Risk identification system and risk identification method
By combining scene perception module, data alignment module and multimodal analysis module, the automatic collection and fusion of multi-source data at the construction site is realized, which solves the shortcomings of traditional construction site risk identification, improves the intelligence and real-time performance of risk identification, and is applicable to smart construction sites and digital construction scenarios.
Patent Information
- Application Number
- CN202511296945.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-11
- Publication Date
- 2026-01-23
AI Technical Summary
Traditional construction site risk identification and scheduling management rely on manual inspections, which has problems such as untimely response, weak risk prediction capabilities, and information silos, and cannot effectively utilize multi-source heterogeneous data for risk identification.
By introducing a scene perception module, a data alignment module, and a multimodal analysis module, the automatic collection and fusion of multi-source heterogeneous data at the construction site is realized. A multi-dimensional alignment of time, space, and task semantics is performed through a data alignment mechanism guided by structured data. A pre-trained multimodal risk identification model is used for risk analysis.
It improves the intelligence and real-time performance of risk identification at construction sites, reduces the lag and omission rate of manual inspections, supports the identification of risks such as unprotected high-altitude operations and conflicts in regional operations, and has good scalability and engineering adaptability.
Smart Images

Figure CN121390845A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of risk identification, and in particular to a risk identification system and a risk identification method. BACKGROUND
[0002] With the acceleration of urbanization construction process, the scale of construction site is increasingly expanding, and the operation process involved is becoming increasingly complex. Safety management and progress control become the key link in the engineering project. The risk identification and dispatch management of the traditional construction site mainly rely on manual patrol, static plan and scattered dispatch process, which has many problems such as untimely response, weak risk prediction ability and information island.
[0003] In recent years, with the application of video monitoring, sensors, unmanned aerial vehicles and other sensing devices, the multi-source heterogeneous data such as video, image, text and voice collected by the construction site has increased significantly. How to identify risks based on multi-source heterogeneous data such as video, image, text and voice is a problem to be solved. SUMMARY
[0004] Therefore, the embodiments of the present application provide a risk identification system and a risk identification method to solve the problem that the construction site cannot be identified based on multi-source data in the prior art.
[0005] The first aspect of the embodiments of the present application provides a risk identification system, which comprises: a scene perception module, configured to collect multi-source data of different dimensions of a construction site; a data alignment module, configured to perform data alignment on the multi-source data of different dimensions based on pre-set structured data, to obtain time matching results, space matching results and task progress matching results, to perform joint coding on the multi-source data based on the time matching results, the space matching results and the task progress matching results, and to obtain the multi-modal feature vector; and a multi-modal analysis module, configured to input the multi-modal feature vector into a pre-trained multi-modal risk identification model to analyze the risk of the construction site and obtain a risk identification result.
[0006] The second aspect of the embodiments of the present application provides a risk identification method, which comprises: collecting multi-source data of different dimensions of a construction site; performing data alignment on the multi-source data of different dimensions based on pre-set structured data, to obtain time matching results, space matching results and task progress matching results, performing joint coding on the multi-source data based on the time matching results, the space matching results and the task progress matching results, and obtaining the multi-modal feature vector; and inputting the multi-modal feature vector into a pre-trained multi-modal risk identification model to analyze the risk of the construction site and obtain a risk identification result.
[0007] The beneficial effects of this application's embodiments compared to existing technologies are as follows: The risk identification system provided in this application's embodiments, by introducing a scene perception module, a data alignment module, and a multimodal analysis module, achieves automatic collection and fusion of multi-source heterogeneous data from construction sites, including images, videos, voice, sensor data, and operational information, effectively improving the system's comprehensive perception capability of the site status. Through a data alignment mechanism guided by structured data, it can achieve multi-dimensional alignment of time, space, and task semantics based on historical schedules, BIM parameters, and construction plans, solving the problem of difficult fusion of multi-source data in traditional methods and improving the accuracy and consistency of data modeling. Employing a pre-trained multimodal risk identification model, it possesses cross-modal analysis capabilities, enabling comprehensive understanding of image, text, and behavioral data from the construction site, improving the intelligence level and scene adaptability of risk identification, and thus realizing risk identification of the construction site based on multi-source data. It achieves real-time early warning and automatic analysis of construction risks, supporting the identification of typical risks such as unprotected high-altitude operations, regional operational conflicts, and violations, effectively reducing the lag and omission rate of relying on manual inspections. The system has good scalability and engineering adaptability, and can be flexibly connected to various sensing devices and business systems according to actual construction projects. It is suitable for various scenarios such as smart construction sites and digital construction, avoiding the problem in related technologies that cannot identify risks at construction sites based on multi-source data. Attached Figure Description
[0008] To more clearly illustrate the technical solutions in the embodiments of this application, 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0009] Figure 1 This is a basic schematic diagram of a risk identification system provided in an embodiment of this application;
[0010] Figure 2 This is a basic schematic diagram of a risk identification method provided in an embodiment of this application;
[0011] Figure 3 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation
[0012] In the following description, specific details such as particular application structures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known applications, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.
