Construction safety risk intelligent control system and method based on knowledge graph

Abstract

The invention discloses a construction safety risk intelligent control system and method based on a knowledge graph, and the system comprises a data collection layer which is used for collecting construction model data, process parameters, real-time monitoring data and historical risk event data of a construction site; the multi-library collaborative knowledge graph layer is used for constructing a knowledge graph based on the data acquired by the data acquisition layer; the intelligent analysis processing layer is used for analyzing the association network of the scheme, the risk and the hidden danger and generating risk early warning information, an optimized construction scheme and a quantitative evaluation report; and the visual decision-making layer is used for receiving and displaying the risk early warning information, the optimized construction scheme and the quantitative evaluation report which are generated by the intelligent analysis processing layer. According to the method, deep and cross-professional coupling risks can be mined, automatic and quantitative optimization of the construction scheme is realized, the dynamic response capability to field changes is enhanced, and the scientificity and efficiency of power transmission construction risk management and control are remarkably improved.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of power transmission engineering and artificial intelligence, and in particular to a knowledge graph-based intelligent control system and method for construction safety risks. Background Technology

[0002] Power transmission projects, especially the construction of high-voltage and ultra-high-voltage transmission lines, are critical infrastructure projects for ensuring energy security. Their construction typically takes place in complex geographical and meteorological environments, involving numerous high-risk operations such as deep foundation pit excavation, high-altitude tower erection, and long-span line laying. Therefore, the management and control of construction safety risks is of paramount importance.

[0003] Currently, risk management in power transmission construction still largely relies on the personal experience of project managers and safety engineers. The preparation and review of construction plans, as well as the identification and assessment of on-site risks, are primarily conducted through traditional methods such as reviewing paper regulations, holding expert review meetings, and conducting on-site inspections. This model has inherent limitations. On the one hand, the comprehensiveness of risk identification is limited by the scope of individual knowledge and experience, making it easy to overlook some uncommon or deep-seated risks caused by the coupling of multiple factors. On the other hand, construction sites are dynamically changing; real-time changes in environmental factors, personnel status, and equipment conditions can generate new risks, and the reliance on periodic inspections often leads to delayed risk response, failing to achieve real-time early warning and dynamic adjustment.

[0004] Furthermore, when developing construction plans, there is a lack of objective and quantifiable standards for evaluating the merits of different plans. Decisions are often based on qualitative judgments, making it difficult to systematically weigh and optimize multiple objectives such as safety, economy, and schedule. A large amount of historical project data, accident reports, and successful experiences are scattered in various archives in unstructured text form, failing to be effectively utilized. This results in low efficiency in knowledge transfer and reuse, leading to the recurrence of similar problems in different projects.

[0005] Therefore, how to systematically integrate multi-source heterogeneous construction data, deeply mine the risk knowledge contained in the data, and based on this, realize the comprehensive, dynamic, and intelligent identification, assessment, and control of construction risks is a technical problem that urgently needs to be solved in the field of power transmission engineering safety management. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this application provides a knowledge graph-based intelligent control system and method for construction safety risks, which improves the comprehensiveness and depth of risk identification through dynamic, quantitative, and forward-looking control of power transmission construction risks.

[0007] To achieve the above objectives, this application provides the following technical solution: The first aspect of this application provides a knowledge graph-based intelligent control system for construction safety risks, comprising: The data acquisition layer is used to collect construction model data, process parameters, real-time monitoring data, and historical risk event data from the construction site. A multi-database collaborative knowledge graph layer constructs a knowledge graph based on the data collected by the data acquisition layer. The intelligent analysis and processing layer is used to analyze the correlation network of plans, risks, and hidden dangers, and generate risk warning information, optimized construction plans, and quantitative assessment reports. The visualization decision-making layer receives and displays the risk warning information, optimized construction plans, and quantitative assessment reports generated by the intelligent analysis and processing layer.

[0008] Furthermore, the knowledge graph includes: a risk database, a process knowledge database, a hidden danger database, and a construction plan database. The risk database, process knowledge database, hidden danger database, and construction plan database establish a two-way dynamic mapping relationship through a preset entity relationship model, and form an association network of plans, risks, and hidden dangers.

[0009] Furthermore, the data acquisition layer includes: The BIM model data extraction unit is used to extract geometric and attribute information from the construction model data. A sensor data acquisition unit is used to acquire the process parameters; A real-time monitoring data acquisition unit is used to acquire the real-time monitoring data; The historical data mining unit is used to process the historical risk event data.

[0010] Furthermore, the intelligent analysis and processing layer includes: The risk identification module is used to identify potential risks and generate the risk warning information. The scheme optimization module is used to generate the optimized construction scheme; The scoring algorithm module is used to calculate the quantitative evaluation report.

[0011] Furthermore, the intelligent analysis and processing layer also includes: The 4D simulation module integrates the construction progress time dimension to dynamically simulate the construction process and output simulation results.

[0012] Furthermore, the visualization decision layer includes: The risk warning unit is used to display the risk warning information; The scheme evaluation report unit is used to display the quantitative evaluation report; The simulation demonstration unit is used to visualize the construction process.

[0013] Furthermore, the method for establishing the bidirectional dynamic mapping relationship is as follows: The risk database and the process knowledge base are linked through process step coding. The process knowledge base and the hidden danger database are linked through risk factor identification; The hazard database and the construction plan database are linked through response measure codes; The construction plan library and the risk library are linked through a plan type identifier.

[0014] Furthermore, the risk identification module uses a graph neural network algorithm to identify risks based on the network of relationships between the scheme, risk, and potential hazard. The scheme optimization module uses a genetic algorithm to find the best construction scheme, and the genetic algorithm uses a roulette wheel selection strategy. The scoring algorithm module is used to calculate the quantitative evaluation report, which includes multi-dimensional scores. The multi-dimensional scoring includes the risk matching degree of the solution, the effectiveness of the response to potential hazards, and the rationality of the schedule.

[0015] According to a second aspect of this application, a knowledge graph-based intelligent control method for construction safety risks is provided, applied to the aforementioned knowledge graph-based intelligent control system for construction safety risks, comprising the following steps: S1. Collect construction model data, process parameters, real-time monitoring data and historical risk event data at the construction site through the data acquisition layer; S2. Based on the data collected in step S1, construct and maintain a knowledge graph consisting of a risk database, a process knowledge database, a hidden danger database, and a construction plan database, and form an association network of plans, risks, and hidden dangers. S3. Analyze the network of relationships between the proposed solutions, risks, and hidden dangers to generate risk warning information, optimized construction plans, and quantitative assessment reports. S4. Output and display the risk warning information, the optimized construction plan, and the quantitative assessment report generated in step S3.

