A pre-shift meeting outline dynamic generation method and system based on a knowledge graph

Abstract

This invention discloses a method and system for dynamically generating pre-shift meeting outlines based on knowledge graphs. The method includes: pre-constructing a construction safety knowledge graph based on multi-source data; acquiring and vectorizing current construction scenario information to activate entity nodes corresponding to the scenario information in the construction safety knowledge graph; using the activated entity nodes as starting points, deriving potential cascading risk transmission paths from the direct risks caused by the current construction scenario; and extracting control measures associated with each risk entity from the construction safety knowledge graph based on the direct risks and potential cascading risk transmission paths, and automatically generating a structured pre-shift meeting outline. This invention enables predictability and depth in risk identification, ensures the accuracy and dynamism of the briefing content, and significantly improves the intelligence level and briefing effect of on-site safety management.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of information technology and safety management, and more particularly to a method and system for dynamically generating pre-shift meeting outlines based on knowledge graphs. More specifically, this invention utilizes artificial intelligence technology, particularly knowledge graphs and graph reasoning algorithms, to achieve a method and related system for automatically generating highly targeted pre-shift safety meeting outlines based on real-time, dynamic construction scenario information. Background Technology

[0002] In engineering construction, energy extraction, equipment maintenance, and other engineering construction fields, the daily pre-shift meeting is the first and crucial line of defense for implementing safety production responsibilities and preventing safety accidents. The core function of the pre-shift meeting is for the team leader to clearly and accurately convey the day's work tasks to all workers, explain key technical points and operating procedures, highlight potential safety risks closely related to the day's work content and working environment, and deploy specific and executable risk control measures.

[0003] However, existing methods for organizing and generating pre-shift meetings generally rely on manual experience or generic templates. Information technology tools mostly only digitize paper-based content, enabling only basic safety briefings and recording. They cannot integrate multi-dimensional data such as specific construction procedures, work locations, personnel status, and dynamic environments for intelligent analysis and precise matching. This results in a serious disconnect between the briefing content and the actual site conditions, incomplete and superficial risk identification, and difficulty in structuring and reusing individual safety knowledge and experience. Consequently, the effectiveness of briefings is greatly reduced, becoming merely a formality, posing numerous challenges to the targetedness, standardization, and knowledge transfer of construction safety management.

[0004] Therefore, there is an urgent need in this field for a technical solution that can break through the current technical bottlenecks. This solution should be able to automatically integrate multi-dimensional on-site dynamic data, intelligently analyze and accurately predict the specific risks of the day's work based on deep knowledge models, and generate differentiated pre-shift briefing content in a more efficient and instructive manner. Summary of the Invention

[0005] The main objective of this invention is to overcome the technical shortcomings of existing technologies, such as the severe disconnect between the content of pre-shift safety briefings and actual construction scenarios, and the reliance on and incompleteness of risk identification based on personal experience. To achieve this objective, this invention provides a method, system, and related products that can automatically and dynamically integrate real-time scenario information and perform deep risk reasoning based on knowledge graphs to accurately generate customized pre-shift meeting outlines. This invention aims to ensure that safety briefing content comprehensively and thoroughly covers all key risk points of the day's work, particularly in anticipating and revealing potential cascading risks.

[0006] Firstly, this application provides a method for dynamically generating pre-shift meeting outlines based on knowledge graphs. This method is executed by a computer system equipped with a processor and memory, and includes: pre-constructing a construction safety knowledge graph containing safety entities and relationships between entities based on multi-source data including construction plans and historical accident cases; acquiring current construction scenario information in real time through an information interface and vectorizing the scenario information to activate entity nodes corresponding to the scenario information in the construction safety knowledge graph; starting from the activated entity nodes, executing a graph reasoning algorithm on the construction safety knowledge graph, traversing preset relationship paths to deduce potential chain risk transmission paths from the direct risks caused by the current construction scenario; and extracting control measures associated with each risk entity in the direct risks and risk transmission paths from the construction safety knowledge graph based on the direct risks and risk transmission paths, and automatically generating a structured pre-shift meeting outline.

[0007] Optionally, in this embodiment, the step of acquiring the current construction scene information in real time through an information interface and vectorizing the scene information to activate the corresponding entity node in the construction safety knowledge graph includes: collecting real-time environmental data of the work location through IoT sensors; combining the preset process information in the daily construction task list, vectorizing the real-time environmental data and the preset process information together to activate the corresponding environmental entity node and process entity node in the knowledge graph. This ensures the real-time and authenticity of the scene information acquisition, solves the drawbacks of traditional safety briefings relying on static information, and achieves structured fusion of multi-source scene data through the collaborative vectorization processing of the two types of information, accurately activating the corresponding environmental entity node and process entity node in the knowledge graph. This lays a solid foundation for subsequent direct risk identification and potential chain risk deduction based on the knowledge graph. The intelligent data processing method replaces manual experience judgment, which also greatly improves the accuracy and efficiency of scene information processing, providing a key prerequisite for finally generating a customized pre-shift meeting outline that comprehensively covers the key risk points of the day's work and can foresee and reveal potential chain risks.

[0008] Optionally, in this embodiment, the relationships between entities in the construction safety knowledge graph include: causal relationships representing the root causes of risks, triggering relationships representing new risks triggered by specific conditions, and aggravating relationships representing the amplification of risk levels by environmental factors. The graph reasoning algorithm traverses along one or more of these relationships to discover the risk transmission paths. By clarifying the three types of relationships between entities—causal, triggering, and aggravating—in the construction safety knowledge graph, the core correlation logic of risk root causes, condition triggering, and environmental amplification is clearly defined. By using the graph reasoning algorithm to traverse along these relationships, starting from activated entity nodes, the risk transmission paths derived from direct risks can be accurately mined, effectively achieving predictability and depth in risk identification.