[0013] A risk identification system and risk identification method according to embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0014] Figure 1 This is a flowchart illustrating a risk identification system provided in an embodiment of this application, such as... Figure 1 As shown, the risk identification system includes: a scene perception module 101, used to collect multi-source data of different dimensions from the construction site; a data alignment module 102, used to align multi-source data of different dimensions based on pre-set structured data to obtain time matching results, spatial matching results, and task progress matching results; and to jointly encode the multi-source data based on the time matching results, spatial matching results, and task progress matching results to obtain the multi-modal feature vector; and a multi-modal analysis module 103, used to input the multi-modal feature vector into a pre-trained multi-modal risk identification model to perform risk analysis on the construction site and obtain risk identification results. This risk identification system can realize the collection, fusion, and risk identification of multi-source data from the construction site.
[0015] Specifically, the scene perception module 101 is used to collect multi-source data from different dimensions at the construction site. This multi-source data includes, but is not limited to, video images, sensor data, construction task records, personnel location information, voice input, and text work order data. This data can come from surveillance cameras, wearable devices, IoT sensors, work platforms, or management systems.
[0016] The data alignment module 102 is used to perform unified processing on the aforementioned multi-source data based on pre-set structured data. The structured data may include historical schedules, construction plans, building information model (BIM) parameters, etc. The data alignment module 102 achieves unified expression of multi-source heterogeneous data and generates multimodal feature vectors representing the current state of the construction site through time tag matching, spatial location mapping, and task semantic alignment.
[0017] Subsequently, the multimodal feature vectors are input into the multimodal analysis module 103, which calls a pre-trained multimodal risk identification model to comprehensively analyze the current construction site status. This model can integrate visual, semantic, and behavioral features, possessing cross-modal understanding capabilities. The risk identification results output by the model can include existing safety hazards (such as unprotected work at heights, equipment conflicts, and incorrect work sequence), and can further quantify the risk level and scope of impact.
[0018] Through the collaboration of the above modules, the system can realize the entire process from multi-dimensional information perception at the construction site to in-depth risk analysis, significantly improving the automation level and real-time performance of safety hazard identification at the construction site.
[0019] According to the solution provided in this embodiment, the risk identification system introduces a scene perception module 101, a data alignment module 102, and a multimodal analysis module 103 to achieve automatic collection and fusion of multi-source heterogeneous data at the construction site, including images, videos, voice, sensor data, and operational information, effectively improving the system's comprehensive perception capability of the site status. Through a data alignment mechanism guided by structured data, multi-dimensional alignment of time, space, and task semantics can be achieved based on historical schedules, BIM parameters, and construction plans, solving the problem of difficult fusion of multi-source data in traditional methods and improving the accuracy and consistency of data modeling. Employing a pre-trained multimodal risk identification model, it possesses cross-modal analysis capabilities, comprehensively understanding image, text, and behavioral data from the construction site, improving the intelligence level and scene adaptability of risk identification, and thus realizing risk identification of the construction site based on multi-source data. Real-time early warning and automatic analysis of construction risks are achieved, supporting the identification of typical risks such as unprotected high-altitude operations, regional operational conflicts, and violations, effectively reducing the lag and omission rate of relying on manual inspections. The system has good scalability and engineering adaptability, and can be flexibly connected to various sensing devices and business systems according to actual construction projects. It is suitable for various scenarios such as smart construction sites and digital construction, avoiding the problem in related technologies that cannot identify risks at construction sites based on multi-source data.
