A third-party construction operation intelligent safety evaluation method and system

Abstract

This invention relates to the field of construction operation technology and discloses a third-party intelligent safety evaluation method and system for construction operations, comprising the following steps: First, collecting multi-source heterogeneous data from the construction site, including visual, structured permits, and dynamic sensing data, constructing a full dataset and using perspective transformation to achieve spatiotemporal alignment between pixel space and geographic space; then, using a large computer vision model to extract construction behavior features, and combining prior knowledge such as pipeline burial depth to identify violations such as exceeding excavation depth limits; furthermore, constructing a quantitative evaluation model based on multiple dimensions such as regional compliance and equipment operation specifications, and calculating a real-time overall safety score; finally, determining the risk level based on the score, and automatically triggering graded control strategies such as early warning push or on-site emergency shutdown. This invention, by constructing a closed-loop system of perception-analysis-evaluation-control, realizes the digital measurement and automated handling of construction risks, effectively improving the safety management level of gas pipeline networks.

Description

Technical Field

[0001] This invention relates to the field of construction operation technology, specifically to a third-party intelligent safety evaluation method and system for construction operations. Background Technology

[0002] With the accelerating pace of urbanization, the coverage of urban gas pipeline networks is constantly expanding, highlighting their crucial role as the "lifeline" of cities. However, the increased investment in urban infrastructure construction has also led to frequent third-party construction work, exposing gas pipelines to severe risks of external damage. Statistics show that third-party construction damage has become the leading cause of serious accidents such as gas pipeline leaks and explosions, severely threatening public safety and the stability of energy supply. Therefore, implementing comprehensive and precise safety management of third-party construction work along pipelines is a critical issue that the gas industry urgently needs to address.

[0003] Traditional third-party construction management models primarily rely on manual inspections and fixed-point monitoring. This approach is not only costly in terms of manpower but also limited by personnel's energy and inspection cycles, making it difficult to achieve 24 / 7 coverage of wide-area pipeline networks without blind spots. It is also highly susceptible to accidents caused by damage during inspection breaks. Furthermore, manual monitoring relies mainly on the experience of on-site personnel for visual judgment, lacking objective and quantifiable evaluation standards. This often results in delayed detection, difficulty in obtaining evidence, and ineffective warnings for excavator operators' violations, leading to significant subjectivity and uncertainty in the management effectiveness.

[0004] Although video surveillance technology has been gradually applied to construction site supervision in recent years, most existing monitoring methods are limited to the acquisition and transmission of video images, or only have basic human or vehicle detection functions. The main drawback of existing technology is its inability to achieve deep integration of visual space and geographic space. Surveillance cameras capture two-dimensional pixel images, while underground pipelines and construction areas exist in a three-dimensional geographic coordinate system, lacking a unified spatial mapping relationship between the two. This makes it impossible for the system to accurately calculate the actual physical distance between construction machinery and underground pipelines, let alone perceive the key risk dimension of excavation depth. Often, monitoring personnel can only make rough estimates through the screen, making it difficult to effectively identify hidden risks such as excavator buckets intruding into pipeline restricted excavation areas or over-excavation.

[0005] Furthermore, current regulatory systems often suffer from data silos, with on-site visual perception data failing to logically integrate with backend administrative licensing data and pipeline GIS data. The system cannot automatically determine whether the current construction scope exceeds approved limits, nor can it create a comprehensive profile of the construction unit based on historical violation records. This lack of multi-dimensional data fusion and quantitative evaluation mechanisms leaves safety management in a reactive state, lacking the ability to dynamically assess and tieredly control risk levels, and unable to achieve automated prevention and intervention at critical moments of accidents.