[0016] Furthermore, the step of analyzing the network of relationships among the formed solutions, risks, and hidden dangers also includes: Risk identification is performed using graph neural network algorithms; A genetic algorithm is used to optimize the construction scheme, and the genetic algorithm adopts a roulette wheel selection strategy.

[0017] According to a third aspect of this application, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the computer program to implement a knowledge graph-based intelligent control method for construction safety risks as described above.

[0018] According to a fourth aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements a knowledge graph-based intelligent control method for construction safety risks as described above.

[0019] This application provides a knowledge graph-based intelligent control system, method, electronic device, and storage medium for construction safety risks. It offers the following advantages: 1. Improved the comprehensiveness and depth of risk identification. This application constructs a multi-database collaborative knowledge graph layer consisting of a risk database, a process knowledge base, a hidden danger database, and a construction plan database, forming a relational network of plans, risks, and hidden dangers, thus structuring discrete construction data. Based on this network, the risk identification module can uncover potential relationships between different entities, thereby identifying deep-seated risks that are easily overlooked in traditional manual inspections, such as those that are cross-disciplinary or caused by the coupling of multiple factors, reducing the probability of risk omission.

[0020] 2. Automated optimization and quantitative evaluation of construction plans have been achieved. This application utilizes a plan optimization module (such as a genetic algorithm) and a scoring algorithm module in the intelligent analysis and processing layer to replace the traditional manual decision-making process. The system can automatically optimize based on a preset objective function, generating construction plans that take into account multiple factors such as safety and schedule. The scoring algorithm module then performs multi-dimensional quantitative evaluation of the plans, providing objective and quantifiable data for decision-making and improving the efficiency and scientific rigor of plan formulation.

[0021] 3. Enhanced dynamic response capability for risk management. This application continuously acquires real-time monitoring data from the construction site through the data acquisition layer. This data can dynamically update the multi-database collaborative knowledge graph layer. When site conditions change, the intelligent analysis and processing layer can re-analyze based on the updated knowledge graph and generate risk warning information in real time. This closed-loop workflow from data acquisition to analysis and decision-making makes risk management no longer a static, one-off activity, but rather capable of dynamic adjustment as the construction progresses. Attached Figure Description

[0022] Figure 1 This is a framework diagram of the knowledge graph-based intelligent control system for construction safety risks proposed in this application. Figure 2 This is a schematic diagram of the data acquisition layer framework of this application; Figure 3This is a schematic diagram of the multi-database collaborative knowledge graph layer framework of this application; Figure 4 This is a schematic diagram of the intelligent analysis and processing layer framework of this application; Figure 5 This is a schematic diagram of the risk identification algorithm flow of this application; Figure 6 This is a schematic diagram of the optimized solution process for this application; Figure 7 This is a schematic diagram of the calculation process for the evaluation report of this application; Figure 8 This is a schematic diagram of the system operation of this application. Detailed Implementation

[0023] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0024] It should be understood that the steps described in the method embodiments of this application may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this application is not limited in this respect.

[0025] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0026] It should be noted that the terms "one" and "multiple" used in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless explicitly stated otherwise in the context, they should be understood as "one or more". "Multiple" should be understood as two or more.

[0027] Example 1 One embodiment of this application provides a knowledge graph-based intelligent control system for construction safety risks. Figure 1 This is a framework diagram of the knowledge graph-based intelligent control system for construction safety risks proposed in this application, such as... Figure 1 As shown, the knowledge graph-based intelligent control system for construction safety risks provided in this application includes: a data acquisition layer, a multi-database collaborative knowledge graph layer, an intelligent analysis and processing layer, and a visualization decision-making layer, wherein... The data acquisition layer is configured to collect various source data from the construction site, specifically including construction model data, process parameters, real-time monitoring data, and historical risk event data. The data acquisition layer serves as the system's data input, providing the data foundation for subsequent knowledge graph construction and intelligent analysis.

[0028] like Figure 2 As shown in this embodiment, the data acquisition layer includes: a BIM model data extraction unit, a sensor data acquisition unit, a real-time monitoring data acquisition unit, and a historical data mining unit. The BIM model data extraction unit extracts structured construction model data from the Building Information Model (BIM), including the three-dimensional geometric information and attribute information such as material and labeling of entities like transmission lines, towers, and foundations. The sensor data acquisition unit collects structured process parameters, such as tower material stress, conductor sag, wind speed, and temperature, using various sensors deployed at the construction site (e.g., stress sensors, displacement sensors, and meteorological sensors). The real-time monitoring data acquisition unit acquires unstructured real-time monitoring data, such as video streams of the construction area and the location coordinate sequences of construction personnel, by interfacing with on-site video monitoring systems and personnel positioning systems. The historical data mining unit extracts and structures historical risk event data, including event descriptions, occurrence times, consequences, and cause analyses, by accessing historical project databases and accident report databases.

[0029] The multi-database collaborative knowledge graph layer connects to the data output of the data acquisition layer. For example... Figure 3 As shown in the embodiment of this application, the multi-database collaborative knowledge graph layer is used to construct and maintain a knowledge graph composed of a risk database, a process knowledge database, a hidden danger database, and a construction plan database based on various types of data collected by the data acquisition layer. This knowledge graph uses entities, relationships, and attributes as basic units to structurally store knowledge in the field of power transmission construction.

[0030] In this knowledge graph, the risk database stores information such as risk classification, level, and triggering conditions. The process knowledge database stores information such as standard construction procedures, resources required for each step, and construction period. The hazard database stores specific descriptions of safety hazards and their corresponding risk factors. The construction plan database stores historical or standard construction plans, including the processes used and countermeasures.

[0031] In this embodiment, the risk database, process knowledge base, hidden danger database, and construction plan database establish bidirectional dynamic mapping relationships through a preset entity relationship model, forming a network linking plans, risks, and hidden dangers. Specifically, the bidirectional dynamic mapping relationships are established as follows: the risk database and process knowledge base are linked through process step coding; the process knowledge base and hidden danger database are linked through risk factor identification; the hidden danger database and construction plan database are linked through countermeasure coding; and the construction plan database and risk database are linked through plan type identification.

[0032] The intelligent analysis and processing layer connects to the multi-database collaborative knowledge graph layer. This layer is used to analyze the relationship network between solutions, risks, and potential hazards, and outputs the analysis results. The intelligent analysis and processing layer performs the core computational tasks of this system, generating risk warning information, optimized construction plans, and quantitative assessment reports.