[0009] Optionally, in this embodiment of the application, before the pre-shift meeting outline is generated, the method further includes: quantifying and ranking the risk values ​​of the multiple risk transmission paths derived from reasoning, so as to scientifically define the control priorities of different risk paths; correspondingly, the generation of the structured pre-shift meeting outline includes: according to the ranking results, prioritizing the generation of risks, control measures and transmission logic corresponding to high-risk paths in a prominent position in the pre-shift meeting outline, which can guide operators to quickly focus on core risks and key control points, strengthen the pertinence and warning of the briefing content, and ensure that high-priority risks are given key attention and implemented.

[0010] In the above implementation process, the quantitative calculation of the risk value includes: comprehensively evaluating the preset risk weights of each entity node constituting the risk transmission path, and the preset transmission probabilities of each relation edge connecting the entity nodes; and obtaining the comprehensive risk value of each risk transmission path through a weighted cumulative algorithm. This fully considers the severity of the risk itself, while also taking into account the possibility of risk transmission, achieving a precise quantitative assessment of potential cascading risks and providing accurate data support for prioritizing risk transmission paths.

[0011] Optionally, in this embodiment of the application, after the pre-shift meeting outline is generated, the method further includes: using the key risk entities identified in the risk transmission path as search keywords, automatically matching and retrieving corresponding visual safety education materials from a preset multimedia database, and associating and pushing the visual safety education materials with the pre-shift meeting outline. This can transform the abstract risk transmission logic and control requirements into intuitive and easy-to-understand content, and strengthen the operators' awareness and vigilance regarding key risks.

[0012] Optionally, in this embodiment, dynamically adjusting the weights of entity nodes or the confidence levels of relationships between entities in the knowledge graph includes: collecting actual construction feedback data after pre-shift meetings, the feedback data including missed risk events, false risk events, and effective on-site temporary control measures; using a Bayesian update algorithm, updating the confidence weights of the connections between entity nodes in the knowledge graph based on the feedback data using posterior probability, to strengthen the verified risk transmission path and weaken false alarm paths; performing natural language processing and cluster analysis on the text description information in the feedback data, automatically generating new risk entity nodes and associated edges in the knowledge graph when the frequency of specific keywords exceeds a preset threshold, achieving incremental updates to the knowledge graph structure. This effectively solves the pain points of missed and false alarms in traditional risk identification, improving the accuracy and predictability of risk identification; performing natural language processing and cluster analysis on the text information in the feedback data to achieve incremental updates of new risk entity nodes and associated edges in the knowledge graph allows the knowledge graph to continuously evolve with construction practice, significantly improving the intelligence level and long-term control effect of safety management.

[0013] Optionally, in this embodiment, a personnel profile sub-graph is established. This sub-graph includes personnel entities as graph nodes, and static and dynamic attributes associated with each entity. The static attributes include job type, qualifications, skill level, and historical violation records. The dynamic attributes include real-time physiological state characteristics obtained through wearable devices. Correspondingly, a daily shift schedule list and its real-time physiological state characteristics are obtained, and these are mapped to the personnel profile sub-graph to activate the corresponding personnel entity nodes. This achieves structured management and dynamic updating of personnel information, effectively identifying specific risks caused by mismatch between personnel and positions or physiological state maladaptation to the environment, enriching the dimensions and depth of risk identification.

[0014] In the above process, cross-domain association reasoning is performed, and the activated personnel entity nodes are respectively matched with the activated process entity nodes and environmental entity nodes for feature calculation. The reasoning algorithm is configured to calculate the person-job fit degree between the personnel entity nodes and the process entity nodes. When the person-job fit degree is lower than a preset safety threshold, an intermediate node representing the risk of insufficient skills is activated in the knowledge graph. The reasoning algorithm is also configured to calculate the physiological fitness degree between the personnel entity nodes and the environmental entity nodes. When the physiological fitness degree shows an abnormality and the corresponding environmental entity attribute is severe weather conditions, a specific risk node representing a decline in the physiological function of a specific person is activated in the knowledge graph. Accordingly, for the identified specific risk nodes, a targeted early warning prompt containing a specific personnel list is generated in the pre-shift meeting outline. This solves the pain point of traditional safety management that easily overlooks the adaptability of personnel to processes and environments, and fills the gap in identifying specific risks caused by person-job mismatch and personnel's physiological state not adapting to harsh environments.

[0015] Secondly, an embodiment of this application provides a dynamic pre-shift meeting outline generation device, comprising: one or more processors; a memory storing instructions executable by the one or more processors; when the instructions are executed, the one or more processors implement the above-described method; specifically, the device may include: a knowledge graph construction module, used to pre-construct a construction safety knowledge graph containing safety entities and relationships between entities based on multi-source data including construction plans and historical accident cases; a scenario information processing module, used to obtain current construction scenario information in real time through an information interface and vectorize the scenario information to activate entity nodes corresponding to the scenario information in the construction safety knowledge graph; a risk reasoning engine, configured to execute a graph reasoning algorithm on the construction safety knowledge graph starting from the activated entity nodes, and derive potential chain risk transmission paths from the direct risks caused by the current construction scenario by traversing preset relationship paths; and an outline generation module, used to extract control measures associated with each risk entity in the direct risks and risk transmission paths from the construction safety knowledge graph based on the direct risks and risk transmission paths, and automatically organize and generate a structured pre-shift meeting outline.