[0006] Therefore, in order to address the above shortcomings, a third-party intelligent safety evaluation method and system for construction operations is proposed. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a third-party intelligent safety evaluation method and system for construction operations. It solves the technical problems in existing third-party construction supervision technologies, such as the lack of accurate mapping between visual monitoring images and physical geographic space, the inability to quantify and monitor hidden risks such as excavation depth and spatial distance, and the lack of multi-source data fusion and objective evaluation mechanisms, which lead to delayed risk perception and strong subjectivity in control.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a third-party intelligent safety evaluation method and system for construction operations. The first aspect of this invention provides an intelligent safety evaluation method for third-party construction operations, comprising: collecting multi-source heterogeneous data from the third-party construction site and establishing a unified data input vector; using a pre-trained computer vision model to extract features from the multi-source heterogeneous data, identifying construction behavior characteristics and violation status; calculating a total safety score for the third-party construction operation based on the construction behavior characteristics and violation status, combined with a preset multi-dimensional quantitative evaluation model; determining the construction risk level based on the total safety score, and implementing a corresponding graded control strategy according to the construction risk level.

[0009] Preferably, the multi-source heterogeneous data From visual data Structured data and dynamic data Composition, represented as: ; Among them, the visual data The structured data is a sequence of keyframes from a video stream acquired by intelligent monitoring equipment and processed with defogging and image stabilization. It includes a feature vector consisting of the construction unit, permitted scope, permitted period, pipe diameter, pipe pressure, and pipe burial depth; the dynamic data Includes real-time GPS coordinates of construction machinery and on-site personnel location information .

[0010] Preferably, when establishing a unified data input vector, the construction permit number is used as the primary key to associate the visual data, structured data, and dynamic data; the pixel coordinates in the visual data are mapped to the geographic coordinate system using a GIS system to achieve spatial alignment between the image area and the pipeline protection zone.

[0011] Preferably, identifying construction behavior characteristics and violations specifically includes regional compliance testing and operational standardization identification. In regional compliance testing, the pipeline centerline is defined as... Calculate the construction object Euclidean distance from the centerline of the pipeline The regional attributes of the construction object are determined based on the Euclidean distance. :

[0012] ; in, This indicates a no-excavation zone. Indicates a restricted excavation area. Indicates the safe zone. and This is a preset distance threshold.

[0013] In operational compliance identification, detecting and mining deep violations. 1. Unauthorized hot work and characteristics of protective facilities; among them, if the depth of the equipment bucket trajectory greater than the pipeline burial depth ,determination If hot work is detected and the distance is less than the preset safety threshold, then... ; Check the clarity of warning signs and the status of guardians on duty To assess protective facilities.

[0014] Preferably, the multi-dimensional quantitative evaluation model is calculated using a linear weighting method, and the total safety score is... Defined as: ; in, Indicates the first Scores for each dimension. The specific dimension calculation rules are as follows:

[0015] Regional compliance score Based on the cumulative duration of the construction object's intrusion into the prohibited or restricted excavation zone. And the deduction coefficient α per unit time is calculated, with a weight of W1: ; Equipment Operation Standards Score (S2): Based on the cumulative number of violations and the deduction coefficient for a single violation Calculate, weight is : ; Integrity score of protective measures (s3): based on the compliance rate of warning signs and the on-duty rate of guardians Calculate, weight is , This is a single deduction value. Indicator function: ; Permission matching score Based on the proportion of the actual construction scope exceeding the permitted scope. and penalty ladder coefficient Calculate, weight is : ; Historical compliance record score Based on the number of historical violations by the construction company and the deduction coefficient for a single historical violation Calculate, weight is : ; Preferably, when determining the construction risk level, a hierarchical mapping function is used. The overall safety score is mapped to four levels: Excellent, Satisfactory, Risky, and High Risk. ; The control strategy implemented based on the construction risk level includes: generating an early warning signal when Level=Risk. The message is sent to the relevant personnel; when Level=HighRisk, an emergency blocking signal is triggered. Activate on-site audible and visual alarms and generate dispatch instructions.