[0033] like Figure 4 As shown in this embodiment, the intelligent analysis and processing layer includes a risk identification module, a scheme optimization module, and a scoring algorithm module. The risk identification module uses a graph neural network algorithm to classify nodes and predict links in the network of relationships between schemes, risks, and hidden dangers, in order to identify potential risks and generate risk warning information. The scheme optimization module uses a genetic algorithm, taking the construction scheme as the optimization object, and iteratively generates an optimized construction scheme by defining a fitness function and performing operations such as selection, crossover, and mutation. The scoring algorithm module is used to calculate a quantitative evaluation report. This report includes multi-dimensional scores, including the scheme risk matching degree, the effectiveness of hidden danger response, and the rationality of the schedule. The scheme risk matching degree score is calculated using the following formula: ; in: Representative solution risk matching score; This represents the risk coverage ratio, which is calculated as the ratio of the number of risk items for which countermeasures are already included in the construction plan to the total number of identified risk items. Represents the risk level weight, which is calculated as the ratio of the sum of the level scores of the covered risk items to the sum of the level scores of all identified risk items; and The preset weighting coefficients, In one embodiment, Set to 0.6. Set to 0.4.

[0034] In another embodiment, the intelligent analysis and processing layer also includes a 4D simulation module. This module combines the construction plan with the time dimension (the fourth dimension), integrates construction progress information, dynamically simulates the construction process, and outputs information such as resource conflicts and safety distance conflicts during the simulation process as analysis results.

[0035] The visualization decision-making layer connects to the output of the intelligent analysis and processing layer. This layer receives and displays risk warning information, optimized construction plans, and quantitative assessment reports to user terminal devices in the form of a graphical interface.

[0036] In one embodiment, the visualization decision-making layer includes: a risk warning unit, a solution evaluation report unit, and a simulation display unit. The risk warning unit displays risk information in the form of highlights and alarm lists. The solution evaluation report unit displays the evaluation report in the form of charts and scorecards. The simulation display unit receives and plays a dynamic construction process generated by the 4D simulation module in the form of a 3D animation. (See appendix) Figure 1 and attached Figure 2 The present application discloses a knowledge graph-based intelligent control system for construction safety risks. Its data acquisition layer is the data acquisition entry point of the entire system. It is responsible for acquiring structured, semi-structured and unstructured data related to power transmission construction sites from multiple heterogeneous sources, and preprocessing and formatting these data for use by the multi-database collaborative knowledge graph layer.

[0037] In one specific embodiment, the data acquisition layer consists of four functional units, which are described in detail below.

[0038] The BIM model data extraction unit is a software module used to parse construction model files. This unit is configured with parsers for general BIM standard formats (such as IFC) or specific software formats (such as RVT). Its execution process is as follows: it receives the specified BIM model file as input, and by calling the application programming interface (API) or parsing library corresponding to the file format, it traverses the model's data structure to extract predefined categories of geometric and attribute information. Geometric information includes the 3D coordinates of vertices, face normals, and volume of the volumes of components such as transmission towers, foundations, and conductors. Attribute information includes the component's unique identifier (GUID), material type (such as Q345B grade for steel, C35 grade for concrete), component status (such as installed, awaiting installation), and preset construction time nodes. This unit outputs the extracted information in a unified JSON object format.

[0039] The sensor data acquisition unit is used to collect and process time-series data from automatic physical sensors. This unit integrates multiple communication protocol stacks, such as Modbus and MQTT, and connects to the field sensor network via a physical interface (such as an RS-485 bus) or a wireless gateway (such as a LoRaWAN gateway). The unit periodically sends polling commands to the sensors or receives data actively reported by the sensors in a subscription mode. The acquired data includes, but is not limited to: stress values ​​of strain gauges at key nodes of the tower, wind speed and direction readings from an aerial anemometer, and angle change data from an inclinometer on the foundation pit slope. For the received raw data, the unit performs data cleaning operations, including filtering out outliers exceeding preset physical thresholds and performing linear interpolation on missing data points. Finally, a precise timestamp is appended to each valid data point, forming a standard time-series data stream.

[0040] The real-time monitoring data acquisition unit is used to acquire real-time dynamic data from existing security and monitoring systems on-site. This unit connects to on-site cameras via standard video streaming protocols (such as RTSP), acquires video streams of designated monitoring areas, and forwards them to the intelligent analysis and processing layer for subsequent image analysis. Simultaneously, this unit uses a RESTful API provided by a personnel / machinery positioning system (such as UWB or BeiDou positioning system) to acquire the real-time coordinate sequences of all registered personnel and large machinery on-site at a preset frequency (e.g., 1Hz). The acquired data is in the form of a data packet containing device ID, timestamp, and three-dimensional coordinates (x, y, z).

[0041] The historical data mining unit is a processing module used to extract structured risk event information from unstructured or semi-structured historical documents. This unit receives text files (such as PDF or DOCX formats) as input, including historical construction logs, safety inspection reports, and accident investigation reports. Its internal processing flow includes: first, converting the document content into plain text using a text extraction component; then, using a Natural Language Processing (NLP) model based on a predefined dictionary to perform Named Entity Recognition (NER) on the text, extracting key entities such as risk type (e.g., falling objects from height), involved equipment (e.g., tower crane), location, and triggering conditions; finally, determining the relationships between entities using a relation extraction algorithm (e.g., determining that a violation was the cause of falling objects from height). This unit ultimately integrates the extracted fragmented information into structured event records, each containing fields such as event ID, occurrence time, risk type, cause, and consequences, and stores them in a database.

[0042] See appendix Figure 1 and attached Figure 3The multi-database collaborative knowledge graph layer of this application is the knowledge core of the system. Its input end is connected to the data acquisition layer, and its output end is connected to the intelligent analysis and processing layer. This layer is responsible for fusing, processing, and structuring the heterogeneous and multi-source data from the data acquisition layer, ultimately forming a professional knowledge graph for the field of power transmission construction risk management.

[0043] In a specific embodiment, the construction and maintenance process of this knowledge graph is as follows: First, the system predefines the data schemas of four core knowledge bases. These four bases are logically independent but interconnected in terms of data, forming the foundation of the knowledge graph. The Risk Base uses risk as its core entity. Each risk entity includes the following attributes: Risk ID (unique identifier), Risk Name (e.g., electric shock accident), Risk Level (e.g., Level I, Level II), Risk Category (e.g., personal injury, equipment damage), Triggering Conditions (describing the specific situation or event that led to the risk), and a description of the consequences. The Process Knowledge Base uses process steps as its core entity. Each process step entity includes the following attributes: Process Step Code (unique identifier), Step Name (e.g., tower material hoisting), Prerequisite Steps, Subsequent Steps, Standard Duration, Required Resource List (including personnel types and quantities; machinery models and quantities; material specifications and quantities), and corresponding quality acceptance standards. The Hazard Base uses hazard as its core entity. Each hazard entity includes the following attributes: Hazard ID (unique identifier), Hazard Description (e.g., someone is standing below the hoisting area), Associated Risk Factor Identifier (indicating the specific risk that triggered it), Inspection Standards, and Suggested Rectification Measures. The construction plan database has a data structure centered around construction plans as the core entity. Each plan entity contains the following attributes: plan ID (unique identifier), plan name, applicable project type, plan type identifier (e.g., standard plan, high-altitude plan), and an ordered list of multiple process step codes, representing the complete construction process of the plan.