[0016] The knowledge graph-based dynamic generation method and system for pre-shift meeting outlines provided in this application, by conducting risk transmission path reasoning on the knowledge graph, breaks through the limitation of traditional risk identification, which can only match "point-like" risks. It can reveal deep-seated and cascading safety risks hidden under multiple apparent factors, realizing the transformation of safety management from "post-event attribution" to "pre-event prediction," greatly improving the predictability and proactivity of risk identification. Based on dynamic factors such as the specific construction tasks of the day, real-time environmental data, and specific personnel configuration, the system generates pre-shift meeting outlines in real time, ensuring the accuracy and dynamism of the briefing content. It completely changes the situation of traditional template-based briefings being out of touch with the actual scenario, significantly improving the pertinence and practical guidance value of safety briefings. The system solidifies and makes explicit industry standards, laws and regulations, and tacit expert experience into a structured knowledge graph, promoting the standardization and intelligentization of safety management. This transforms high-quality safety analysis capabilities from relying on a few senior managers into a stable, reliable, and replicable system capability, ensuring the standardization and high quality of pre-shift meetings. The system automatically generates detailed outlines and intelligently matches and pushes supporting visual multimedia materials, which greatly reduces the burden of pre-meeting preparation for team leaders, allowing them to focus on the interaction and confirmation of the briefing process. It also improves the understanding, memory, and acceptance of front-line workers through vivid and intuitive briefing methods, effectively improving the effectiveness and efficiency of briefing work and comprehensively promoting the intelligent transformation of construction site safety management. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the functional module structure of a pre-shift meeting outline dynamic generation system according to an embodiment of the present invention;

[0019] Figure 2 This is a flowchart illustrating a method for dynamically generating pre-shift meeting outlines based on knowledge graphs, according to an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. For example, the flowcharts and block diagrams in the drawings illustrate the architecture, functions, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, program segment, or part of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions. In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0024] Example 1

[0025] This embodiment details a method for dynamically generating pre-shift meeting outlines based on knowledge graphs. This method typically runs in a server / client architecture, with the core algorithm engine deployed on the server side and the client side (such as a mobile app or PC webpage) serving as the user interface. Please refer to [link to relevant documentation]. Figure 2 The method for dynamically generating pre-shift meeting outlines based on knowledge graphs includes the following steps: S201~S205.

[0026] S201: Constructing a construction safety knowledge graph;

[0027] S202: Real-time acquisition of construction scene information and vectorization activation;

[0028] S203: Execute risk transmission path reasoning;

[0029] S204: Generate differentiated pre-shift meeting outlines and push out supporting materials;

[0030] S205: Execution and feedback closed loop.

[0031] Step 201: Construct a construction safety knowledge graph.

[0032] This step is fundamental to the entire methodology and is typically completed in the initial phase of system deployment, continuously updated and iterated as new data is integrated. The goal of this step is to build a comprehensive and accurate knowledge graph model that reflects the field of construction safety.

[0033] Specifically, the system's data layer collects data from multi-source, heterogeneous databases and documents. These data sources include at least: structured data: such as construction plans (WBS decomposition) in project management systems, equipment ledgers (equipment models, maintenance records), personnel information databases (job types, qualification certificates, historical violation records), and on-site sensor data aggregated through Internet of Things (IoT) gateways (such as tower crane anemometers, foundation pit temperature and humidity sensors, gas concentration detectors, etc.). After collection, data cleaning and deduplication are required to filter out invalid data (such as expired qualification information, records of malfunctioning equipment) to ensure the accuracy and timeliness of the data; semi-structured / unstructured data: such as specialized construction plans in Word and PDF formats, safety technical briefing documents, national and industry safety production standards (such as the "Technical Specification for Safety of High-Altitude Operations in Building Construction" JGJ80-2016), internal safety management systems of enterprises, historical accident investigation reports, safety inspection records, etc. For this type of data, format standardization processing can be performed first, text extraction from scanned PDFs can be completed using OCR technology, and the document content can be structurally segmented using paragraph splitting technology to prepare for subsequent knowledge extraction.

[0034] Then, the knowledge graph construction module (see [link]) is used. Figure 1Module 110 processes the collected data. By employing Natural Language Processing (NLP) technologies, such as Named Entity Recognition (NER) and Relation Extraction (RE), core safety entities, including processes, equipment, materials, environmental factors, personnel roles, potential risks, and control measures, are automatically extracted from the preprocessed data. Simultaneously, entity disambiguation and normalization are performed on synonymous entities (such as "high-altitude operation" and "working at height") to ensure the uniqueness of the entities in the graph. Regarding the relationships between entities, in addition to accurately extracting the five core relationship types—causation, triggering, aggravation, inhibition, and response—implicit relationships can be mined by combining historical accident cases. For example, a chain causal relationship of "illegal material stacking → pressure imbalance on the pit sidewall → collapse accident" can be extracted from accident investigation reports to further enrich the knowledge density of the graph.

[0035] As an example, step 201 includes the following: the relationships between entities in the construction safety knowledge graph include: causal relationships to characterize the root causes of risks, triggering relationships to characterize new risks triggered by specific conditions, and aggravating relationships to characterize the amplification of risk levels by environmental factors. The graph reasoning algorithm traverses along one or more of the above relationships to discover potential chain risk transmission paths.

[0036] Specifically, the types of entities may include, but are not limited to: processes (such as steel beam welding, scaffolding erection), equipment (such as tower cranes, welding machines), materials (such as oxygen cylinders, insulation materials), environmental factors (such as windy weather, high temperature environment, confined space), personnel roles (such as welders, crane operators), potential risks (such as falls from heights, fires, electric shocks, falling objects), and control measures (such as wearing safety belts, providing fire extinguishers, and assigning supervisors).

[0037] The types of relationships are used to describe the complex logic between entities, including at least: Cause: indicating that one entity is the root cause of another entity, such as (improper operation) - [cause] -> (equipment damage); Trigger: indicating that the occurrence of a specific entity or condition will trigger a certain risk, such as (working at height) - [trigger] -> (risk of falling from height); Aggravate: indicating that a certain environmental factor will significantly increase the probability or severity of an existing risk, such as (strong winds) - [aggravate] -> (risk of falling from height); Inhibit / suppress: indicating that a certain control measure can effectively reduce or eliminate a risk, such as (wearing a safety belt) - [inhibit] -> (risk of falling from height); Cope_with: indicating that a certain emergency plan or equipment is used to deal with a risk event that has already occurred, such as (fire extinguisher) - [cope_with] -> (fire risk).

