Chemical hidden danger management method and system based on sequential network analysis

Abstract

The invention discloses a chemical hidden danger management method and system based on sequential network analysis, and relates to the technical field of hidden danger management. Compared with a previous hidden danger checking method, the hidden danger checking method solves the problems that a traditional hidden danger checking method depends on manpower, an existing static knowledge graph and an existing system have many defects such as incapability of reflecting dynamic characteristics, lack of time sequence analysis and the like, and difficulty in accurately judging hidden dangers and providing dynamic query services; modeling is carried out on chemical enterprise data, a time sequence knowledge graph is constructed on the basis, the dynamic process of entities and relationships thereof changing along with time can be recorded and analyzed, and the dynamic characteristics of the data are reflected; possible hidden dangers are accurately judged according to time sequence analysis and changes of risk points along with time; a complex network analysis technology is utilized to deeply analyze the time sequence knowledge graph, and a dynamic automatic reminding mechanism is combined to help managers to better manage and prevent safety problems; and an intelligent question-answering system is further integrated to help managers to carry out intelligent centralized management.

Description

Technical Field

[0001] The present invention relates to the technical field of hidden danger management, and in particular to a chemical industry hidden danger management method and system based on temporal network analysis. Background Art

[0002] With the rapid development of the chemical industry, the production scale and complexity of chemical companies are increasing, and safety issues are becoming increasingly prominent. Chemical accidents cause huge casualties and property losses. How to effectively manage and control hidden dangers has become an important task at present. However, traditional hidden danger detection methods mainly rely on manual inspection and experience judgment, which is not only time-consuming and labor-intensive, but also prone to omissions and misjudgments, and it is difficult to meet the high requirements of modern chemical production for safety management.

[0003] A knowledge graph is a graph data structure consisting of nodes and edges, where each node represents an entity in the real world and each edge represents the relationship between entities. By integrating and associating data through knowledge graphs, it is possible to reveal the information features and relationships hidden in the data, and provide deeper analysis and insights. The prior art has applied knowledge graphs in hidden danger management and accident analysis. For example, patent application number "CN115544272A" discloses "a method for constructing a knowledge graph of chemical accident causes based on an attention mechanism"; patent application number "CN116010581A" discloses "a knowledge graph question-and-answer method and system based on a power grid hidden danger investigation scenario"; patent application number "CN112948566B" discloses "a method and device for constructing a chemical knowledge graph and an intelligent question-and-answer method and device".

[0004] Complex network is a theory and method for studying network structure and dynamic behavior, which is widely used in many fields, such as social networks, biological networks, the Internet and transportation networks. Complex network analysis techniques include node centrality analysis, community detection, graph embedding, etc., which can reveal important nodes in the network, discover communities in the structure, and perform dimensionality reduction and feature extraction of high-dimensional data, which helps to find the clustering areas of hidden dangers and potential risk hotspots. For example, the patent application number "CN202410585855.9" discloses "an accident cause analysis method based on complex network theory."

[0005] Although knowledge graphs and complex networks play an important role in hidden danger management and accident analysis, and can provide a deep understanding of network structure and dynamic behavior, most existing applications are static, that is, the entities and relationships in the graph do not change after they are built, and it is difficult to capture the dynamic process of nodes and relationships in the network changing over time. This limitation makes it difficult to provide real-time risk assessment and hidden danger troubleshooting functions in a dynamic environment, and cannot meet the needs of modern enterprises for dynamic hidden danger management.

[0006] In order to solve the above problems, the present invention proposes a chemical industry hidden danger management method and system based on temporal network analysis. The present invention applies the temporal knowledge graph (Temporal Knowledge Graph, TKG) to construct a hidden danger management system for chemical enterprises. By introducing the time dimension, the dynamic process of entities and their relationships changing over time is recorded and analyzed, and a three-dimensional knowledge graph containing time attributes is constructed to intuitively display the dynamic changes of entities, events and their relationships, and provide comprehensive support. In addition, the present invention combines complex network analysis techniques (such as node centrality analysis, community detection, graph embedding, etc.) to conduct in-depth analysis of the temporal knowledge graph, identify key nodes and hidden danger aggregation areas, perform graph dimension reduction and feature extraction, and improve analysis and visualization effects. In order to ensure the timeliness and effectiveness of hidden danger management, the present invention designs and implements a dynamic automatic reminder mechanism, automatically generates reminder information according to the results of complex network analysis, dynamically identifies high-risk areas and potential hidden dangers by real-time monitoring of changes in nodes and relationships in the graph, and automatically reminds managers through text messages, emails or internal notification systems of enterprises to ensure timely attention to important hidden dangers and risk points. By introducing time-series knowledge graph technology, complex network analysis technology and dynamic automatic reminder mechanism, the present invention realizes dynamic hidden danger management, improves management accuracy and efficiency, provides strong support for the safe production of chemical enterprises, and significantly improves the overall safety level of enterprises. Summary of the invention