[0044] The knowledge graph construction process includes entity extraction, relation extraction, and knowledge fusion. When new data is input into the data acquisition layer 100, such as a new construction log, the knowledge graph layer 200 first uses named entity recognition technology to extract entities related to the four databases mentioned above, such as identifying the foundation excavation process. Then, through relation extraction technology, the relationships between entities are determined, such as extracting the causal relationship between "heavy rain" (triggering condition) and "water accumulation in the foundation pit" (hazard) from the sentence "heavy rain caused water accumulation in the foundation pit." Finally, through knowledge fusion technology, the newly extracted triples (entity-relationship-entity) are compared and merged with the existing knowledge graph to achieve dynamic knowledge updates, including adding new entities and relations or updating the attributes of existing entities.

[0045] The core feature of this knowledge graph lies in its internal bidirectional dynamic mapping relationship. This relationship enables four independent knowledge bases to work collaboratively, forming a closely interconnected network of solutions, risks, and hidden dangers. The specific establishment methods are as follows: The connection between the risk base and the process knowledge base is established through process step coding. In the risk base, each risk entity is associated with one or more process step codes that trigger that risk. Conversely, in the process knowledge base, each process step also points to a risk entity that requires focused prevention in that step. For example, the tower material hoisting step is associated with the risk of falling objects from heights. The connection between the process knowledge base and the hidden danger base is established through risk factor identification. In each step of the process knowledge base, in addition to being associated with macro-level risks, it is further refined to specific risk factors. These risk factor identifications directly link to specific hidden danger entities in the hidden danger base. For example, the tower material hoisting step is not only associated with the risk of falling objects from heights, but also, through the risk factor of unstable hoisting, is associated with specific hidden dangers in the hidden danger base such as excessive wear of slings and incorrect binding methods. The connection between the hidden danger base and the construction solution base is established through countermeasure coding. Each hazard entity in the hazard database comes with recommended rectification or prevention measures, each assigned a unique response code. Similarly, in the construction plan database, a specific construction plan includes a series of safety measures, which also use these codes. This allows the system to quickly retrieve which specific hazards a plan can address. The connection between the construction plan database and the risk database is established through plan type identifiers. Plans in the construction plan database are categorized and labeled, such as construction plans for high-altitude areas. In the risk database, risks like altitude sickness are specifically highlighted for their strong correlation with the high-altitude plan type identifier. This connection enables the system to prioritize and match appropriate specialized plans when facing specific environments or project types.

[0046] Through the aforementioned multi-layered and multi-dimensional mapping relationships, the multi-database collaborative knowledge graph layer weaves the originally isolated data points and knowledge points into a semantically rich and logically rigorous network structure, providing the intelligent analysis and processing layer with high-quality, computable analysis objects.

[0047] See appendix Figure 4 - Appendix Figure 7 The intelligent analysis and processing layer of this application is the core of the system's calculation and analysis. It receives the associated network of schemes, risks, and hidden dangers from the multi-database collaborative knowledge graph layer as input, processes it through multiple internally integrated algorithm modules, and finally outputs risk warning information, optimized construction schemes, and quantitative assessment reports to the visualization decision layer.

[0048] In one specific embodiment, the intelligent analysis and processing layer includes the following modules: The risk identification module performs deep risk identification based on knowledge graphs. Its core employs graph neural network (GNN) algorithms, such as graph convolutional networks (GCNs). The module's execution flow is as follows: First, the association network generated by the multi-database collaborative knowledge graph layer is converted into a graph data structure that can be processed by the GNN model. This structure includes a node feature matrix X and an adjacency matrix A. Each row in the node feature matrix X represents a vectorized representation of an entity (such as a specific construction worker, a piece of equipment, or a process step), and this vector is encoded by the entity's attributes (such as the worker's job type or the equipment's age). The adjacency matrix A describes the connections between entities.

[0049] Next, the graph data is fed into a pre-trained GCN model. This model contains multiple graph convolutional layers. In each layer, each node aggregates information from its neighbors to update its own representation. This process can be represented by the following formula: ; in, It is the first The node representation matrix of the layer; . It adds a self-loop adjacency matrix; It is the identity matrix; yes The degree matrix; It is the first The trainable weight matrix of the layer; It is a non-linear activation function (such as ReLU); through multi-layer stacking, the model can capture high-order dependencies between nodes in the graph.

[0050] Finally, the trained model is used for risk identification, specifically link prediction. The module takes two non-directly connected nodes in the graph (e.g., a novice operator and an aerial work platform) as input and predicts the probability of a high-risk relationship (i.e., a link exists) between them by calculating the dot product of their final node representations or using a shallow neural network. When this probability exceeds a preset threshold (e.g., 0.8), the risk identification module generates a specific risk warning and outputs it.

[0051] The scheme optimization module is used to automatically optimize existing construction schemes. Its core utilizes a genetic algorithm. The module's execution flow is as follows: First, coding is performed. A construction plan is abstracted as a chromosome, with the basic units in the plan (such as the execution order of process steps, the construction teams assigned to specific steps, and the types of machinery used) as genes.

[0052] Secondly, initialize the population. A set of initial construction schemes is generated randomly or based on a historical scheme library, serving as the first generation of the population. Then, iterative optimization loop begins. In each generation: Fitness assessment: Calculate the fitness score for each scheme (chromosome) in the population. The fitness function is a multi-objective function, for example: ; in, This is the total project duration; It is the total cost; It is a comprehensive risk index assessed through the risk identification module 310; These are the weighting coefficients for each item; the higher the fitness score, the better the solution.

[0053] Selection: A roulette wheel selection strategy is used. The probability of each option being selected is proportional to its fitness score. Options with higher fitness scores are more likely to be selected for reproduction.

[0054] Crossover: Randomly pair up selected parents and exchange some of their genes with a certain crossover probability (e.g., exchange the order of construction steps in two schemes) to generate new offspring schemes.

[0055] Mutation: Certain genes in the offspring population are randomly modified with a low mutation probability (e.g., replacing a construction machine in a certain step with another model) to increase population diversity and avoid getting trapped in local optima. This process is repeated until a preset number of iterations is reached or the average fitness of the population converges. Finally, the module outputs the solution with the highest fitness in the current population as the optimized construction solution.