[0020] In some examples, the structured data includes historical schedule structured data, building information model parameter structured data, and construction plan structured data. The data alignment module 102 is also used to perform time-dimensional label matching on the multi-source data based on the historical schedule structured data to determine the time matching result; to perform spatial location mapping on the multi-source data using the building information model parameter structured data to determine the spatial matching result; to perform task semantic alignment on the multi-source data using the construction plan structured data to determine the task progress matching result; and to perform joint encoding on the multi-source data based on the time matching result, spatial matching result, and task progress matching result to obtain a multimodal feature vector.
[0021] Specifically, the data alignment module 102 first uses structured data from historical schedules to perform time-dimensional label matching on the collected multi-source data. The historical schedule contains the start time, end time, and construction stage identifier for each construction task. The data alignment module 102 extracts timestamp information from the multi-source data and compares the timestamp with the task time periods in the schedule to determine the corresponding construction stage and obtain the time matching result.
[0022] The data alignment module 102 uses the structured data of building information model parameters to perform spatial location mapping of multi-source data. The structured data of BIM parameters includes the three-dimensional coordinates of components, spatial hierarchy information, and component numbers. Based on information such as image shooting angle, sensor deployment coordinates, or spatial reference objects in video frames, the data alignment module 102 maps the multi-source data to specific BIM components or areas to determine the spatial matching result.
[0023] The data alignment module 102 uses structured construction plan data to perform task semantic alignment on multi-source data. The construction plan includes a list of tasks to be performed within the current time period, work areas, responsible personnel, and other information. The data alignment module 102 analyzes behavioral keywords or object categories in video content, audio data, or text records to determine whether they match the tasks in the construction plan, and determines the task progress matching result accordingly.
[0024] Finally, based on the aforementioned time matching results, spatial matching results, and task progress matching results, the data alignment module 102 jointly encodes the multi-source data to obtain a unified multimodal feature vector. This feature vector is used by the subsequent multimodal analysis module 103 to achieve comprehensive identification and assessment of risks at the construction site.
[0025] In some examples, the data alignment module 102 is also used to establish cross-modal mapping relationships between entity objects contained in multi-source data based on time matching results, spatial matching results, and task progress matching results, so as to achieve consistent identification of the same construction entity under different modalities. The entity objects include at least one of construction personnel, construction equipment, and work areas.
[0026] Specifically, the multi-source data collected by the system may include: human or equipment targets extracted from images or video streams, names and roles appearing in voice data (such as "Team Leader Zhang" or "welder"), personnel numbers, work areas or equipment codes mentioned in text data (such as construction logs or task records), and unique identifiers of personnel or equipment from access control or positioning systems (such as ID numbers, RFID tags, work card numbers, etc.).
[0027] The data alignment module 102 first filters out multimodal data fragments that appear within the same time period based on the time matching results; then, based on the spatial matching results, it determines whether these data involve the same BIM component or construction area; and finally, combined with the task progress matching results, it analyzes whether the behaviors reflected in the data belong to the same work task.
[0028] When two or more data fragments from different modalities show significant consistency across the three dimensions mentioned above, the system can determine that they involve the same entity and establish a unified cross-modal mapping relationship within the system. For example:
[0029] If the voice data identifies “Team Leader Zhang is tying steel bars”, and the image analysis identifies a person wearing a work badge with the number 105, and the time and space match, then the system can map “Team Leader Zhang” = “ID 105”.
[0030] For example, if the video identifies that the tower crane equipment number "CR-22" is in operation, and the scheduling log records the lifting operation for that period as "CR-22 lifting template", the system can establish a semantic mapping between the equipment entity and the image.
[0031] This cross-modal mapping relationship can be used for subsequent applications such as individual-level risk identification, behavioral anomaly detection, and liability attribution analysis.
[0032] In some examples, the multimodal analysis module 103 is used to fuse multimodal feature vectors with information from historical operation data, historical case libraries and construction plan documents to construct a unified representation reflecting the behavior, location and state sequence of entities at the construction site. Through a pre-trained multimodal risk identification model, it outputs risk identification results, schedule deviation prediction information and abnormal behavior analysis results to identify potential risks, predict schedule deviations and detect abnormal behaviors at the site.
[0033] Specifically, the multimodal analysis module 103 receives the multimodal feature vector generated by the data alignment module 102 and fuses it with multiple historical information sources. These historical information sources include, but are not limited to: historical work data (such as execution records of completed tasks and worker shift data), a construction case library (including past accidents, abnormal events, and their contexts), and construction plan documents (such as weekly plans and responsibility assignment tables). By introducing the aforementioned structured and unstructured information, the system can establish richer scenario contexts.