[0038] Finally, all extracted entities are used as nodes and relationships as edges to construct a massive, network-like construction safety knowledge graph. Based on project scale and application requirements, this graph is loaded into high-performance graph databases such as Neo4j and JanusGraph. Neo4j is suitable for rapid deployment and visualization queries in small to medium-sized projects, while JanusGraph is adapted to the distributed storage and high-concurrency query requirements of large-scale engineering projects. At the same time, a dynamic update interface for the knowledge graph is reserved to support the subsequent integration of construction feedback data and the iterative optimization of knowledge, so as to facilitate efficient graph queries and risk reasoning in the future.

[0039] Step 202: Acquire construction scene information in real time and perform vectorized activation.

[0040] This step reflects the dynamic nature of the method. Before the start of a typical workday, the team leader logs into the system via a client app. The system first automatically retrieves the team's scheduled work tasks for the day from the construction planning system via an API interface.

[0041] As an example, step S202 includes the following: collecting real-time environmental data of the work location through IoT sensors; combining the real-time environmental data with the preset process information in the daily construction task list, and vectorizing the preset process information together to activate the corresponding environmental entity nodes and process entity nodes in the knowledge graph.

[0042] To illustrate with a specific scenario: In a construction project, the core task of the welding team (3 people, all of whom are certified welders) on that day was to carry out steel beam welding work on the outer wall of the core tube on the 15th floor of the main structure.

[0043] At this time, the scene information processing module (see...) Figure 1 Module 120 in the project initiated multi-dimensional data collection and integration: At the task information level, the process was extracted from the construction plan as "steel beam welding," the work type was labeled as "high-altitude operation" or "hot work," and the work location was locked as "15th floor core tube exterior wall." Simultaneously, the corresponding basic safety requirements for the operation were retrieved. At the personnel information level, the project personnel information database was synchronized, and the validity of the welding qualifications and historical violation records of the three members of the welding team were verified. All qualifications were confirmed as "qualified," and information archiving was completed. At the environmental information level, a "IoT device priority + third-party API backup" collection strategy was adopted. Real-time data collected by the IoT anemometers and temperature and humidity sensors deployed on-site was prioritized. If a sensor malfunctioned or data was missing, a third-party meteorological service API was called to obtain the day's weather forecast (e.g., "daytime weather is 6..."). (High wind speeds and high humidity) and at the same time, the system verifies the validity of the collected data and removes abnormal values; at the temporary information level, the system provides a visual interactive interface, allowing team leaders to quickly select unplanned temporary situations through drop-down options or manually input them (such as "Flammable insulation materials are temporarily piled up below the work site"). The interface has preset options for common temporary scenarios to improve operational efficiency, while also retaining a manual input channel to adapt to special situations.

[0044] After data collection, the discrete information is first preprocessed to unify the data format and remove invalid and redundant content. Then, models such as Word2Vec or BERT are used to transform the task, personnel, environment, and temporary information into standardized scene feature vectors. During the transformation process, entity normalization is used to ensure that information such as "high-altitude operation" and "hot work" are consistent with the preset entity descriptions in the knowledge graph. Subsequently, the feature vector is matched with the entity vectors in the knowledge graph to accurately activate the corresponding nodes. Nodes such as "high-altitude operation", "hot work", "windy weather", "steel beam welding", "electric welder", and "flammable insulation materials" are marked as "activated". At the same time, the system performs preliminary correlation verification on the activated nodes and eliminates redundant nodes that are not related to the day's work.

[0045] The scene information processing module also records the data source priority and collection timestamp of each activated node, providing a basis for weight determination in subsequent risk reasoning, ensuring that the activated nodes not only fully cover key information on site, but also accurately support the next step of cross-domain correlation reasoning.

[0046] Step 203: Perform risk transmission path reasoning.

[0047] This step is the core innovation of this invention, and it is powered by a risk reasoning engine (see [link]). Figure 1Module 130 in the engine is executed. Starting from the node activated in step 202, the engine executes a multi-step, weighted graph reasoning algorithm on the knowledge graph to discover all the important risk paths that may be possible from the current scenario.

[0048] Specifically, the first step is direct risk identification. Starting from the activated "process" type nodes, the engine traverses along the -[trigger]-> relationship between entities in the knowledge graph to accurately identify surface risks strongly related to the current scenario: from the "high-altitude operation" node, it directly links to the "high-altitude fall risk" node; from the "hot work" and "steel beam welding" nodes, combined with the characteristics of electric welding operations, it derives "fire risk," "scalding risk," and "electric shock risk" (related to the electric welding machine usage scenario). At the same time, the identified direct risk nodes are initially marked to eliminate redundant risks unrelated to the day's work scenario.

[0049] The process then moves to the chain and compound risk reasoning (deep reasoning) stage, uncovering hidden and compound risk transmission chains: In the aggravation effect reasoning, the engine traverses the -[aggravate]-> relationship, discovering the reinforcing effect of the "strong wind weather" node on "high-altitude fall risk," and dynamically adjusts the correlation strength of the aggravation relationship based on real-time wind speed data, with the aggravation weight corresponding to level 6 gusts being higher than that of regular light wind weather; In the chain reaction reasoning, hidden paths are uncovered through multi-step traversal, starting from the "strong wind weather" node, passing through the "cause" relationship to the intermediate node of "welding shielding gas failure," and then pointing to "welding quality defect risk" through the "further cause" relationship, filling the safety-quality cross-risk loophole that is easily overlooked by human experience; In the compound risk reasoning, the engine matches the multi-node joint triggering rules preset in the knowledge graph, identifies the "handheld tool fall risk" triggered by the combined effect of "high-altitude operation" and "strong wind weather," and then deduces along the causal relationship to "object strike risk," forming a complete compound risk chain.