[0007] The purpose of the present invention is to propose a chemical industry hidden danger management method and system based on time series network analysis to solve the problems raised in the background technology:

[0008] Traditional hidden danger detection methods rely on manual inspection and empirical judgment, which makes it difficult to efficiently integrate and analyze data from different sources, and it is impossible to fully grasp the dynamic changes of risk factors; the existing static knowledge graph cannot reflect the dynamic characteristics of things and relationships in the real world that change over time, and it is difficult to provide accurate risk assessment and hidden danger detection functions; the lack of dynamic tracking of hidden dangers and risk factors that change over time makes it difficult to accurately judge potential hidden dangers, resulting in inaccurate inspections and missed hidden dangers; the existing system lacks the ability to perform time series analysis on historical data, and is unable to analyze the occurrence and development trends of hidden dangers based on the time dimension; the existing question-and-answer system based on the knowledge graph cannot effectively process information containing the time dimension, and it is difficult to provide users with dynamic hidden danger and accident information query services.

[0009] In order to achieve the above object, the present invention adopts the following technical solutions:

[0010] A chemical industry hidden danger management system based on time series network analysis, comprising:

[0011] Data acquisition module: used to collect and pre-process multi-source data of chemical enterprises;

[0012] Information processing module: used to define entities and relationships and perform knowledge modeling;

[0013] Graph construction module: used to build a time-series knowledge graph and update it in real time over time;

[0014] Network analysis module: Analyze based on the constructed time series knowledge graph to reveal the distribution of hidden dangers and risk points and their changing patterns;

[0015] Intelligent question-answering module: used to provide intelligent question-answering services based on the timeline in combination with the time-series knowledge graph;

[0016] Hidden danger inspection module: used to judge hidden dangers and conduct comprehensive inspections based on the time-series knowledge graph and intelligent question-answering module.

[0017] A chemical industry hidden danger management method based on time series network analysis includes the following steps:

[0018] S1: Multi-source data collection: Use the data acquisition module to collect multi-source data of chemical companies, establish a corresponding database, and pre-process the collected data;

[0019] S2: Hazard knowledge modeling: The information processing module defines the key entities and the relationships between entities in the chemical industry hazard knowledge graph;

[0020] S3: Building a temporal knowledge graph: The graph building module models the graph data according to the pre-defined ontology and determines the data structure and mode of the image database;

[0021] S4: Time series knowledge graph analysis: The network analysis module conducts in-depth analysis of the time series knowledge graph based on complex network analysis technology, and automatically generates and sends reminder information to managers based on the dynamic automatic reminder mechanism according to the network analysis results;

[0022] S5: Intelligent question-answering system: Integrates an intelligent question-answering module based on a temporal knowledge graph, uses the temporal knowledge graph to query based on the timeline, parses user query requests, retrieves relevant temporal data and returns corresponding answers.

[0023] Preferably, the preprocessing in S1 includes data cleaning and sorting operations to remove duplicate values, blank values ​​and abnormal values.

[0024] Preferably, the key entities in the chemical industry hidden danger knowledge graph in S2 include enterprise name, hidden danger, risk point and potential accident, and each entity corresponds to different attributes.

[0025] Preferably, the relationship between the entities in S2 includes hidden dangers in the enterprise, hidden dangers in the risk points of the enterprise, potential accidents faced by the enterprise, hidden dangers caused by failure of safety measures at risk points, potential accidents caused by loss of control of risk points, and potential accidents caused by the release of hidden dangers.

[0026] Preferably, the steps of constructing a time series knowledge graph in S3 are as follows:

[0027] S3.1: Obtain basic atlas data;

[0028] S3.2: Associate the data of hidden dangers, risk points and potential accidents in different time periods with the entities in the basic graph data to form time series graph data;

[0029] S3.3: Use these time-series graph data to build a target knowledge graph with time attributes to record and analyze the dynamic process of entities and their relationships changing over time.