[0056] The scoring algorithm module is used to objectively and multidimensionally quantify the construction plan and generate a quantitative assessment report. Its assessment dimensions include the plan's risk matching degree, the effectiveness of hazard mitigation, and the rationality of the schedule. The calculation of the plan's risk matching degree has been detailed above. The score for the effectiveness of hazard mitigation is obtained by retrieving the codes of mitigation measures included in the plan, statistically analyzing the number and level of hazards in the hazard database that the plan can cover, and then performing a weighted calculation. The score for the rationality of the schedule is obtained by comparing the planned construction period of the plan with the standard construction period in the process knowledge base, and combining this with the resource conflict detection results from the 4D simulation module. The module integrates the scores from all dimensions to form a comprehensive assessment report that includes a radar chart, specific sub-scores, and improvement suggestions.

[0057] The 4D simulation module integrates the 3D spatial information of the BIM model with the time schedule information of the construction plan (the fourth dimension). The module receives a construction plan (including the sequence of process steps and the planned start and end times of each step) and the corresponding BIM model as input. It associates the installation status of each component with the timeline. The core function of the module is dynamic collision detection. During the simulation, it progresses in discrete time steps, detecting geometric interference between all moving objects (such as hoisted components and moving machinery) and fixed objects (such as existing structures and scaffolding), as well as preset safety zones (such as the safety distance envelope under high-voltage lines). Once a collision or intrusion is detected, the module records the time, location, and objects involved in the conflict, and uses this as a negative evaluation indicator, feeding it back to the scoring algorithm module or highlighting it directly in the visualization decision layer 400.

[0058] See appendix Figure 1 and attached Figure 7 The visualization decision-making layer of this application serves as the front-end interface for system-user interaction. Its function is to receive various analysis results from the intelligent analysis and processing layer and present them to project managers in a structured and graphical manner, providing direct data support for their decision-making. This layer is typically deployed on the user's terminal device, such as a PC workstation, tablet computer, or large-screen display system.

[0059] In one specific embodiment, the user interface (UI) of the visualization decision layer is divided into several functional areas, each implemented by a different display unit: The Risk Warning Unit is specifically designed to display risk warning information generated by the risk identification module. In one embodiment, this unit exists in a prominent position on the interface (e.g., at the top or right sidebar) as a scrollable list. Each warning message is displayed as a separate entry, containing the following fields: Risk level: indicated by icons of different colors (e.g., red for high risk, orange for medium risk, and yellow for low risk).

[0060] Warning content: A concise text description explaining the specific details of the risk, such as: Construction worker Zhang San (employee number 075) was working at the edge of the foundation pit (coordinates X, Y) when no safety rope connection signal was detected, posing a risk of falling from a height.

[0061] Related Entities: Clickable links pointing to specific personnel, equipment, or locations in the BIM model involved in the risk. Clicking these links will highlight these entities in other parts of the interface.

[0062] Timestamp: The time the alert was generated. When a new high-risk alert is generated, this unit will trigger an audible alarm, and the corresponding alert entry will be pinned to the top and flash to attract the user's attention.

[0063] The Solution Evaluation Report section displays the quantitative evaluation report generated by the scoring algorithm module. It is not a single view, but rather an interactive dashboard.

[0064] Main view: Using a radar chart, it visually compares the scores of the current solution and the optimized solution across multiple dimensions, including risk matching, effectiveness of hazard mitigation, schedule rationality, and cost control. Users can clearly see in which dimensions the optimized solution has improved.

[0065] Detailed data tables: Located below the radar chart or expandable by clicking on a dimension, these tables display detailed scoring data and calculation basis. For example, clicking on "Solution Risk Matching" will list all identified risk items, their levels, and whether each solution covers these risks.

[0066] Improvement suggestion list: Based on the dimensions with lower scores, the system automatically extracts and generates specific improvement suggestion text from the knowledge base. For example, if the hazard response effectiveness score is low, the suggestion list will show a suggestion to add a pre-shift inspection item for sling wear.

[0067] The simulation display unit is a window that integrates a 3D rendering engine to visualize the construction process. Its data comes from the output of the 4D simulation module.

[0068] 3D Scene: The BIM model of the project is loaded into the unit, creating a 3D environment consistent with the actual construction site.

[0069] Timeline control: A timeline is located at the bottom of the window, allowing users to drag the slider to view the construction status at any given moment. As the timeline moves, the building components in the 3D scene will change from an incomplete state to an installed state according to the progress of the construction plan.

[0070] Dynamic elements: Construction machinery (such as tower cranes and concrete pump trucks) and construction workers (represented by simplified models) will be dynamically demonstrated according to the paths and times set in the plan.

[0071] Conflict Highlighting: When the timeline reaches the moment when a conflict is detected by the 4D simulation module, the simulation animation pauses, and the object involved in the conflict (e.g., the crane boom and the existing structure) is highlighted in red, with a pop-up window displaying conflict details. This unit allows users to intuitively preview the entire construction process and identify potential spatiotemporal logic errors and safety issues—something that cannot be achieved with 2D drawings and static plans.

[0072] These three units work together in the same user interface, providing users with a comprehensive and multi-layered decision-making view, from macro-level risk situation to micro-level solution details, and even dynamic construction process simulation.

[0073] Example 2 One embodiment of this application provides a knowledge graph-based intelligent control method for construction safety risks. This method is implemented through a aforementioned knowledge graph-based intelligent control system for construction safety risks. The execution steps of this method will be described in detail below using a specific power transmission construction scenario as an example.

[0074] The scenario for the proposed implementation is as follows: foundation excavation and concrete pouring operations are being carried out on a power transmission tower (model NJ2-30) in a high-altitude area (e.g., 4500 meters above sea level).

[0075] The method includes the following steps: Step S1: Data Acquisition First, the system's data acquisition layer performs data acquisition operations. For the defined scenario, the BIM model data extraction unit retrieves and parses the BIM model of the NJ2-30 iron tower and its foundation from the project database, extracting the foundation's dimensions, depth, reinforcement configuration, required concrete grade (e.g., C35 frost-resistant type), and the digital elevation model (DEM) data of the surrounding terrain.

[0076] The sensor data acquisition unit is activated and receives parameters from sensors deployed on-site. This includes data from displacement sensors installed on the slope, readings from temperature sensors embedded in the ground to monitor changes in the permafrost layer, and real-time wind speed, air pressure, and oxygen content data sent by the weather station.

[0077] The real-time monitoring data acquisition unit obtains video streams of key work areas from the on-site video surveillance system via API interface, and obtains real-time coordinate data of construction personnel and large machinery from UWB (Ultra-Wideband) positioning tags worn by personnel.

[0078] The historical data mining unit retrieves historical risk event data related to keywords such as high altitude, permafrost, and strong winds from the system's historical database, and extracts structured records of similar events that have occurred in similar projects, such as altitude sickness leading to personnel operational errors and permafrost thawing causing small-scale slope slippage.