[0034] The multimodal analysis module 103 further establishes time-series representations of the behavior, location, and state of key entities (such as construction personnel, equipment, and work areas) at the construction site. Specifically, it can construct a temporal feature chain of "who, where, and what" based on target detection results in video data, location information in sensor data, and operation instructions in voice / text, for subsequent analysis.
[0035] Based on this, the multimodal analysis module 103 calls a pre-trained multimodal risk identification model to perform risk analysis on the above fused representation. This model can be a Transformer architecture, a graph neural network, or a multimodal deep neural network with a fused attention mechanism, and has the ability to fuse cross-modal features and understand context.
[0036] Ultimately, the multimodal analysis module 103 can output one or a combination of the following analysis results: risk identification results, such as detecting "working at height without wearing a safety belt" or "lacking edge protection"; schedule deviation prediction results, such as predicting "Building 4 is expected to be delayed by 2 days" based on the current site conditions; and abnormal behavior analysis results, such as identifying "multiple workers gathered without corresponding work order records" or "a certain piece of equipment in an abnormal start-stop state for an extended period". These analysis results can be used to trigger alarm mechanisms, generate daily reports, or be integrated with the project management system to achieve risk prevention and resource optimization.
[0037] In some examples, the system also includes a decision suggestion module, which generates rectification strategies based on the risk identification results, and the rectification strategies are used to guide on-site risk management and construction adjustments.
[0038] Specifically, after receiving the risk identification results, the decision suggestion module first analyzes information such as risk type, affected area, and related entities. For example, if the risk of "unprotected work at height" is identified, the corresponding work area, the identification of the construction personnel involved, and the risk level label are extracted.
[0039] Based on this, the decision-making suggestion module can match applicable rectification suggestions based on a preset rule base or strategy template. The rule base may include industry safety standards, enterprise-level construction standards, safety management regulations, etc. For example, if the identification result is "unprotected work at height", the system can automatically generate "suspend construction of this work unit and notify the safety officer to check the protective facilities"; if the identification result is "multiple workers gathered but no work order record", the system can generate "check the current personnel distribution and work schedule table to identify illegal work behavior"; if it is "equipment conflict risk", it can suggest "adjust the work sequence, prioritize the completion of concrete pouring in area A before starting hoisting in area B".
[0040] In some implementations, the decision-making suggestion module can also combine historical case databases and construction plan documents to make a comprehensive judgment on the current stage of risk and resource conflicts, thereby generating dynamic, multi-step rectification suggestions, such as a combination strategy of "immediate suspension + investigation + rearrangement of construction schedule".
[0041] The generated rectification strategies can be presented to project leaders and safety supervisors through the management system interface, or automatically pushed to the mobile terminals of relevant responsible persons, realizing a closed-loop response from risk identification to risk handling.
[0042] In some examples, the decision recommendation module includes a large language model (such as a Transformer model with thought chain capabilities) reasoning unit, which generates rectification strategies based on risk identification results, construction specification data, scheduling rules and resource constraint information, and outputs interpretable reasons corresponding to the rectification strategies.
[0043] Specifically, the input to the large language model inference unit includes the risk identification results output by the multimodal analysis module 103, as well as construction specification data, scheduling rules, and resource constraint information pre-loaded or dynamically invoked in the system. The construction specification data may include national or industry safety standard clauses, the scheduling rules may include task priority and work sequence constraints, and the resource constraint information may cover equipment availability, material delivery time, personnel schedules, etc.
[0044] The large language model inference unit first structures or semanticizes the above information to construct contextual prompts, generating language model input templates. For example, the model can receive the following prompt sequence: Risk identification result: There is currently "working at height without wearing a safety belt"; Current work area: East side of Building 4; Scheduling plan: The main steel reinforcement binding in this area is scheduled to be completed today; Rainfall forecast: Moderate rain is expected this afternoon; Safety regulations require: Safety belts must be worn when working above 6 meters; Objective: Propose rectification suggestions and explain the reasons.
[0045] Based on the above prompts, the large language model reasoning unit performs multiple rounds of reasoning, comprehensively considering risk level, task importance, resource scheduling and safety regulations, and outputs specific rectification suggestions and explainable reasons.