[0050] After inference is complete, the engine initiates the risk path quantification and ranking process, which can calculate the comprehensive risk value (Rpath) of each path based on the following weighted cumulative algorithm:

[0051]

[0052] Among them, R path B represents the comprehensive risk value of the risk transmission path; B is the basic risk coefficient, which can be adjusted according to the type of operation (e.g., the coefficient for hot work is higher than that for routine operations); Nodes(path) is the set of all entity nodes on the risk transmission path; w iThe pre-defined inherent risk weights for node i are defined. For example, the weight w of the risk node "fall from height" will be much higher than the weight of "welding quality defects." These weight values ​​can be pre-defined by safety experts or obtained by analyzing historical accident data. Edges (path) is the set of all relational edges on the risk transmission path. j The p-value is the preset transmission probability or confidence level for relation edge j. For example, the p-value of the edge (strong wind weather) - [aggravate] -> (risk of falling from height) will dynamically increase as the wind force level increases.

[0053] After calculation using the formula, the engine sorts all risk paths in descending order of comprehensive risk value, eliminates duplicate paths and extremely low-risk paths, and finally outputs a graded risk list (e.g., fall from height risk (high risk), fire risk (medium risk), object strike risk (medium risk), welding quality defect risk (general), electric shock risk (general)). At the same time, it associates the transmission logic and key influencing factors of each path, providing priority support and content basis for the accurate generation of the subsequent pre-shift meeting outline.

[0054] Step 204: Generate a differentiated pre-shift meeting outline and send out supporting materials.

[0055] Specifically, the outline generation module (see...) Figure 1 After receiving the sorted risk list, module 140 prioritizes generating the risks corresponding to the high-risk paths, their transmission logic, and the corresponding control measures extracted from the knowledge graph in a prominent position in the pre-shift meeting outline template based on the risk sorting results.

[0056] The descriptions were optimized to better reflect real-world scenarios: For the high-risk "fall from height risk," control nodes related to the -[inhibit]-> relationship were retrieved from the knowledge graph and transformed into concrete measures such as "the team leader must 100% check the integrity of safety belts, safety ropes, and hooks before work," "strictly implement the 'high-hanging, low-use' principle, and strictly prohibit safety ropes from being tangled or knotted," and "immediately stop high-altitude work if the instantaneous wind force exceeds level 6." Simultaneously, a risk warning analysis was added: "Today there are level 6 gusts at the site; personnel standing at height are unstable, and the fall risk level is extremely high." For the medium-risk "fire risk," considering the temporary scenario of flammable insulation materials below the work site, the following measures were extracted: "clear flammable materials within a 5-meter radius below and cover them with non-combustible materials," "set up a fire catcher," "equip with two 8 kg ABC dry powder fire extinguishers," and "assign a dedicated person to monitor the entire process." Measures such as these are used to clarify the requirements for scenario-based management and control; in response to the hidden chain risks of "welding quality defects", personalized special reminders are generated such as "There is a strong wind today. When welding, pay attention to the effect of the protective gas and take windproof measures when necessary (such as building a temporary windproof shed). Check the weld formation in real time to prevent quality problems such as porosity and slag inclusions."

[0057] After the content is filled in, the module is sorted out and output according to a standardized structure. The overall framework covers the meeting time and location, participants and work teams, the main work tasks for the day (specifically, welding of the steel beams of the outer wall of the 15th-floor core tube), core risk warning and control measures (described in sections according to risk level, with each risk accompanied by warning analysis and implementable measures), emergency response plan (supplementing the initial response procedures and emergency contact methods for falls from heights and fires), and special precautions (summarizing personalized risk reminders), ensuring a clear structure and highlighting key points, making it easy for team leaders to quickly explain.

[0058] The system initiates an intelligent process for pushing supporting materials, using key risk entities such as "fall from height," "hot work," "safety belts," and "fire extinguishers" as search keywords. It accurately matches suitable materials from a pre-set multimedia database, prioritizing content that is 3-5 minutes long and highly practical. This includes short videos on "The Correct Use of Safety Belts," dynamic illustrations on "Fire Extinguisher Inspection and Usage Procedures," example images on "Fall Prevention Measures for Tools Used in High-Altitude Operations," and teaching segments on "Welding Wind Protection Operation Specifications." These materials are pushed to the team leader's client APP interface simultaneously with the outline, in the form of links that can be directly projected or high-definition thumbnails. This allows for one-click projection and playback during pre-shift meetings, replacing dry text explanations with intuitive and vivid visuals. This strengthens workers' memory and understanding of risks and control measures, improving the actual effectiveness of safety briefings.

[0059] Step 205: Execution and feedback loop.

[0060] A user interface can be provided to receive feedback from team leaders regarding adjustments or confirmations to the generated pre-shift meeting outline; and the feedback information can be used to dynamically adjust the weights of entity nodes or the confidence of relationships between entities in the knowledge graph, thereby achieving continuous optimization and learning of the knowledge graph model.

[0061] Specifically, when team leaders hold pre-shift meetings on-site, they conduct safety briefings based on the structured outlines generated by the system and the visual materials pushed to them. At the same time, they can flexibly adjust and supplement the content of the outline according to the unexpected situation on site. Taking welding team operations as an example, if it is found that the power cord of the welding machine temporarily laid on site is damaged or the insulation layer is aged during the briefing, the team leader can quickly and manually add the risk item through the "Temporary Risk Supplement" module of the client APP. The interface has preset risk classification labels (such as "Temporary Power Risk"), supports one-click selection and brief filling in of control suggestions (such as "Replace the damaged power cord immediately, check the insulation of the line before operation"). The system saves the modified content in real time and synchronizes it to the outline copy to ensure that the briefing content is consistent with the actual situation on site.