[0079] Step S2: Knowledge Graph Construction and Association Network Formation - The multi-database collaborative knowledge graph layer receives all data from the data acquisition layer. The system integrates and updates the newly acquired data with the existing four knowledge bases (risk base, process knowledge base, hidden danger base, and construction plan base). For example, real-time acquired low oxygen content data will trigger risk entities related to high-altitude hypoxia in the knowledge graph. The geometric features of deep foundation pits in the BIM model will be associated with foundation pit wall instability entities in the hidden danger base.

[0080] Subsequently, based on a pre-defined entity relationship model, the system dynamically constructs a network of connections between the solutions, risks, and hazards related to this operation within the updated knowledge graph. This network uses the NJ2-30 tower foundation construction solution as its central node, connecting to activated risk entities (such as falling objects from heights, mechanical injuries, and altitude sickness) and hazard entities (such as workers not wearing safety harnesses or people within the excavator's swing radius) through multiple paths. This is a weighted directed graph, where edge weights can be assigned based on the frequency of risk occurrences or the severity of their consequences in historical data.

[0081] Step S3: Intelligent Analysis and Processing The intelligent analysis and processing layer calculates and analyzes the correlation network generated in step S2.

[0082] The risk identification module takes this interconnected network as input and feeds it into a pre-trained Graph Convolutional Neural Network (GNN) model. This model learns the characteristics of nodes in the network and the graph's topology to predict links. For example, if the model detects a potentially high-risk link between worker A, the pit edge node, and the nighttime construction node, even without any explicit violation, the system will generate a warning message indicating a potential risk of personnel falling into the pit under poor visibility conditions at night.

[0083] The scheme optimization module initiates a genetic algorithm to optimize existing standard construction schemes. The algorithm encodes the sequence of construction steps, resource (personnel, machinery) allocation, and work / rest schedules as genes. Its fitness function is set as follows: ; in For the total construction period, For total cost, This is the comprehensive risk index assessed through the risk identification module. , where is the weighting coefficient. The algorithm generates an optimized solution through multiple iterations, for example: suggesting that concrete pouring operations be scheduled during the midday period when wind speeds are lowest, and that additional mandatory rest time be provided for nighttime construction workers to cope with altitude sickness.

[0084] The scoring algorithm module quantitatively evaluates the solutions before and after optimization. Based on the aforementioned formula, the module calculates the risk-matching degree of the solutions and generates a structured quantitative evaluation report by combining other dimensions (such as the effectiveness of hazard mitigation and the rationality of the schedule). The report clearly indicates that the original solution's coverage rate for measures to address the risk of permafrost thawing was 70%, while the optimized solution, by adding temperature monitoring and pre-heating measures, increased the coverage rate to 95%.

[0085] Meanwhile, the 4D simulation module receives the optimized construction plan and, combined with the BIM model and construction schedule, generates a dynamic 3D construction process animation. This simulation can intuitively predict the movement trajectory of large machinery (such as cranes and concrete pump trucks) and automatically detect whether their working area conflicts with personnel activity areas or the safety distance of high-voltage lines.

[0086] Step S4: Visualization Output and Decision Making The visualization decision layer receives all output results from the intelligent analysis and processing layer.

[0087] On the user terminal's graphical interface, the risk warning unit displays the risk warning information generated in step S3 as a highlighted red pop-up. The scheme evaluation report unit compares and displays the scores of the original scheme and the optimized scheme in terms of risk, cost, and schedule using radar charts and data tables. The simulation display unit plays a construction process animation generated by the 4D simulation module in a separate window, allowing users to drag the timeline to view the construction status and potential conflict points at any given moment.

[0088] Ultimately, project managers make decisions based on this intuitive and quantifiable information, select and adopt the optimized construction plan, and adjust the focus of on-site safety supervision according to risk warnings.

[0089] Example 3 This embodiment is based on a 220kV transmission line foundation construction project of a provincial power company, and constructs a complete multi-database linkage knowledge graph intelligent management and control system.

[0090] Step 1: Building the data acquisition layer.

[0091] The sensor data acquisition unit is equipped with 12 environmental monitoring sensors, including 4 temperature sensors (measurement range -40℃ to 80℃, accuracy ±0.5℃), 4 humidity sensors (measurement range 0 to 100%RH, accuracy ±3%RH), 2 wind speed sensors (measurement range 0 to 60m / s, accuracy ±0.3m / s), and 2 soil density sensors (measurement range 1.2 to 2.8g / cm³, accuracy ±0.05g / cm³). Data acquisition is set to once every 5 minutes, and data is transmitted to the central database in real time.

[0092] The BIM model data extraction unit constructs a 3D model of the tower foundation based on the Bentley MicroStation platform. It includes 16 key construction steps, such as excavation, rebar tying, and concrete pouring. Each step contains attribute information such as process parameters, quality requirements, and safety specifications. The model accuracy is set to LOD300 level, with geometric accuracy errors controlled within ±10mm.

[0093] The historical data mining unit extracted 1,247 risk event records from similar projects between 2018 and 2023 from the company's historical project database. These records included information such as event type, occurrence time, process steps, cause analysis, and corrective measures. Data preprocessing employed data cleaning algorithms to remove duplicate and invalid records, ultimately yielding 1,089 valid samples.

[0094] The real-time monitoring data acquisition unit integrates a construction site video surveillance system, deploying eight high-definition cameras to cover key areas such as the foundation pit operation area, material storage area, and equipment operation area. The video resolution is set to 1920×1080, with a frame rate of 25fps, supporting 24-hour continuous monitoring and intelligent recognition.

[0095] Step 2: Establish a multi-database collaborative knowledge graph layer.

[0096] The risk database covers five major risk elements: geological risk, meteorological risk, personnel risk, equipment risk, and process risk. Each risk category is further subdivided into 3-5 subcategories, totaling 23 specific risk types. Risk levels are divided into four levels: extremely high (weight 0.9-1.0), high (weight 0.7-0.9), medium (weight 0.5-0.7), and low (weight 0.3-0.5).

[0097] The process knowledge base establishes a knowledge system covering four major process modules: foundation pit excavation, reinforcement engineering, concrete engineering, and backfilling engineering. Each module includes information such as operating procedures, technical standards, and key quality control points. The process step coding adopts a three-level classification system, such as "01-02-03" representing foundation pit excavation - earthwork excavation - mechanical excavation.

[0098] The hazard database establishes a hazard classification system based on risk factors, including four major categories: safety hazards, quality hazards, schedule hazards, and environmental hazards. Each category of hazard corresponds to 3-8 specific manifestations. Hazard levels are divided into three levels according to their impact: major, significant, and minor, with corresponding response measures of immediate work stoppage, rectification within a specified time, and continuous monitoring, respectively.

[0099] The construction plan database stores 156 standard construction plans and 89 optimized plans, covering construction scenarios with different geological conditions, meteorological environments, and project scales. The plan type identification adopts a coding system, including dimensions such as geological type code, meteorological type code, and scale type code.