[0030] Preferably, the complex network analysis technology in S4 performs an in-depth analysis of the temporal knowledge graph as follows:

[0031] S4.1: Extract all nodes and their relationships from the time series knowledge graph through node centrality analysis, calculate different centrality indicators of the nodes, identify key nodes, and conduct further analysis and monitoring;

[0032] S4.2: Use community detection algorithm to divide the graph, identify the hidden danger concentration areas, and determine the concentration degree and potential risk of hidden dangers by analyzing the characteristics of each community;

[0033] S4.3: Use graph embedding algorithm to embed nodes, reduce the high-dimensional graph structure to low-dimensional vector space, and use dimensionality reduction technology to reduce the dimension of node embedding for visual display.

[0034] Preferably, the dynamic reminder mechanism in S4 is as follows:

[0035] A: By real-time monitoring of node and relationship changes, the latest data changes are collected from the time-series knowledge graph regularly or in real time to update the graph information;

[0036] B: Using complex network analysis technology, we can calculate node centrality and community detection results in real time to identify high-risk areas and potential risks;

[0037] C: Define reminder rules;

[0038] D: Based on the analysis results and predefined rules, the system automatically generates reminder content;

[0039] E: Automatically send reminders via SMS, email or the company's internal notification system.

[0040] Preferably, the intelligent question-answering module described in S5 integrates a BERT pre-trained model, develops an algorithm to identify the intent and key information of user queries; converts the user's natural language query into a structured query representation based on a natural language parser; and defines a time series data model in a graph database, including time attributes and a historical log mechanism, which are used to record the time changes of entities and relationships, and perform queries based on a time series query optimization strategy; uses index optimization and query path selection technology to perform information retrieval, and based on a data aggregation mechanism, provides corresponding information in combination with historical hidden danger data and safety inspection records.

[0041] Preferably, in S3-S5, the time series knowledge graph and the intelligent question and answer module are combined, and the hidden danger inspection module determines the chemical hidden danger according to the change of risk points over time.

[0042] Compared with the prior art, the present invention provides a chemical industry hidden danger management method and system based on time series network analysis, which has the following beneficial effects:

[0043] 1. Dynamic hidden danger management: The time-series knowledge graph of the present invention can reflect the dynamic characteristics of things and relationships in the real world that change over time, making enterprise hidden danger management more accurate and comprehensive. By recording and analyzing the dynamic process of entities and their relationships changing over time, the timeliness and effectiveness of hidden danger management are improved.

[0044] 2. Accurate hidden danger inspection: This invention uses the time series knowledge graph to accurately judge the possible hidden dangers according to the changes of risk points over time, ensure the comprehensiveness and accuracy of the inspection, and avoid the omission of hidden dangers. By dynamically tracking risk points, it provides accurate hidden danger inspection and risk management strategies, which improves the effectiveness of hidden danger inspection.

[0045] 3. Time series analysis capability: Through the time series knowledge graph, the present invention can perform time series analysis on historical data, analyze the occurrence and development trend of hidden dangers based on the time dimension, and provide forward-looking risk management strategies. This capability enables enterprises to better prevent and control potential risks.

[0046] 4. Complex network analysis technology: The present invention uses complex network analysis technology to conduct in-depth analysis of the time series knowledge graph. Key nodes are identified through node centrality analysis, hidden danger clustering areas are discovered through community detection, and graph embedding technology is used to reduce the dimension and extract features of the graph for better analysis and visualization. This complex network analysis method can not only fully reveal the distribution of hidden dangers and risk points and their changing patterns, but also help managers better understand and prevent potential safety issues.

[0047] 5. Dynamic automatic reminder mechanism: The present invention designs and implements a dynamic automatic reminder mechanism, which automatically generates and sends reminder information to managers based on the results of complex network analysis. This mechanism can monitor the changes of nodes and relationships in the graph in real time, dynamically identify high-risk areas and potential hidden dangers, and automatically remind managers through SMS, email or internal notification system of the enterprise, ensuring that managers can pay attention to important hidden dangers and risk points in a timely manner.

[0048] 6. Intelligent question-answering system: The present invention integrates an intelligent question-answering system based on a time-series knowledge graph, and users can query hidden dangers and accident information within a specific time period through natural language. The system can parse query requests, retrieve relevant data and return accurate answers, helping inspectors obtain the required information and make wise decisions.

[0049] 7. Real-time data update: The data of the time-series knowledge graph constructed by the present invention can be updated in real time as time goes by, ensuring that the latest data can be reflected in the graph in a timely manner. The real-time update mechanism provides the latest and most comprehensive information, improving the system's response speed and decision-making accuracy.