[0062] After the meeting, the team leader clicked the "Disclosure Completed" button on the APP to complete the electronic signature confirmation. The system automatically archived all process data, including the final outline, manually added content, multimedia material usage records, meeting duration, and electronic signature ledger of participants. Among them, the manually added "Power cord damage risk" was marked as high-value on-site feedback data and stored in the dedicated feedback database as "Pending Review / Pending Analysis".

[0063] The system then initiates a dual-path feedback processing mechanism: In the semi-automatic learning chain, the system summarizes all on-site feedback data at fixed intervals (e.g., daily / weekly), filters out manually added new risks and control measures, and pushes them to safety engineers for review. The engineers determine the effectiveness of the feedback based on industry standards and the actual project situation. If it is confirmed that the "power line damage risk" is strongly correlated with the hot work scenario, a "power line damage risk" entity is added to the knowledge graph, associated with nodes such as "hot work" and "temporary power supply," and control measures such as "checking line insulation" and "replacing damaged cables" are added to complete the manual-assisted update of the knowledge graph.

[0064] In the automatic learning process, the system performs cluster analysis and frequency statistics on the feedback data using a preset algorithm, and sets a reasonable feedback frequency threshold (such as when 3 or more work groups in the same work scenario all supplement the same type of risk). When the threshold is reached, the knowledge graph parameters are automatically adjusted. For example, if multiple work groups supplement "equipment leakage risk" when working in the rain, the system will automatically increase the confidence (pj value) of the relationship (rainy day) - [trigger] -> (equipment leakage risk) and update the calculation logic of the inference engine.

[0065] Through the above operations, the entire technical solution forms a complete closed loop from knowledge graph construction, scene perception reasoning, outline generation and push, to on-site feedback and optimization. This allows the system to continuously absorb on-site practical experience, gradually reduce the frequency of manual supplementation, and continuously improve the accuracy and intelligence level of subsequent safety briefings, thus consolidating the foundation for long-term safety management.

[0066] Example 2

[0067] This embodiment provides a device for dynamically generating pre-shift meeting outlines. Please refer to [link / reference]. Figure 1 This device is the hardware and software entity that implements the above-described methods. It can be one or more servers, or a functional system embedded in a larger security management platform. Structurally, the device mainly includes: a knowledge graph construction module 110, a scene information processing module 120, a risk reasoning engine 130, and an outline generation module 140. In addition, the device also includes basic hardware components such as a processor, memory, and communication interfaces.

[0068] Knowledge Graph Construction Module 110: This module can physically consist of a set of software programs running on a server. It is responsible for executing all the functions of step 201 in Embodiment 1. Internally, this module can be further divided into a data acquisition unit, a knowledge extraction unit, and a graph storage unit. The data acquisition unit connects to external data sources (such as BIM systems and ERP systems) through a configured data interface; the knowledge extraction unit integrates an NLP algorithm library for processing text data; and the graph storage unit directly interacts with the graph database, performing add, delete, modify, and query operations on the graph. This module typically runs periodically in the background to maintain the timeliness of the knowledge graph.

[0069] Scene Information Processing Module 120: This module serves as the portal for real-time interaction between the system and the outside world. It receives user input and task information from client apps or web applications via communication interfaces (such as RESTful APIs) and subscribes to real-time data from field sensors via IoT interfaces (such as the MQTT protocol). At the core of this module is a vectorization processor that converts the received multi-source, heterogeneous scene information into standardized digital signals (feature vectors) that the risk inference engine 130 can understand, thereby activating corresponding nodes in the knowledge graph.

[0070] Risk Reasoning Engine 130: This is the "brain" of the entire device, its functions implemented by a complex algorithm program running on a high-performance processor. It receives activation signals from the scene information processing module 120 and then performs graph traversal and path discovery algorithms on the knowledge graph loaded in memory. The risk value quantification calculation formula described in Embodiment 1 is embedded within this engine, enabling rapid and accurate prioritization of multiple discovered risk paths. The performance of this engine directly determines the system's response speed from receiving scene information to outputting a risk list.

[0071] Outline Generation Module 140: This module receives the risk ranking results output by the risk reasoning engine 130. Internally, it includes a content management system and a template engine. The content management system is responsible for storing and managing text descriptions and multimedia material links associated with each risk entity and control measure entity. The template engine is responsible for dynamically and structurally populating this content into a preset pre-shift meeting outline format and generating the final report file (such as HTML, PDF, or JSON format), which is then sent back to the user client via the communication interface.

[0072] In this embodiment, the aforementioned modules work collaboratively. When the team leader requests the generation of an outline on the mobile app, the request is sent via the network to the scene information processing module 120, which processes it and then transmits an activation signal to the risk reasoning engine 130. After completing the reasoning calculation on the knowledge graph, the engine 130 sends the result to the outline generation module 140, which generates the final outline and returns it to the app. The entire process can be completed within seconds, providing efficient decision support for on-site management.

[0073] Example 3

[0074] This embodiment, building upon Embodiment 1, further introduces a refined management mechanism based on individual personnel differences. In traditional safety management, risk identification is often limited to the dimensions of "events" and "objects," such as considering only the risk of "high-altitude operations" themselves, while ignoring the differences in the condition of the "people" performing the work. In fact, the same working environment can result in significantly different risk levels for workers with different skill levels and physical conditions. Therefore, this embodiment proposes a multimodal knowledge graph reasoning method that integrates personnel profile data.

[0075] First, in the knowledge graph construction phase, this embodiment adds a personnel profile sub-graph. This sub-graph uses specific workers as core entity nodes and associates two types of attribute data. The first type is static attribute data, mainly from the enterprise's human resource management system and training and assessment system, specifically including personnel's age, job type, certification status, skill level rating, and historical violation records. The second type is dynamic attribute data, mainly from the construction site's real-name gate system and smart wearable devices (such as smart bracelets and smart safety helmets), specifically including personnel's daily entry time, previous night's sleep duration, real-time heart rate, blood pressure trend, and physiological indicators such as body surface temperature. After cleaning and standardizing the above multi-source data, the system maps it to the attribute values ​​of personnel entities in the knowledge graph, thereby constructing a dynamically updated digital twin personnel model.