[0100] The mapping relationships between the various databases are established through a MySQL database. The risk database is linked to the process knowledge database through the process step coding field, the process knowledge database is linked to the hidden danger database through the risk factor identification field, the hidden danger database is linked to the construction plan database through the countermeasure coding field, and the construction plan database is linked to the risk database through the plan type identification field, forming a complete closed-loop mapping system.

[0101] Step 3: Intelligent analysis and processing.

[0102] The risk identification module implements a graph neural network algorithm based on the TensorFlow deep learning framework. The network structure consists of three graph convolutional layers with 128, 64, and 32 neurons in the hidden layers, respectively. The activation function is ReLU, the learning rate is set to 0.001, the batch size is 32, and the training iterations are set to 500. Risk propagation path weights are calculated based on the similarity and historical association strength between nodes, with a weight threshold set to 0.7. Nodes exceeding this threshold are marked as potential risk points.

[0103] The scheme optimization module uses a genetic algorithm to find the best scheme. The population size is set to 50 individuals, with each individual representing a construction scheme configuration. The crossover probability is set to 0.8, the mutation probability is set to 0.1, and the selection strategy is roulette wheel selection. The fitness function comprehensively considers three factors: risk level, cost-effectiveness, and schedule rationality, with weights of 0.5, 0.3, and 0.2, respectively. The termination condition is set as no significant improvement for 20 consecutive generations or reaching the maximum number of iterations of 200 generations.

[0104] The 4D simulation module builds a 3D simulation environment based on the Unity3D engine. The imported BIM model contains geometric information and construction sequence of key processes such as foundation excavation, rebar tying, and concrete pouring. The time dimension is set in hours, with a simulation time step of 0.5 hours, and the total simulation duration covers the entire construction cycle (usually 15-30 days). The risk status update frequency is synchronized with the actual monitoring data, refreshing every 5 minutes.

[0105] The scoring algorithm module implements multi-dimensional quantitative scoring. The formula for calculating the risk matching degree of the solution is: Matching degree score = Risk coverage rate × 0.6 + Risk level weight × 0.4, where risk coverage rate = Number of identified risks / Total number of potential risks, and the risk level weight is calculated based on the weighted average of the number of risks at each level. The effectiveness score for hazard response is based on the completeness and relevance of the response measures, and the formula is: Effectiveness score = Measure coverage × 0.7 + Measure relevance × 0.3. The schedule rationality score is calculated based on critical path analysis and the degree of resource allocation optimization.

[0106] Example 4 Based on Example 3, this embodiment adopts optimized algorithm parameters and enhanced functional modules for the construction of 500kV transmission lines under complex geological conditions.

[0107] Optimize parameter settings: The neural network learning rate was optimized to an adaptive learning rate, initially set to 0.001, with a decay rate of 0.95 every 100 iterations. The number of network layers was increased to 5, and the number of neurons in the hidden layers was adjusted to 256, 128, 64, 32, and 16. A Dropout mechanism was introduced, with a dropout rate set to 0.2, effectively improving the model's generalization ability. The risk propagation path weight threshold was dynamically adjusted according to the project complexity, set to 0.6 for complex projects and maintained at 0.7 for general projects.

[0108] The genetic algorithm population was expanded to 100 individuals, employing an elite retention strategy with a retention rate of 10%. Multi-point crossover and non-uniform mutation operators were introduced, with the number of crossover points set to 2-4 randomly selected, and the mutation intensity decreasing with each iteration. Environmental factors were incorporated into the fitness function, with weights adjusted to: risk level 0.4, cost-effectiveness 0.25, schedule rationality 0.2, and environmental impact 0.15.

[0109] Enhanced functionality modules: A dedicated geological risk identification module has been added, integrating multi-source geological information such as geological survey data, borehole data, and groundwater level monitoring data. A geological risk assessment model has been established, employing a fuzzy comprehensive evaluation method, considering 11 evaluation factors including soil type, groundwater conditions, and seismic intensity. Geological risk levels are divided into 5 levels, corresponding to different foundation pit support schemes and excavation techniques.

[0110] The enhanced 4D simulation module integrates VR virtual reality technology, supporting an immersive construction process experience. The VR device uses an HTC Vive Pro 2 with a resolution of 2448×2448 (per eye), a refresh rate of 90Hz, and a field of view of 110°. The virtual environment includes realistically scaled construction scenes and supports multi-user collaborative operation and real-time interaction.

[0111] An intelligent early warning subsystem has been added, establishing an early warning model based on real-time monitoring data and historical trend analysis. Early warning levels are divided into four categories: blue (caution), yellow (warning), orange (danger), and red (emergency), each corresponding to different response measures. Early warning trigger conditions include 12 scenarios such as deteriorating weather conditions, equipment malfunctions, and personnel violations.

[0112] Performance comparison data: Compared to the basic implementation, the preferred implementation improves risk identification accuracy from 95% to 97.8%, and reduces the false alarm rate from 7.2% to 4.1%. Solution optimization efficiency is further improved, with the average optimization time reduced from 15-20 minutes to 8-12 minutes. Overall scoring accuracy is improved by 8.5%, and user satisfaction increases from 87% to 94%.

[0113] The VR simulation module achieved a user experience score of 4.7 out of 5, demonstrating significantly better immersion and interactivity than traditional 3D display methods. The intelligent early warning response time averages 3.2 minutes, with an accuracy rate of 91.5%, effectively reducing the probability of safety incidents.

[0114] Example 5 This embodiment is designed for the special application scenario of offshore wind power transmission line construction, and adapts the basic system architecture accordingly.

[0115] Offshore construction environment adaptation The data acquisition layer includes marine environmental monitoring equipment, such as a wave height meter (measurement range 0-15m, accuracy ±0.1m), an ocean current velocity meter (measurement range 0-5m / s, accuracy ±0.05m / s), a seawater salinity meter (measurement range 0-40‰, accuracy ±0.1‰), and a tide prediction system (prediction accuracy ±5cm). Considering the limitations of maritime communication conditions, a hybrid 4G / 5G network is used to ensure data transmission stability.

[0116] The knowledge graph base has been expanded to reflect the characteristics of offshore construction. The risk base has been expanded to include eight marine-specific risks, such as wind and wave risk, tidal risk, marine corrosion risk, and ship collision risk. The process knowledge base has been expanded to include specialized offshore construction processes, such as steel pipe pile construction, submarine cable laying, and offshore substation installation.

[0117] Special algorithm adaptation The risk identification module adds a marine environment prediction algorithm, which predicts sea state changes over the next 72 hours based on numerical weather prediction and ocean dynamics models. An LSTM (Long Short-Term Memory) network is used, with the input layer containing 16 marine meteorological parameters such as wind speed, wind direction, air pressure, and sea surface temperature. The hidden layers are set to three layers with 64, 32, and 16 neurons respectively.