[0050] 8. Modular design of the system: The modules in the system of the present invention work together to effectively integrate and analyze multi-source data, fully grasp the dynamic changes of risk factors, and improve the flexibility and scalability of the system. This modular design makes the system easy to maintain and expand, and can adapt to the application of chemical companies of different scales and complexities. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is the system architecture diagram mentioned in Example 1 of the present invention;

[0052] Figure 2 This is a system block diagram mentioned in Embodiment 1 of the present invention;

[0053] Figure 3 This is a schematic diagram of knowledge modeling mentioned in Embodiment 2 of the present invention;

[0054] Figure 4 A flowchart for constructing the knowledge graph and complex network mentioned in Example 2 of the present invention;

[0055] Figure 5 This is a schematic diagram of the temporal knowledge graph mentioned in Example 2 of the present invention;

[0056] Figure 6 This is a technical diagram of the intelligent question-answering system query mentioned in Example 2 of the present invention.

[0057] Meaning of the symbols in the figure:

[0058] 100. Data acquisition module; 200. Information processing module; 300. Graph construction module; 400. Network analysis module; 500. Intelligent question and answer module; 600. Hidden danger inspection module. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0060] The present invention models the data of chemical enterprises and constructs a time series knowledge graph based on it, which can record and analyze the dynamic process of entities and their relationships changing over time, reflecting the dynamic characteristics of data; it can also accurately judge the potential hidden dangers based on the time series analysis and the changes of risk points over time; it also uses complex network analysis technology to conduct in-depth analysis of the time series knowledge graph, and combines it with a dynamic automatic reminder mechanism to help managers better manage and prevent safety issues; it further integrates an intelligent question-and-answer system to help managers conduct intelligent centralized management. Specifically, it includes the following contents.

[0061] Embodiment 1:

[0062] See also Figure 1-2 The present invention provides a chemical industry hidden danger management system based on time series network analysis, comprising:

[0063] Data acquisition module 100: used to collect and pre-process multi-source data of chemical enterprises;

[0064] Information processing module 200: used to define entities and relationships in the knowledge graph of hidden dangers in the chemical industry, and to perform ontology modeling and graph data structure design;

[0065] Graph construction module 300: used to construct a three-dimensional knowledge graph containing time attributes based on Neo4j, and update it in real time over time to improve the efficiency and accuracy of data integration and analysis;

[0066] Network analysis module 400: performs analysis based on the constructed time series knowledge graph to reveal the distribution of hidden dangers and risk points and their changing patterns, and provides forward-looking risk management strategies;

[0067] Intelligent question-answering module 500: used to provide intelligent question-answering services based on the time axis in combination with the time series knowledge graph. Users can query the hidden dangers and accident information within a specific time period through natural language. The system parses the query request, retrieves the relevant time series data and returns accurate answers, significantly improving the accuracy and practicality of information query;

[0068] Hidden danger inspection module 600: used to judge hidden dangers and conduct comprehensive inspection based on the time series knowledge graph and intelligent question and answer module 500.

[0069] Embodiment 2:

[0070] See also Figure 3-6 The present invention provides a chemical industry hidden danger management method based on time series network analysis, comprising the following steps:

[0071] S1: Multi-source data collection: Use the data acquisition module 100 to collect multi-source data of chemical enterprises, establish a corresponding database, and pre-process the collected data; the details are as follows:

[0072] In the hidden danger management of chemical enterprises, data sources are extensive and diverse, including but not limited to enterprise operation data, safety inspection records, accident reports, etc. Through the interface connection with the enterprise's internal database, safety supervision system, and external environmental protection and industry supervision database, the system can collect necessary historical and real-time data. In addition, in order to ensure the comprehensiveness and multi-angle analysis of the data, employee safety behavior records, equipment maintenance records, and historical accident data are also included. Due to the diverse sources of data, the format and structure may be inconsistent, so data integration is required. First, use the Talend data integration tool to extract raw data from multiple heterogeneous data sources, configure Talend tasks, and regularly extract data from each data source to temporary storage. Next, define data conversion rules in Talend to unify the data format and encoding of different data sources, for example, convert all time data to ISO 8601 standard format. Then, load the converted data into a unified data warehouse, and configure Talend tasks to regularly load the processed data into the target database.

[0073] During the data cleaning phase, OpenRefine tools were used to perform operations such as format standardization, deduplication, error correction, and missing value filling. Data cleaning rules were defined in OpenRefine to unify the field formats of different data sources, for example, all text fields were converted to UTF-8 encoding. Duplicate records were removed by matching key fields (such as equipment ID, inspection date, etc.), and unique constraints were set to automatically detect and delete duplicate records. Common errors were automatically identified and corrected using OpenRefine’s anomaly detection function, and corrections were made manually or automatically. For missing data, the nearest neighbor interpolation method (KNN) was used to fill in missing inspection records.