[0016] A second aspect of this invention provides a third-party construction operation intelligent safety evaluation system, comprising: a data acquisition and processing module for acquiring multi-source heterogeneous data from the third-party construction operation site and establishing a unified data input vector; a behavior intelligent analysis module for extracting features from the multi-source heterogeneous data using a pre-trained computer vision model to identify construction behavior characteristics and violation status; a quantitative evaluation calculation module for calculating the overall safety score of the third-party construction operation based on the construction behavior characteristics and violation status, combined with a preset multi-dimensional quantitative evaluation model; and a risk classification and control module for determining the construction risk level based on the overall safety score and executing corresponding classification and control strategies according to the construction risk level.

[0017] This invention provides a third-party intelligent safety evaluation method and system for construction operations. It has the following beneficial effects:

[0018] 1. This invention overcomes the limitation of traditional video surveillance, which can only present two-dimensional planar images, by establishing a mapping model between the pixel coordinate system and the geographic coordinate system based on the homography matrix. This achieves spatiotemporal alignment between visual data and underground pipeline GIS data. This technique enables the system to directly and accurately calculate the Euclidean distance between the construction machinery and the pipeline centerline in physical space. Therefore, without relying on visual estimation by on-site personnel, it can accurately determine whether an excavator has entered a prohibited or restricted excavation zone, effectively solving the technical problem that visual surveillance cannot quantitatively perceive spatial location.

[0019] 2. This invention innovatively introduces a digging depth monitoring logic based on the geometry of the robotic arm, which can calculate the cutting depth of the excavator bucket tip in real time and dynamically compare it with the pipeline burial depth. This mechanism fills the gap in existing technology for underground depth perception, allowing the monitoring line of sight to penetrate the ground surface and identify the risk of exceeding the depth limit during the digging stage before the bucket contacts the pipeline. This transforms post-event accountability into pre-event prevention, greatly reducing the probability of pipeline damage accidents caused by blindly digging too deep.

[0020] 3. This invention constructs a quantitative evaluation mathematical model encompassing multiple dimensions, including regional compliance, permit matching, and historical credit, transforming discrete and fuzzy construction site performance into continuous and objective digital scoring. Through linear weighted calculation, this method eliminates judgment biases caused by individual experience differences and subjective emotions in traditional manual inspections. It not only focuses on current violations but also integrates the scope of administrative approvals and the historical profile of construction units, providing gas management departments with a scientific, fair, and traceable standardized evaluation system.

[0021] 4. This invention designs a closed-loop response mechanism for risk grading and automated control, directly mapping the calculated total safety score to differentiated control strategies. When a high-risk state is determined, the system can skip the manual reporting process and trigger the on-site audible and visual alarm devices via the IoT interface in milliseconds, simultaneously generating an emergency blocking work order. This automated "from perception to action" link significantly reduces emergency response time, ensuring effective physical intervention at the critical point of an accident.

[0022] 5. This invention integrates a model adaptive iteration mechanism based on negative sample feedback, using manually reviewed data to continuously fine-tune the behavior recognition model. This design enables the system to self-evolve, continuously adapting to complex and changing construction environments and lighting conditions as operating time increases, dynamically correcting evaluation weights and discrimination thresholds, thereby effectively suppressing false alarms and false negatives, and ensuring the robustness and reliability of the system in long-term unattended environments. Attached Figure Description

[0023] Figure 1 This is the main flowchart of the intelligent safety evaluation method for third-party construction operations of the present invention; Figure 2 This is a functional module block diagram of the third-party intelligent safety evaluation system for construction operations according to the present invention. Detailed Implementation

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see the appendix Figure 1-2 This invention provides a third-party intelligent safety evaluation method and system for construction operations. This method realizes the digital measurement and automated control of the safety status of construction sites by constructing a closed-loop system of real-time perception, intelligent analysis, quantitative evaluation and risk warning.