[0046] For example, the model might generate the following suggested output: Corrective action: "Suspend work at height in this area, notify the safety officer to conduct an on-site inspection, and postpone the task until the next morning." Explanatory reason: "Due to rainfall today and the personnel not wearing safety belts, it does not comply with Article 5.2 of the 'Safety Regulations for Work at Heights,' so it is recommended to postpone the work to ensure safety compliance."
[0047] In some implementations, the inference unit also supports supplementary inference based on natural language questions input by the user. For example, the user can query "What tasks need to be reordered?" or "Why is it recommended to pause the job?" The system will dynamically generate a response based on the current context, further enhancing the interpretability and flexibility of human-computer interaction.
[0048] The rectification strategies and their rationale can be displayed in the front end of the construction management system with pictures and text, or pushed to the mobile terminals of relevant responsible persons to assist in real-time decision-making and execution at the construction site.
[0049] In some examples, the system also includes a scheduling and interaction module, which receives the rectification strategy output by the decision suggestion module and generates response information corresponding to the interaction information based on the interaction information and the rectification strategy.
[0050] Specifically, the scheduling and interaction module receives the rectification strategy output by the decision suggestion module and generates response information based on the user's input interaction information. Interaction information can include user requests provided through natural language commands, voice input, or image uploads. For example, a user can ask "Why should the hoisting task be paused?" via voice, or click on a work area in the management terminal to ask a natural language question such as "Request to adjust the work sequence".
[0051] Upon receiving the aforementioned interactive information, the scheduling and interaction module can invoke the language understanding component to perform semantic parsing of the input information. Combining this with the current system status, the generated rectification strategy, and the scheduling context, it can generate structured response information. Response information may include, but is not limited to: natural language explanations generated for user queries, such as "Due to the current thunderstorm warning and incomplete protective facilities in the area, it is recommended to suspend hoisting operations"; assignment suggestions generated for task adjustment requests, such as "The formwork installation task for Area B has been postponed to July 18th and will be handled by the construction personnel of Group C"; and status responses generated for progress confirmation requests, such as "The rebar tying work for Building 4 is 80% complete and is expected to end today."
[0052] In some implementations, the scheduling and interaction module can also push the generated response information to relevant construction personnel or managers through terminal applications, wearable devices or scheduling dashboards, and collect task execution feedback in real time (such as completion status, whether it is delayed, whether it is modified). After sorting the feedback information together with the user interaction records, it is sent back to the analysis module or decision suggestion module for the next round of risk assessment and strategy optimization.
[0053] Through the above methods, the scheduling and interaction module not only realizes multi-round human-computer interaction capabilities, but also builds a closed-loop mechanism from suggestion generation to execution feedback, thereby improving the intelligent management and real-time response capabilities of the construction site.
[0054] This application also provides a risk identification method, the method comprising:
[0055] S201. Collect multi-source data from different dimensions at the construction site;
[0056] S202. Based on pre-set structured data, perform data alignment on multi-source data of different dimensions to obtain time matching results, spatial matching results and task progress matching results. Based on the time matching results, spatial matching results and task progress matching results, perform joint encoding on the multi-source data to obtain the multimodal feature vector.
[0057] S203. Input the multimodal feature vectors into the pre-trained multimodal risk identification model to perform risk analysis on the construction site and obtain the risk identification results.
[0058] In some examples, the structured data includes historical schedule structured data, building information model (BIM) parameter structured data, and construction plan structured data. Based on pre-set structured data, data alignment is performed on multi-source data of different dimensions to obtain time matching results, spatial matching results, and task progress matching results. Based on the time matching results, spatial matching results, and task progress matching results, the multi-source data is jointly encoded to obtain the multimodal feature vector. This includes: performing time-dimensional label matching on the multi-source data based on the historical schedule structured data to determine the time matching result; using the BIM parameter structured data to perform spatial location mapping on the multi-source data to determine the spatial matching result; using the construction plan structured data to perform task semantic alignment on the multi-source data to determine the task progress matching result; and jointly encoding the multi-source data based on the time matching result, spatial matching result, and task progress matching result to obtain the multimodal feature vector.
[0059] In some examples, the method also includes establishing cross-modal mapping relationships between entity objects contained in multi-source data based on time matching results, spatial matching results, and task progress matching results.