[0118] The optimization scheme takes into account tidal window constraints and adds a time window optimization algorithm. Construction operations must be carried out under suitable sea conditions: wave height less than 1.5m, wind speed less than 12m / s, and visibility greater than 1000m. The sea condition adaptability weight is increased to 15% in the optimization objective function.

[0119] The 4D simulation integrates dynamic marine environmental effects, including wave animations, tidal changes, and weather system evolution. Simulation accuracy is improved to the minute level, with real-time synchronization of marine forecast data, providing precise environmental predictions for offshore construction.

[0120] Application effect verification This modified embodiment was applied during the construction of a 220kV submarine cable for an offshore wind power project, successfully identifying 12 instances of sudden sea state changes and preventing potential construction accidents. The utilization rate of the construction window increased from 65% in the traditional approach to 83%, significantly improving construction efficiency. The safety factor for offshore operations increased by 32%, earning high praise from the project owner and supervision unit.

[0121] Example 6 In this embodiment, an electronic device is also provided, including a processor and a memory.

[0122] Memory is used to store non-transitory computer-readable instructions.

[0123] The processor is used to execute non-transitory computer-readable instructions, which, when executed by the processor, can perform one or more steps of the knowledge graph-based intelligent control method for construction safety risks described above. The memory and processor can be interconnected via a bus system and / or other forms of connection.

[0124] Example 7 In this embodiment, a computer-readable storage medium is also provided.

[0125] Computer-readable storage media are used to store non-transitory computer-readable instructions.

[0126] For example, when a non-transitory computer-readable instruction is executed by a computer, one or more steps in the knowledge graph-based intelligent control method for construction safety risks described above can be performed.

[0127] For example, this storage medium can be used in the aforementioned electronic device. For example, the storage medium can be a memory in the electronic device. For example, relevant descriptions of the storage medium can be found in the corresponding descriptions of the memory in the aforementioned electronic device, and will not be repeated here.

[0128] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0129] The aforementioned knowledge graph-based intelligent control system for construction safety risks, along with its functional layers / modules / units and the risk control methods it implements, can all be implemented using computer program instructions. These computer programs can be stored in a computer-readable storage medium or transmitted over a network. When these programs are run on a computing device, the device executes the methods disclosed in this application.

Claims

1. A knowledge graph-based intelligent control system for construction safety risks, characterized in that, include: The data acquisition layer is used to collect construction model data, process parameters, real-time monitoring data, and historical risk event data from the construction site. A multi-database collaborative knowledge graph layer constructs a knowledge graph based on the data collected by the data acquisition layer. The intelligent analysis and processing layer is used to analyze the correlation network of plans, risks, and hidden dangers, and generate risk warning information, optimized construction plans, and quantitative assessment reports. The visualization decision-making layer receives and displays the risk warning information, optimized construction plans, and quantitative assessment reports generated by the intelligent analysis and processing layer.

2. The knowledge graph-based intelligent control system for construction safety risks according to claim 1, characterized in that, The knowledge graph includes a risk database, a process knowledge database, a hidden danger database, and a construction plan database. The risk database, process knowledge database, hidden danger database, and construction plan database establish a two-way dynamic mapping relationship through a preset entity relationship model, and form an association network of plans, risks, and hidden dangers.

3. The knowledge graph-based intelligent control system for construction safety risks according to claim 1, characterized in that, The data acquisition layer includes: The BIM model data extraction unit is used to extract geometric and attribute information from the construction model data. A sensor data acquisition unit is used to acquire the process parameters; A real-time monitoring data acquisition unit is used to acquire the real-time monitoring data; The historical data mining unit is used to process the historical risk event data.

4. The knowledge graph-based intelligent control system for construction safety risks according to claim 1, characterized in that, The intelligent analysis and processing layer includes: The risk identification module is used to identify potential risks and generate the risk warning information. The scheme optimization module is used to generate the optimized construction scheme; The scoring algorithm module is used to calculate the quantitative evaluation report.

5. The knowledge graph-based intelligent control system for construction safety risks according to claim 1, characterized in that, The intelligent analysis and processing layer also includes: The 4D simulation module integrates the construction progress time dimension to dynamically simulate the construction process and output simulation results.

6. The knowledge graph-based intelligent control system for construction safety risks according to claim 1, characterized in that, The visualization decision-making layer includes: The risk warning unit is used to display the risk warning information; The scheme evaluation report unit is used to display the quantitative evaluation report; The simulation demonstration unit is used to visualize the construction process.

7. The knowledge graph-based intelligent control system for construction safety risks according to claim 2, characterized in that, The bidirectional dynamic mapping relationship is established as follows: The risk database and the process knowledge base are linked through process step coding. The process knowledge base and the hidden danger database are linked through risk factor identification; The hazard database and the construction plan database are linked through response measure codes; The construction plan library and the risk library are linked through a plan type identifier.

8. The knowledge graph-based intelligent control system for construction safety risks according to claim 4, characterized in that, The risk identification module uses a graph neural network algorithm to identify risks based on the correlation network of the scheme, risk, and hidden danger. The scheme optimization module uses a genetic algorithm to find the best construction scheme, and the genetic algorithm uses a roulette wheel selection strategy. The scoring algorithm module is used to calculate the quantitative evaluation report, which includes multi-dimensional scores. The multi-dimensional scoring includes the risk matching degree of the solution, the effectiveness of the response to potential hazards, and the rationality of the schedule.

9. A knowledge graph-based intelligent control method for construction safety risks, applied to the knowledge graph-based intelligent control system for construction safety risks as described in any one of claims 1-8, characterized in that, Includes the following steps: S1. Collect construction model data, process parameters, real-time monitoring data and historical risk event data at the construction site through the data acquisition layer; S2. Based on the data collected in step S1, construct and maintain a knowledge graph consisting of a risk database, a process knowledge database, a hidden danger database, and a construction plan database, and form an association network of plans, risks, and hidden dangers. S3. Analyze the network of relationships between the proposed solutions, risks, and hidden dangers to generate risk warning information, optimized construction plans, and quantitative assessment reports. S4. Output and display the risk warning information, the optimized construction plan, and the quantitative assessment report generated in step S3.

10. The intelligent control method for construction safety risks based on knowledge graphs according to claim 9, characterized in that, The step of analyzing the network of relationships between the formed solutions, risks, and hidden dangers further includes: Risk identification is performed using graph neural network algorithms; A genetic algorithm is used to optimize the construction scheme, and the genetic algorithm adopts a roulette wheel selection strategy.

11. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor is configured to execute the computer program, it implements the knowledge graph-based intelligent control method for construction safety risks as described in claim 9 or 10.

12. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the knowledge graph-based intelligent control method for construction safety risks as described in claim 9 or 10.

Images