[0026] The third-party intelligent safety evaluation method for construction operations mainly includes the following steps: First, the system performs the acquisition and vectorization processing of multi-source heterogeneous data. For third-party construction sites, the system does not simply receive video streams but actively aggregates data from three dimensions: visual, structured permits, and dynamic sensing. To achieve a unified understanding of complex construction scenarios by the computer, a full dataset needs to be constructed. The dataset is a collection of the following three subsets:

[0027] In the formula, The visual data originates from intelligent monitoring equipment deployed around the construction site, specifically as a sequence of keyframes in a video stream after time synchronization and image preprocessing. This represents structured data, which originates from the construction permit database and GIS pipeline database approved by relevant departments, and contains feature vectors that can define compliance boundaries; This represents dynamic data, sourced from positioning terminals mounted on construction machinery and handheld terminals used by on-site monitoring personnel, providing real-time spatial location information. By constructing the aforementioned unified data input vector, the system maps the physical construction scene into a computable digital spatial model.

[0028] After data acquisition, the system performs feature extraction and behavior recognition based on a large-scale computer vision model. This step utilizes a pre-trained computer vision model, fine-tuned with gas industry data, as the core inference engine. The system then processes the entire dataset described above. The input model focuses on construction objects in keyframes through an attention mechanism. During this process, the system uses GIS geographic information technology to map pixel coordinates in visual data to a geographic coordinate system, thereby spatially aligning dynamically extracted construction behavior features with static pipeline protection zone geofences to identify specific violations such as "operation in no-excavation zones," "excessive mechanical excavation," and "absence of monitoring personnel."

[0029] Subsequently, a multi-dimensional quantitative evaluation and hierarchical calculation process is implemented. Unlike the traditional binary "compliance / non-compliance" judgment, this step introduces a quantitative mathematical model to precisely measure construction risks. Based on the behavioral characteristics and non-compliance status identified in the previous steps, the system independently scores the construction site across five orthogonal dimensions: regional compliance, equipment operation standards, completeness of protective measures, permit matching degree, and historical compliance records. A linear weighted average is then used to calculate the current total safety score for the construction site. This score reflects the real-time safety level of the construction site within a specific time window.

[0030] Finally, the system executes risk level assessment and hierarchical control strategies. The system has a pre-defined hierarchical mapping function that discretizes the continuous overall safety score into four mutually exclusive risk levels: Excellent, Qualified, Risky, and High-Risk. Based on the determined risk level, the system automatically triggers differentiated control procedures: for low-risk situations, the system maintains silent monitoring and logs; for medium-risk situations, the system generates an early warning signal and pushes it to the mobile terminals of relevant personnel, prompting rectification; for high-risk situations, the system immediately triggers an emergency blocking signal, on the one hand, linking with on-site audible and visual alarm devices to force work to stop, and on the other hand, generating a highest-priority dispatch instruction to assign the nearest inspection personnel to the site.

[0031] The system is used to perform the above evaluation methods and includes a data acquisition and processing module, a behavioral intelligence analysis module, a quantitative evaluation calculation module, and a risk classification and control module.

[0032] The data acquisition and processing module is configured as the system's perception front end, responsible for connecting to on-site cameras, sensors, and the backend database. It extracts visual, structured, and dynamic data in real time, and performs preprocessing tasks such as defogging, image stabilization, and timestamp synchronization, ultimately outputting a standardized full dataset. .

[0033] The behavioral intelligence analysis module is connected to the data acquisition and processing module and is configured as the system's inference core, incorporating a large-scale computer vision model and coordinate transformation algorithm. This module receives the full dataset, extracts multi-scale features through a feature pyramid network, and combines prior knowledge such as pipeline burial depth and protected area boundaries to output a structured behavioral feature description containing target category, location relationship, and action attributes.

[0034] The quantitative evaluation calculation module is connected to the behavioral intelligence analysis module and is configured as the system's decision operator, storing the weight parameters and deduction logic of the multi-dimensional quantitative evaluation model. Based on the received behavioral characteristics, this module calculates the score for each evaluation dimension and aggregates them to generate a total safety score, thus realizing the conversion from qualitative features to quantitative values.