[0060] In some examples, multimodal feature vectors are input into a pre-trained multimodal risk identification model to perform risk analysis on the construction site and obtain risk identification results. This includes: fusing multimodal feature vectors with information from historical operation data, historical case databases, and construction plan documents to construct a unified representation reflecting the entity behavior, location, and state sequence of the construction site; and outputting risk identification results, schedule deviation prediction information, and abnormal behavior analysis results through the pre-trained multimodal risk identification model.
[0061] In some examples, the method also includes generating remediation strategies based on risk identification results, which are used to guide on-site risk management and construction adjustments.
[0062] In some examples, rectification strategies are generated based on risk identification results, including: generating rectification strategies based on risk identification results, construction specification data, scheduling rules and resource constraint information, and outputting interpretable reasons corresponding to the rectification strategies.
[0063] In some examples, the method also includes: receiving the rectification strategy output by the decision suggestion module, and generating response information corresponding to the interaction information based on the interaction information and the rectification strategy.
[0064] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.
[0065] Figure 3 This is a schematic diagram of the electronic device 3 provided in an embodiment of this application. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various system embodiments and the functions of each method embodiment described above.
[0066] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or different components.
[0067] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
[0068] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or RAM. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. The memory 302 can also include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.
[0069] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0070] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, user code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in a computer-readable medium can be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.
[0071] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A risk identification system, characterized in that, The system includes: The scene perception module is used to collect multi-source data from different dimensions at the construction site; The data alignment module is used to perform data alignment on the multi-source data of different dimensions based on pre-set structured data to obtain time matching results, spatial matching results and task progress matching results. Based on the time matching results, the spatial matching results and the task progress matching results, the multi-source data is jointly encoded to obtain the multimodal feature vector. The multimodal analysis module is used to input the multimodal feature vectors into a pre-trained multimodal risk identification model to perform risk analysis on the construction site and obtain risk identification results.
2. The system according to claim 1, characterized in that, The structured data includes historical schedule structured data, building information model parameter structured data, and construction plan structured data. The data alignment module is also used to perform time-dimensional label matching on the multi-source data based on the historical schedule structured data to determine the time matching result. Using the structured data of the building information model parameters, spatial location mapping is performed on the multi-source data to determine the spatial matching result; Using the structured data of the construction plan, the multi-source data is semantically aligned to determine the task progress matching result.
3. The system according to claim 2, characterized in that, The data alignment module is also used to establish cross-modal mapping relationships between entity objects contained in the multi-source data based on the time matching results, the spatial matching results, and the task progress matching results.
4. The system according to claim 1, characterized in that, The multimodal analysis module is used to integrate the multimodal feature vectors with information from historical operation data, historical case databases and construction plan documents to construct a unified representation reflecting the behavior, location and state sequence of entities at the construction site. Through a pre-trained multimodal risk identification model, it outputs the risk identification results, schedule deviation prediction information and abnormal behavior analysis results.
5. The system according to claim 1, characterized in that, The system also includes a decision suggestion module, which is used to generate rectification strategies based on the risk identification results, and the rectification strategies are used to guide on-site risk management and construction adjustments.
6. The system according to claim 5, characterized in that, The decision suggestion module includes a large language model reasoning unit, which is used to generate the rectification strategy based on the risk identification results, construction specification data, scheduling rules and resource constraint information, and output the interpretable reasons corresponding to the rectification strategy.
7. The system according to claim 5, characterized in that, The system further includes a scheduling and interaction module, which receives the rectification strategy output by the decision suggestion module and generates response information corresponding to the interaction information based on the interaction information and the rectification strategy.
8. A risk identification method, characterized in that, The method includes: Collect multi-source data from different dimensions at the construction site; Based on pre-set structured data, the multi-source data of different dimensions are aligned to obtain time matching results, spatial matching results and task progress matching results. Based on the time matching results, spatial matching results and task progress matching results, the multi-source data are jointly encoded to obtain the multimodal feature vector. The multimodal feature vectors are input into a pre-trained multimodal risk identification model to perform risk analysis on the construction site and obtain risk identification results.
9. The method according to claim 8, characterized in that, The method further includes: Based on the risk identification results, a rectification strategy is generated, which is used to guide on-site risk management and construction adjustments.
10. The method according to claim 9, characterized in that, The method further includes: Receive the rectification strategy and generate response information corresponding to the interaction information based on the interaction information and the rectification strategy.
Citation Information
Cited By
Intelligent material scheduling decision-making method and system based on large language model
CN122048072A
Intelligent material scheduling decision method and system based on large language model
CN122048072B