[0074] In the outlier detection and processing stage, statistical methods such as standard deviation analysis and box plots are used to identify outliers in the data. Using libraries such as pandas and numpy in Python, statistical indicators such as the mean, standard deviation, and quartiles of the data set are calculated, and Python scripts are written to read the cleaned data for statistical analysis. Outliers are identified through methods such as the 3σ principle and box plots. The 3σ principle and box plot analysis are implemented in Python scripts to mark outliers. Combined with business logic and historical data, manual review and automated correction are performed. The automated part uses a random forest model to predict and replace abnormal data points. The sklearn library is called in the Python script to train a random forest model to predict and replace outliers.

[0075] In order to meet the needs of subsequent knowledge graph construction, necessary data conversion is performed. Numerical data is scaled to a specified range (such as 0 to 1) to ensure the balance of different features in the analysis, and a Python script is written to use the sklearn.preprocessing library to normalize the data. Numerical data is standardized (Z-score standardization) to improve data comparability, and the sklearn.preprocessing library is used in the Python script to perform Z-score standardization. According to the needs of graph modeling, key data is feature extracted and encoded. For example, the accident type, hidden danger level, etc. are extracted and one-hot encoded in the Python script using pandas and sklearn libraries.

[0076] S2: Hidden danger knowledge modeling: The information processing module 200 defines the key entities and the relationships between entities in the hidden danger knowledge graph of the chemical industry; the details are as follows:

[0077] Knowledge modeling diagram Figure 3, the present invention identifies key entity categories in the hidden danger management of chemical enterprises, including "enterprise", "hidden danger", "risk point" and "potential accident", and defines detailed attributes for each entity category. For example, the attributes of the "hidden danger" entity include hidden danger description, discovery date, severity, current status and expected resolution time, etc. The specific implementation steps are first to determine the key entity categories by analyzing the business processes and documents of hidden danger management in chemical enterprises, and then use the UML class diagram tool StarUML to define the detailed attributes of each entity to ensure that the attribute description is comprehensive and accurate. Next, define the relationship types between entities. These relationships not only describe the connection mode between entities, but also reflect the dependency and interactivity between them, such as "hidden dangers in enterprises", and give attributes to relationships such as the start time, impact degree and status of the relationship. The relationship types between entities are identified through business process analysis and expert interviews, and the relationship types and their attributes are defined in the UML class diagram. The entities and relationships are modeled using graph database modeling tools (such as Neo4j) to ensure the structure and relevance of the data.

[0078] When constructing an ontology model for hidden danger management, use an ontology editing tool (such as Protégé) to design the ontology structure, define classes, subclasses, attributes, relationships, and their constraint logic, such as defining the "hidden danger" class and its subclasses "minor hidden danger" and "serious hidden danger", etc., to ensure the rigor and scalability of the model. Then use SPARQL or SWRL in Protégé to write ontology rules to ensure the logical consistency and integrity of entities and relationships, such as defining the rule "If the severity of the hidden danger is high, its expected resolution time should be less than 7 days."

[0079] In terms of extracting knowledge from unstructured data using natural language processing (NLP) tools, we preprocessed the unstructured data, converted the text data into structured data using technologies such as word segmentation, part-of-speech tagging, and named entity recognition, and used the NLP toolkit NLTK for data preprocessing. Then we used the BERT (Bidirectional Encoder Representations from Transformers) model to extract entities and relationships from unstructured text, build a knowledge graph, and train a custom NLP model to ensure the accuracy and completeness of the extraction results. Finally, we fused the extracted knowledge with the structured data, used the data fusion tool Dedupe to handle data conflicts and redundancies, formed a unified knowledge base, and stored the fused data in the graph database Neo4j to ensure data query efficiency and storage structure optimization.

[0080] S3: Constructing a temporal knowledge graph: The graph construction module 300 models the graph data according to a predefined ontology and determines the data structure and mode of the image database; the details are as follows:

[0081] The process of building a temporal knowledge graph is as follows: Figure 4 As shown, this embodiment selects Neo4j as the graph database, and its excellent performance has obvious advantages in processing complex relationships and dynamic data. Neo4j's graph structure is intuitive and easy to expand, which is very suitable for realizing complex entity relationships and dynamic changes in hidden danger management. First, by installing and configuring the Neo4j database, adjust the memory allocation and cache settings to support the processing of large amounts of data and high-frequency data updates. Create indexes for key attributes (such as entity ID, relationship type, etc.) to improve query performance, and define unique constraints and foreign key constraints to ensure data integrity and consistency, such as creating a unique identifier for each hidden danger entity. Next, define time attributes for each entity and relationship, including time tags for creation time, modification time, and historical status, such as adding created_at, updated_at, and status_history attributes to the "hidden danger" entity. Design and implement a mechanism for recording historical changes in entities and relationships. Each time an entity or relationship changes, automatically update its historical log, and use Neo4j's triggers or scheduled tasks to record historical status when data changes.