[0035] The risk grading and control module is connected to the quantitative evaluation and calculation module and is configured as the system's execution terminal. This module maintains a risk level mapping table and a control strategy library. Based on the input score, it determines the current risk level and sends corresponding control commands or early warning information to downstream alarm devices or communication gateways via an IoT interface.

[0036] To achieve accurate perception of all elements at third-party construction sites, the system needs to process data encompassing high-dimensional visual information, static business attribute information, and real-time spatiotemporal trajectory information. Establishing a unified data input vector is the foundation for subsequent intelligent analysis and quantitative evaluation.

[0037] In this invention, a full dataset is defined. From visual data Structured data and dynamic data Together they constitute. For visual data... This data originates from intelligent monitoring equipment deployed around the construction site. The intelligent monitoring equipment collects raw video streams. Considering the complex environment of construction sites, often accompanied by dust or equipment vibration, the system first performs dehazing processing based on dark channel prior and electronic image stabilization based on optical flow on the original video stream. The processed video stream is then sampled at preset time intervals to extract keyframe sequences. ,in Representing the Image frame data at each moment.

[0038] For structured data The data primarily originates from the administrative approval database and pipeline geographic information system of the gas management department. The system retrieves relevant information through a data interface to construct structured feature vectors. This feature vector contains the construction unit. Permitted scope of work Permitted operation period Pipe diameter Pipeline operating pressure and pipeline burial depth These attribute data provide benchmark parameters for determining whether construction activities are compliant, especially pipeline burial depth. This will be directly used as the threshold for subsequent judgment of violations in mining depth.

[0039] For dynamic data It includes real-time location information of construction machinery and personnel. Using positioning terminals installed on large machinery such as excavators and cranes, the real-time coordinate sequence of the construction machinery is obtained. ; Use the smart badges or handheld terminals worn by on-site monitoring personnel to obtain personnel location information. All coordinates are uniformly based on the WGS-84 coordinate system, including longitude, latitude, and elevation data, and are used to locate the work entity in geospatial space.

[0040] After acquiring the three types of data mentioned above, the system uses the construction permit number as a unique primary key to logically associate the unstructured visual data, structured permit data, and streaming dynamic data, forming data packets corresponding to the same construction event. However, the visual data is in a pixel coordinate system, while the pipe locations in the structured data and the personnel and machinery locations in the dynamic data are in a geographic coordinate system, exhibiting spatial heterogeneity. Therefore, this embodiment employs a coordinate mapping algorithm based on perspective transformation to achieve precise alignment from pixel space to geographic space.

[0041] The system selects several fixed control points in the keyframe image and obtains their pixel coordinates in the image. and their corresponding real geographic coordinates Based on these control point pairs, the homography matrix or perspective transformation matrix is ​​solved using the least squares method. The matrix It describes the mapping relationship from a two-dimensional image plane projection to a three-dimensional geographic surface.

[0042] For any pixel detected in the image Convert it to homogeneous coordinate form By transforming the matrix Calculate its corresponding intermediate vector in the geographic coordinate system. : In the formula, for The transformation matrix is ​​then used. After obtaining the intermediate vector, the actual geographic longitude corresponding to the pixel is calculated through normalization. and latitude :

[0043] Through the aforementioned mapping transformation, the system can directly project the geofence data of the pipeline protection zone onto the video surveillance screen, or conversely, map the pixel location of the excavator identified in the video onto the electronic map. This spatiotemporal alignment mechanism allows the system to directly calculate the physical distance between the visually identified object and the underground concealed pipeline, thus providing accurate quantitative evidence for subsequent determination of whether the construction object has encroached on the no-excavation zone, eliminating the errors caused by relying solely on visual distance estimation in traditional monitoring.

[0044] In this invention, a large-scale computer vision model fine-tuned with domain-specific data is used as the core inference engine. The system inputs pre-processed keyframes from the video stream into the model, extracting high-dimensional features of the image using convolutional neural networks or Transforer architectures. The model's primary task is to perform target detection and instance segmentation on key elements in the construction scene, identifying objects including various construction machinery, workers, safety facilities, and abnormal environmental features. For each identified construction object... The model outputs its pixel region and category label in the image.

