one embodiment, the computer device can be a device dedicated to the processing of an ethylene plant system, or it can be a control device for the overall chemical production. As shown in Figure 10, the computer device includes a memory 1001 and one or more processors 1002.
[0115] The memory 1001 is used to store a knowledge graph of the ethylene plant and computer programs executed by the processor 1002. The memory 1001 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and programs required to run instant messaging functions, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.
[0116] Memory 1001 may be volatile memory, such as random-access memory (RAM); memory 1001 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 1001 may be any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 1001 may be a combination of the above-described memories.
[0117] The processor 1002 may include one or more central processing units (CPUs) or digital processing units, etc. The processor 1002 is used to implement the various steps of the aforementioned ethylene plant malfunction handling system when calling the computer program stored in the memory 1001.
[0118] This application embodiment does not limit the specific connection medium between the memory 1001 and the processor 1002. In Figure 10, the memory 1001 and the processor 1002 are connected via a bus 1003, which is depicted as a thick line in Figure 10. The connection methods between other components are only illustrative and are not intended to be limiting. The bus 1003 can be divided into address bus, data bus, control bus, etc. For ease of description, only one thick line is used in Figure 10, but it does not indicate that there is only one bus or one type of bus.
Claims
1. A knowledge graph based ethylene plant abnormality handling system, characterized in that, The system includes: An embedding learning module is used to perform representation learning on the knowledge graph of the ethylene plant based on a variety of preset embedding strategies, and obtain the vector representation corresponding to the knowledge graph; the knowledge graph contains multiple entities of the ethylene plant and the relationships between the entities, and each entity corresponds to the equipment, sensors and control logic of the ethylene plant respectively; the vector representation represents the feature vector of each entity and relationship in the knowledge graph. The reasoning and prediction module is used to perform anomaly reasoning based on the vector representation to obtain anomaly prediction results for the ethylene unit and corresponding solutions.
2. The system of claim 1, wherein, The system also includes a map construction module, used for: Obtain the device data of the ethylene unit; The knowledge graph is constructed based on the device data.
3. The system of claim 1, wherein, The system also includes: The interactive question-and-answer module is used to respond to questions about the ethylene plant input by the target object and output corresponding question-and-answer results based on the knowledge graph.
4. The system of claim 2, wherein, The map construction module is specifically used for: Based on the device data, the ontology file corresponding to the knowledge graph is determined. The ontology file includes custom classes, relational attributes, and data attributes. The classes represent the physical entities, operational logic, and abnormal phenomena involved in the operation of the ethylene device. The relational attributes represent the associations between the classes. The data attributes represent the attribute information corresponding to each class. The unstructured data in the device data is transformed into structured data to obtain the triple data of the knowledge graph, and the knowledge graph is visualized.
5. The system of claim 2, wherein, The various embedding strategies include geometric model embedding strategies, language model embedding strategies, and graph neural network embedding strategies; among them... The geometric model embedding strategy represents the triplet data of the knowledge graph through a geometric model. The language model embedding strategy is characterized by embedding text information in the device data based on natural language processing technology and combined with a pre-trained language model to obtain the corresponding semantic representation. The graph neural network embedding strategy is characterized by: processing the entities and relationships of the knowledge graph based on the graph neural network to determine the high-order relationships between entities in the knowledge graph; and generating vector representations of each entity through a graph convolutional network.
6. The system of claim 1, wherein, The inference and prediction module is specifically used for: Based on a preset anomaly detection training set and the vector representation, anomaly reasoning is performed on the ethylene unit; the anomaly detection training set includes ethylene unit data labeled with normal and abnormal states. When an abnormal situation occurs in the ethylene unit, the abnormality prediction result and the solution are output based on the knowledge graph.
7. The system of claim 5, wherein, The inference and prediction module is also used for: When the knowledge graph does not include the solution, a solution corresponding to the anomaly prediction result is generated based on a preset decision algorithm; Based on the current operating status of the ethylene plant, a personalized recommendation for the solution is made.
8. The system of claim 1, wherein, The device data includes at least the equipment information, sensor data, process parameters, and control loop information of the ethylene plant.
9. The system of claim 3, wherein, The interactive question-and-answer module is specifically used for: Keyword extraction was performed on the problems of the ethylene unit to determine the corresponding anomaly description information; Based on the knowledge graph and the reasoning and prediction module, the question-and-answer results corresponding to the anomaly description information are obtained and fed back to the target object; the question-and-answer results include the corresponding anomaly prediction results and solutions.
10. The system of claim 9, wherein, The system also includes a visualization module, which is used to display the question-and-answer results and the equipment status of the ethylene plant to the target object based on a visual interface.
11. The system of claim 1, wherein, The system construction method includes: Obtain the requirements analysis information of the ethylene unit's abnormal handling system, including user requirements information and system requirements information; Based on the aforementioned requirements analysis information, a system development environment is constructed, and based on the system development environment, the system architecture of the ethylene plant anomaly handling system is constructed; the system architecture includes a data layer, an application service layer, and a user layer; Based on a pre-defined modular development strategy, the functional modules of the ethylene unit's anomaly handling system are constructed sequentially.
12. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The memory is used to store a knowledge graph of the ethylene plant; When the processor executes the computer program, it implements the steps of the system according to any one of claims 1 to 11.