[0082] To ensure that the information in the graph database is always up to date and consistent with other business systems such as risk assessment and hidden danger investigation and reporting systems, develop data synchronization interfaces with these systems and use APIs to achieve real-time data synchronization. Run data consistency checkers regularly, call API interfaces through scheduled tasks, compare data differences and make corrections. Use Cypher query language to develop complex time series queries to track changes in entities and relationships over time, such as querying the evolution of hidden dangers or the activity history of risk points within a specific time period. The effect diagram is shown below. Figure 5 The specific implementation can be written into a Cypher query statement, for example:

[0083] MATCH(h:Hazard)-[r:HAS_STATUS]->(s:Status)

[0084] WHEREr.start_date>=$start_date AND r.end_date<=$end_date

[0085] RETURN h,r,s

[0086] At the same time, Python scripts are integrated for statistics and analysis to conduct more in-depth time series data analysis and pattern recognition. Python's Neo4j driver py2neo is used to connect to the database, execute Cypher queries and process the results, and use pandas and matplotlib libraries for data analysis and visualization. The Neo4j Bloom tool is used to display the data in the time series knowledge graph, so that users can intuitively see the relationship between hidden dangers, risk points and measures and their changes over time.

[0087] S4: Time series knowledge graph analysis: The network analysis module 400 performs in-depth analysis on the time series knowledge graph based on complex network analysis technology, and automatically generates and sends reminder information to the manager based on the network analysis results based on a dynamic automatic reminder mechanism; the details are as follows:

[0088] This embodiment uses complex network analysis techniques (such as node centrality analysis, community detection, graph embedding, etc.) to conduct in-depth analysis of the time series knowledge graph to effectively identify and manage hidden dangers in chemical companies. First, by extracting all nodes (such as equipment, hidden dangers, risk points, etc.) and their relationships from the time series knowledge graph, and using Python's NetworkX library to calculate different centrality indicators of nodes (including degree centrality, betweenness centrality, proximity centrality, etc.), key nodes such as high-risk hidden dangers and important equipment are identified for further analysis and monitoring. Secondly, the community detection algorithm is used to divide the graph, identify the hidden danger gathering area, and determine the concentration degree and potential risk of its hidden dangers by analyzing the characteristics of each community. In addition, the graph embedding algorithm is used to embed the nodes, and the high-dimensional graph structure is reduced to a low-dimensional vector space for feature extraction and visualization, and the node embedding is reduced in dimension using dimensionality reduction techniques for visual display. This complex network analysis method can not only fully reveal the distribution and changing laws of hidden dangers and risk points, but also help managers better understand and prevent potential safety issues, and significantly improve the accuracy and efficiency of hidden danger management in chemical companies through more in-depth and intuitive analysis.

[0089] A dynamic automatic reminder mechanism was designed and implemented. According to the results of complex network analysis, reminder information was automatically generated and sent to managers. First, by real-time monitoring of node and relationship changes, the system collects the latest data changes from the time series knowledge graph regularly or in real time to update the graph information. Using complex network analysis technology, the node centrality and community detection results are calculated in real time to identify high-risk areas and potential hidden dangers. Then, reminder rules are defined. For example, when the centrality index of a node exceeds a certain threshold, or the hidden danger density of a community reaches a certain level, a reminder is triggered. According to the analysis results and predefined rules, the system automatically generates reminder content, including detailed information such as high-risk hidden dangers, important equipment, and hidden danger clustering areas. Finally, an appropriate notification method is selected to send reminder information through SMS, email, or the company's internal notification system to realize automatic sending of reminders. In this way, managers can obtain important hidden danger and risk information in a timely manner and take corresponding measures to ensure the timeliness and effectiveness of hidden danger management. Through this dynamic automatic reminder mechanism, the system can significantly improve the response speed and decision-making accuracy of hidden danger management in chemical enterprises, and help managers prevent and control potential safety problems more efficiently.