[0045] Based on the mapping relationship between the pixel coordinate system and the geographic coordinate system established in the previous embodiments, the system can calculate the location of the identified object in physical space. To determine regional compliance, the system first reads the pipeline geographic information recorded in the structured data and reconstructs the pipeline centerline in virtual geographic space. Subsequently, the system calculates the construction object in real time. The geometric center or critical work point and the pipeline centerline Euclidean distance between This distance calculation is based on a unified geographic coordinate projection, eliminating errors caused by camera perspective distortion.

[0046] Based on the calculated Euclidean distance, the system divides the work area containing the construction object into geofences with different risk levels. Define the region determination function. as follows:

[0047] In the formula, No-excavation zones, indicating areas where mechanical operations are strictly prohibited. This indicates a restricted excavation area that requires manual assistance or involves limited operations. Indicates the safe work area; and This is based on a preset distance threshold according to gas pipeline protection regulations. The system continuously tracks the status of the construction site, and when it determines that the construction site is located... or At that time, a timer is started to accumulate the duration of its stay and operation in the area, which will serve as the input for subsequent quantitative evaluation.

[0048] When identifying specific violations in excavation operations, this invention specifically targets the monitoring of "excavation depth," a hidden risk characteristic. Since underground pipelines have a defined burial depth, the system utilizes pose estimation technology from computer vision to identify key nodes of the construction machinery's boom, forearm, and bucket. Combined with prior geometric parameters of the boom length or binocular vision depth information, it estimates the vertical cutting depth of the bucket tip relative to the ground plane. The system will compare the real-time excavation depth with the pipe burial depth recorded in the structured data. Perform a comparison. When detected Values ​​exceeding Furthermore, when the location is within the projection area above the pipeline, the system determines that there is a high-risk deep violation and generates a corresponding binary feature flag.

[0049] Furthermore, the system executes a multi-task identification process in parallel to assess the effectiveness of hot work operations and safety protection measures. For hot work violation identification, the model determines the location of the hot work point by capturing the characteristics of welding arc light, open flame, or continuous smoke. If the distance between this location and the pipeline centerline is less than the preset safe fire prevention distance, the system determines it as a violation of hot work regulations. For the integrity identification of protective facilities, the model detects whether standardized warning signs are set up on site and calculates the pixel clarity of the sign patterns. To assess its visibility; simultaneously, using facial recognition or workwear recognition technology, the system detects and continuously tracks monitoring personnel in sequence frames, and generates monitoring personnel on-duty status indicators by analyzing their on-duty duration and off-duty frequency. All the identified behavioral characteristics and violation statuses are structured into standard message bodies and then transferred to the subsequent quantitative evaluation module for comprehensive scoring.

[0050] In this invention, a total safety score for third-party construction work is defined. The total score is the sum of scores from each independent evaluation dimension. The system pre-sets five orthogonal evaluation dimensions, corresponding to regional compliance, equipment operation standards, completeness of protective measures, license matching degree, and historical compliance records. The formula for calculating the total score is as follows:

[0051] In the formula, Indicates the first The system calculates scores for each dimension and assigns a corresponding weight value to each dimension. Furthermore, the sum of the initial weights for all dimensions equals the maximum score. The specific calculation logic for each dimension is as follows.

[0052] For the regional compliance dimension, the system focuses on assessing instances of construction machinery or personnel encroaching on the pipeline protection zone. The score for this dimension is... It depends on the duration of the violation. The system calculates the cumulative time that the construction object has been in the prohibited or restricted excavation zone. And introduce a deduction coefficient per unit time. The calculation formula is:

[0053] In the formula, This represents the base weight score for this dimension; function This is used to ensure that the score for this dimension is non-negative; that is, when the deduction exceeds the base weight, the score for this item is zero, thus preventing negative scores from affecting the evaluation of other dimensions. This logic reflects a time-cumulative penalty mechanism for persistent violations.