[0076] Secondly, during the scenario information acquisition and activation phase, the system not only acquires environmental and process information but also simultaneously obtains the daily shift schedule and its real-time status data. For example, when the system identifies the daily task as "removal of support beams in deep foundation pits" and the ambient temperature as "35 degrees Celsius," it will simultaneously read the physiological data of all shift members involved in the task. If it detects that a shift member's real-time heart rate is consistently higher than the normal baseline, or that their sleep duration last night was less than 5 hours, the system will mark the entity node corresponding to that person in the knowledge graph as "high load state" or "fatigue state," and use it as a new input source for graph reasoning.

[0077] Subsequently, the risk reasoning engine performs cross-domain correlation reasoning. Unlike the single "environment-process" reasoning in Example 1, this example performs "human-machine-environment" ternary feature fusion reasoning. Specifically, the reasoning process includes calculations in two core dimensions:

[0078] The first aspect is the calculation and reasoning of the person-job fit. The system searches the graph for the skill level attribute of the personnel entity and the difficulty level attribute of the current process entity. If it finds that a worker is a "junior worker" or "newly hired" and the process assigned to him is a "high-risk and complex operation" (such as high-pressure welding), the reasoning algorithm will determine that the person-job fit is lower than the preset safety threshold, thereby triggering the "skill deficiency risk" node in the graph and deriving potential "operational error" or "process defect" risk paths.

[0079] Secondly, there is the calculation and reasoning of environmental adaptability. The system jointly analyzes the physiological state attributes of personnel and the conditional attributes of environmental entities. For example, in the aforementioned high-temperature deep foundation pit operation scenario, for a worker with normal physical indicators, the graph reasoning might only result in a routine heatstroke risk; however, for workers marked as "fatigued" or with a "history of hypertension," the reasoning engine will search along specific pathological aggravation paths, activating high-risk nodes such as "sudden fainting risk" or "cardiovascular and cerebrovascular accident risk." This reasoning mechanism refines the granularity of risk management from the "team level" to the "individual level."

[0080] Finally, when generating the pre-shift meeting outline, the system will generate differentiated control content based on the above reasoning results. The generated outline not only includes general work safety requirements but also adds a section on "Key Personnel and Targeted Measures." For example, the outline will clearly state: "Given today's high temperature, monitoring data shows that team members Zhang and Li have slightly decreased physical function indicators. It is recommended that the team leader avoid assigning these personnel to high-intensity climbing operations and increase their rest frequency." In this way, the present invention can assist team leaders in rationally allocating human resources, eliminating potential safety hazards at the outset, and truly realizing people-oriented intelligent safety management.

[0081] Example 4

[0082] This embodiment focuses on a feedback-driven adaptive evolution mechanism for knowledge graphs. Considering the complex and ever-changing nature of construction site environments and the continuous emergence of new technologies and equipment, pre-built knowledge graphs inevitably suffer from incomplete coverage or outdated weights. To address this technical bottleneck, this embodiment designs a closed-loop feedback module in the system, using actual execution data from daily pre-shift meetings to reverse-optimize the knowledge graph model, enabling it to continuously learn and self-evolve.

[0083] Specifically, the mechanism operates as follows:

[0084] The first step is the collection of multimodal feedback data. The system provides a structured feedback interface in the client application, allowing team leaders to submit data after pre-shift meetings or at the end of each day. Feedback data mainly includes three categories: first, missed risk events, i.e., a certain type of hidden danger actually occurred on site (e.g., "slippery slope after rain"), but was not mentioned in the system-generated outline; second, false risk events, i.e., the system indicated a high-risk event (e.g., "deep foundation pit collapse"), but actual on-site monitoring data showed that the working conditions were extremely stable and the risk did not materialize; and third, effective control measures, i.e., temporary and effective safety measures taken by the team leader on site (e.g., "laying anti-slip mats"). In addition, the system also supports voice input, allowing team leaders to verbally record special situations on site, which the system automatically converts into text data.

[0085] The second step is parameter optimization based on the Bayesian update algorithm. For existing entity relationship paths in the knowledge graph, the system uses statistical methods to quantitatively analyze the feedback data. The system maintains the confidence weight of each relationship edge in the graph. When the system receives a large number of "valid" or "occurrence" feedbacks about a certain risk path, the algorithm will use the Bayesian formula to calculate the posterior probability, thereby increasing the weight of that path and making it easier to activate and rank it higher in future inferences. Conversely, if a risk path is frequently marked as a "false alarm," the algorithm will automatically reduce its weight or set a higher activation threshold to reduce invalid interference to users and improve the system's accuracy.

[0086] The third step is new knowledge mining and graph completion based on natural language processing. For unstructured descriptive information submitted by team leaders via text or voice, the system first uses word segmentation and named entity recognition (NER) technology for preprocessing to extract potential entity nouns (such as "grass mat" and "ramp") and action descriptions (such as "laying" and "slipping"). Next, the system uses clustering analysis algorithms to mine the feedback data accumulated over a period of time. If it finds that the frequency of a specific keyword combination (such as "rainy day" and "slippery ramp") in the feedback from different teams and at different time periods exceeds a preset significance threshold, the system will determine that a new causal logic has been discovered.

[0087] At this point, the knowledge graph construction module will automatically execute an incremental update procedure, generating new risk entity nodes (such as "slippery risk") and control measure nodes (such as "laying anti-slip mats") in the knowledge graph, and establishing triggering relationship edges from environmental nodes (such as "rainfall") to the new risk nodes. Through this collective intelligence evolution, the system can quickly transform the personal experience of an experienced team leader into explicit knowledge shared by the system. As usage time increases, the system will accumulate increasingly rich on-site practical experience, and the pre-shift meeting outlines it generates will become increasingly aligned with the actual needs of frontline construction, thus forming a virtuous cycle of intelligent ecosystem.