[0090] S5: Intelligent question-answering system: Integrate the intelligent question-answering module 500 based on the time series knowledge graph, use the time series knowledge graph to query based on the time axis, parse the user's query request, retrieve relevant time series data and return the corresponding answer. The details are as follows:

[0091] The process of intelligent question answering system is as follows Figure 6 As shown, the intelligent question-answering module 500 of this embodiment integrates the BERT pre-training model and develops an algorithm to identify the intention and key information of the user's query, such as distinguishing whether the query is a hidden danger status, a risk point status, or other content. First, by selecting a suitable BERT pre-training model and fine-tuning it on the chemical enterprise hidden danger management data set to improve the model's understanding of specific fields. Use the PyTorch framework to load and train the model, develop a BERT-based intention recognition algorithm, use transfer learning technology, and train the model by annotating the data set to identify the intention and key information of the user's query, such as classifying the query into hidden danger status query, risk point status query, and other categories through classification tasks.

[0092] Next, we developed a natural language parser to convert the user's natural language query into a structured query representation. We used NLP techniques, including semantic analysis and keyword extraction, to identify key entities and relationships in the query, and implemented a query conversion engine to convert the structured query representation into Cypher query language, process complex query structures, and generate efficient Cypher query statements.

[0093] In terms of processing time series data queries, define a time series data model in the graph database, including time attributes and historical log mechanisms, to ensure that the time changes of entities and relationships can be recorded and queried. Develop time series query optimization strategies to ensure that queries can be executed efficiently. Use index optimization and query path selection techniques to improve the efficiency and accuracy of data retrieval, such as using time range indexes to speed up time series data queries, and implement data aggregation mechanisms to combine historical hidden danger data and recent safety inspection records to provide comprehensive information, such as aggregating hidden danger status changes at different time points to generate trend reports.

[0094] The system operation process is as follows: the system first receives the user's natural language input. Through semantic analysis and keyword extraction in natural language processing (NLP) technology, the system analyzes the text input by the user, identifies its intent and key information, and further identifies entities in the input (such as specific hidden danger types, dates, etc.) to accurately locate the user's query needs. The user's natural language query is converted into Cypher query language. Before executing the query, the system will optimize the query command by selecting the most effective query path and index to improve the efficiency and accuracy of data retrieval. The system executes the optimized query in the graph database, retrieves data related to the user's query, such as specific information about the hidden danger, historical records, and possible related data, and performs necessary logical judgments, such as whether the query results need to be further processed or whether additional input is required.

[0095] The query results are further aggregated, such as calculated, sorted, or filtered, to meet the specific needs of the user, and the processed data is presented to the user in an easy-to-understand format. The system provides a feedback mechanism, and users can evaluate the query results. The system uses user feedback to learn and optimize, and adjust the query processing process and result display method.

[0096] Based on the time series knowledge graph and the hidden danger inspection module 500, the hidden danger inspection uses algorithms to generate customized inspection checklists based on historical data and current inspection requirements to help inspectors identify and record hidden dangers on site. By analyzing the historical hidden danger data in the time series knowledge graph, high-risk areas and common hidden danger types are identified. Combined with current inspection requirements and conditions, the graph database query and data analysis technology are used to write Cypher query statements in Neo4j to dynamically generate inspection checklists.

[0097] The machine learning model was developed using Python's Scikit-learn library. A decision tree model was trained to assess the priority and severity of hidden dangers based on multiple dimensions (such as severity, frequency, and detection time). The model was updated regularly to improve prediction accuracy. The specific implementation included regularly obtaining the latest hidden danger data, retraining the model, and verifying its performance. The model's hyperparameter optimization was achieved through grid search.

[0098] Using a content filtering-based recommendation system, we use natural language processing technology to extract key features from descriptions and automatically match current hidden danger features with historical solutions. By using NLP technology to convert hidden danger description text into vector representation, we calculate the similarity between current hidden danger and historical hidden danger descriptions, and recommend the most relevant solutions.

[0099] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A chemical industry hidden danger management system based on time series network analysis, characterized in that: include: Data acquisition module (100): used to collect and pre-process multi-source data of chemical enterprises; Information processing module (200): used to define entities and relationships and perform knowledge modeling; Graph construction module (300): used to construct a time series knowledge graph and update it in real time over time; Network analysis module (400): performs analysis based on the constructed time series knowledge graph to reveal the distribution of hidden dangers and risk points and their changing patterns; Intelligent question and answer module (500): used to provide intelligent question and answer service based on the time axis in combination with the time series knowledge graph; Hidden danger inspection module (600): used to judge hidden dangers and conduct comprehensive inspection based on the time series knowledge graph and intelligent question and answer module (500).

2. A chemical industry hidden danger management method based on time series network analysis implemented by the system according to claim 1, characterized in that: The steps include: S1: Multi-source data collection: using the data acquisition module (100) to collect multi-source data of chemical enterprises, and establish a corresponding database, and also pre-process the collected data; S2: Hidden danger knowledge modeling: The information processing module (200) defines key entities in the knowledge graph of hidden dangers in the chemical industry and the relationships between the entities; S3: constructing a temporal knowledge graph: the graph construction module (300) models the graph data according to a predefined ontology and determines the data structure and mode of the image database; S4: Time series knowledge graph analysis: The network analysis module (400) performs in-depth analysis on the time series knowledge graph based on complex network analysis technology, and automatically generates and sends reminder information to the manager based on the network analysis results based on a dynamic automatic reminder mechanism; S5: Intelligent question-answering system: Integrate the intelligent question-answering module (500) based on the time series knowledge graph, use the time series knowledge graph to query based on the time axis, parse the user's query request, retrieve relevant time series data and return the corresponding answer.

3. A chemical industry hidden danger management method based on temporal network analysis according to claim 2, characterized in that: The preprocessing in S1 includes data cleaning and sorting operations to remove duplicate values, blank values ​​and abnormal values.

4. A chemical industry hidden danger management method based on temporal network analysis according to claim 2, characterized in that: The key entities in the chemical industry hidden danger knowledge graph in S2 include enterprise name, hidden danger, risk point and potential accident, and each entity corresponds to different attributes.

5. A chemical industry hidden danger management method based on temporal network analysis according to claim 4, characterized in that: The relationships between the entities in S2 include hidden dangers in the enterprise, hidden dangers in the enterprise's risk points, the enterprise facing potential accidents, hidden dangers caused by failure of safety measures at risk points, potential accidents caused by loss of control of risk points, and potential accidents caused by hidden dangers.

6. A chemical industry hidden danger management method based on temporal network analysis according to claim 5, characterized in that: The steps for constructing the time series knowledge graph in S3 are as follows: S3.1: Obtain basic atlas data; S3.2: Associate the data of hidden dangers, risk points and potential accidents in different time periods with the entities in the basic graph data to form time series graph data; S3.3: Use these time-series graph data to build a target knowledge graph with time attributes to record and analyze the dynamic process of entities and their relationships changing over time.

7. A chemical industry hidden danger management method based on temporal network analysis according to claim 6, characterized in that: The detailed process of in-depth analysis of the time series knowledge graph by the complex network analysis technology in S4 is as follows: S4.1: Extract all nodes and their relationships from the time series knowledge graph through node centrality analysis, calculate different centrality indicators of the nodes, identify key nodes, and conduct further analysis and monitoring; S4.2: Use community detection algorithm to divide the graph, identify the hidden danger concentration areas, and determine the concentration degree and potential risk of hidden dangers by analyzing the characteristics of each community; S4.3: Use graph embedding algorithm to embed nodes, reduce the high-dimensional graph structure to low-dimensional vector space, and use dimensionality reduction technology to reduce the dimension of node embedding for visual display.

8. A chemical industry hidden danger management method based on temporal network analysis according to claim 7, characterized in that: The dynamic reminder mechanism described in S4 is as follows: A: By real-time monitoring of node and relationship changes, the latest data changes are collected from the time-series knowledge graph regularly or in real time to update the graph information; B: Using complex network analysis technology, we can calculate node centrality and community detection results in real time to identify high-risk areas and potential risks; C: Define reminder rules; D: Based on the analysis results and predefined rules, the system automatically generates reminder content; E: Automatically send reminders via SMS, email or the company's internal notification system.

9. A chemical industry hidden danger management method based on temporal network analysis according to claim 8, characterized in that: The intelligent question-answering module (500) described in S5 integrates the BERT pre-training model and develops an algorithm to identify the intention and key information of the user's query; converts the user's natural language query into a structured query representation based on a natural language parser; and defines a time series data model in a graph database, including time attributes and a historical log mechanism, which are used to record the time changes of entities and relationships, and perform queries based on a time series query optimization strategy; uses index optimization and query path selection technology to perform information retrieval, and based on a data aggregation mechanism, combines historical hidden danger data and safety inspection records to provide corresponding information.

10. A chemical industry hidden danger management method based on temporal network analysis according to claim 9, characterized in that: In S3-S5, the time series knowledge graph and the intelligent question and answer module (500) are combined, and the hidden danger inspection module (600) determines the chemical hidden danger according to the change of the risk point over time.

Images