[0088] In the embodiments provided in this application, it should be understood that the disclosed systems and devices can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some communication interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0089] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0090] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0091] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0092] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A knowledge graph-based pre-shift meeting outline dynamic generation method, characterized in that, include: Based on multi-source data including construction plans and historical accident cases, a construction safety knowledge graph containing safety entities and the relationships between entities is pre-constructed. The current construction scenario information is obtained in real time through the information interface, and the scenario information is vectorized to activate the entity nodes corresponding to the scenario information in the construction safety knowledge graph. Starting from the activated entity node, a graph reasoning algorithm is executed on the construction safety knowledge graph. By traversing the preset relationship paths, the potential chain risk transmission path is derived from the direct risks caused by the current construction scenario. Based on the direct risks and the risk transmission paths, control measures associated with each risk entity in the direct risks and the risk transmission paths are extracted from the construction safety knowledge graph, and a structured pre-shift meeting outline is automatically generated.

2. The method according to claim 1, characterized in that, The step of acquiring current construction scene information in real time through an information interface and vectorizing the scene information to activate entity nodes corresponding to the scene information in the construction safety knowledge graph includes: Real-time environmental data of the work location is collected through IoT sensors; By combining the preset process information in the daily construction task list, the real-time environmental data and the preset process information are vectorized together to activate the corresponding environmental entity nodes and process entity nodes in the knowledge graph.

3. The method according to claim 1, characterized in that, The relationships between entities in the construction safety knowledge graph include: causal relationships to characterize the root causes of risks, triggering relationships to characterize new risks triggered by specific conditions, and aggravating relationships to characterize the amplification of risk levels by environmental factors. The graph reasoning algorithm traverses along one or more of the above relationships to discover the risk transmission path.

4. The method according to claim 1, characterized in that, Before the pre-shift meeting outline is generated, the process also includes: quantifying and ranking the risk values ​​of the multiple risk transmission paths derived from reasoning. Accordingly, the generation of a structured pre-shift meeting outline includes: based on the sorting results, prioritizing the generation of risks, control measures, and transmission logic corresponding to high-risk value paths and placing them in a prominent position on the pre-shift meeting outline.

5. The method according to claim 4, characterized in that, The quantitative calculation of the risk value includes: The preset risk weights of each entity node constituting the risk transmission path and the preset transmission probabilities of each relation edge connecting the entity nodes are comprehensively evaluated. The comprehensive risk value for each of the aforementioned risk transmission paths is obtained through a weighted cumulative algorithm.

6. The method according to claim 1, characterized in that, After the pre-shift meeting outline is generated, the process also includes: using the key risk entities identified in the risk transmission path as search keywords, automatically matching and retrieving corresponding visual safety education materials from a preset multimedia database, and associating and pushing the visual safety education materials with the pre-shift meeting outline.

7. The method according to claim 1, characterized in that, The method further includes: establishing a personnel profile sub-graph, which includes personnel entities as graph nodes, as well as static and dynamic attributes associated with the personnel entities; wherein, the static attributes include job type, qualification, skill level and historical violation records, and the dynamic attributes include real-time physiological state characteristics obtained through wearable devices. Accordingly, the list of personnel scheduled for the day and their real-time physiological status characteristics are obtained, and the list of personnel and physiological status characteristics are mapped to the personnel profile sub-graph to activate the corresponding personnel entity nodes.

8. The method according to claim 7, characterized in that, The method further includes: performing cross-domain association reasoning, and performing feature matching calculations between the activated personnel entity nodes and the activated process entity nodes and environment entity nodes respectively; The reasoning algorithm is configured to calculate the person-job matching degree between the personnel entity node and the process entity node. When the person-job matching degree is lower than a preset safety threshold, the intermediate node representing the risk of insufficient skills is activated in the knowledge graph. The reasoning algorithm is also configured to calculate the physiological fitness between the personnel entity node and the environmental entity node. When the physiological fitness shows an abnormality and the corresponding environmental entity attribute is severe weather conditions, a specific risk node representing the decline in the physiological function of a specific person is activated in the knowledge graph. Accordingly, for the identified specific risk nodes, a targeted early warning prompt containing a list of specific personnel is generated in the pre-shift meeting outline.

9. The method according to claim 1, characterized in that, The method further includes dynamically adjusting the weights of entity nodes or the confidence levels of relationships between entities in the knowledge graph: Collect actual construction feedback data after the pre-shift meeting. The feedback data includes missed risk events, falsely reported risk events, and effective control measures temporarily taken on site. Using the Bayesian update algorithm, the confidence weights of the connection edges between entity nodes in the knowledge graph are updated with posterior probabilities based on the feedback data, so as to strengthen the verified risk transmission path and weaken the false alarm path. Natural language processing and cluster analysis are performed on the text description information in the feedback data. When the frequency of a specific keyword exceeds a preset threshold, new risk entity nodes and related edges are automatically generated in the knowledge graph to achieve incremental updates of the knowledge graph structure.

10. A device for dynamically generating pre-shift meeting outlines, characterized in that, include: The knowledge graph construction module is used to pre-build a construction safety knowledge graph containing safety entities and the relationships between entities based on multi-source data including construction plans and historical accident cases. The scene information processing module is used to obtain the current construction scene information in real time through the information interface, and to vectorize the scene information in order to activate the entity nodes corresponding to the scene information in the construction safety knowledge graph. The risk reasoning engine is configured to start from the activated entity node and execute a graph reasoning algorithm on the construction safety knowledge graph. By traversing the preset relationship paths, it derives the potential chain risk transmission path from the direct risks caused by the current construction scenario. The outline generation module is used to extract control measures associated with each risk entity in the direct risk and the risk transmission path from the construction safety knowledge graph based on the direct risk and the risk transmission path, and automatically organize and generate a structured pre-shift meeting outline.

Images