[0054] Regarding equipment operation procedures, the system assesses the frequency of specific violations during construction. The score for this dimension is... Primarily based on the cumulative number of violations. Calculations are performed. The violations include, but are not limited to, digging deeper than the pipeline's burial depth or conducting hot work in a fire-restricted area. The system sets a penalty coefficient for each violation. The calculation formula is:

[0055] In the formula, This is the base weight score for this dimension. Unlike regional compliance, this dimension focuses on the counting and penalty of discrete violations; each identified and confirmed violation will directly lead to a step-by-step decrease in the score.

[0056] Regarding the integrity of protective measures, the system assesses the configuration of on-site safety facilities. The score for this dimension is... A point deduction system is used for missing items, mainly assessing the compliance rate of warning signs. and the on-duty rate of guardians If warning signs are missing, blurred, or supervisors are absent from their posts, the standard is considered not met. Individual deduction values ​​are set. The calculation formula is:

[0057] In the formula, This is the base weight score for this dimension; This is an indicative function; its value is 1 when the condition within the parentheses is met (i.e., not met), and 0 otherwise. This formula ensures that the absence of any safety measure will be directly reflected in the final evaluation.

[0058] Regarding the permit matching dimension, the system evaluates the consistency between the actual construction scope and the approved permit scope. The score for this dimension is... Based on the proportion of the actual construction area exceeding the permitted area Calculations are performed. To reflect strict control over construction beyond permitted scope, the model employs a tiered penalty logic, setting a penalty tier coefficient λλ, calculated as follows:

[0059] In the formula, This is the base weight score for this dimension; This represents the floor function. The physical meaning of this formula is that whenever the excess proportion increases by a preset step, a corresponding score is deducted, thereby quantifying the degree to which the construction activity deviates from the administrative approval boundary.

[0060] For the historical compliance record dimension, the system introduces a credit profile for the construction unit. The score for this dimension... Based on the number of violations recorded by the construction company in the historical database Confirmed. Set the penalty coefficient for a single historical violation. The calculation formula is:

[0061] In the formula, This serves as the base weight score for this dimension. The introduction of this dimension extends the focus from "managing tasks" to "managing people," subjecting construction companies with poor historical credit records to more stringent initial risk assessments upon entering the site, thereby forcing them to improve their daily management. Through the combined calculation of these five dimensions, the system ultimately outputs an objective and quantifiable overall safety score, providing data support for subsequent tiered management and control.

[0062] In this invention, a hierarchical mapping function is established. The overall safety score of third-party construction work The risk is divided into four mutually exclusive risk zones. The system has three preset threshold levels: an excellent threshold, a low threshold, and a high threshold. Qualified threshold and risk threshold Based on the comparison between the overall safety score and the aforementioned threshold values, the current risk level (Level) of the construction site is determined.

[0063] In the formula, "Excellent" represents a state where construction meets standards and safety is under control; "Qualified" represents a state where minor flaws exist but are within acceptable limits; "Risk" represents a state where there are obvious violations requiring rectification; and "High Risk" represents a state where there are serious violations and are highly likely to cause accidents. Through this mapping function, the system intuitively transforms complex scoring results into business language that is easy for management to understand.

[0064] Based on the determined risk level, the system is configured with automated hierarchical control strategy execution logic. When the risk level is determined to be Excellent or Qualified, the system maintains the normal monitoring mode, only archiving the relevant scoring data and on-site screenshots to the log database as the basis for subsequent credit evaluation, without triggering active interference.

[0065] When the risk level is determined to be "Risk", the system generates an early warning signal. The signal includes a snapshot image of the violation, the time of occurrence, and the corresponding point deduction details, which is instantly pushed to the mobile terminals of the construction site supervisor and gas inspector via mobile communication network. The system also starts a rectification timer; if no feedback indicating rectification completion is received within the preset time or the score does not rise back to the acceptable range, the warning level is automatically upgraded.

[0066] When the risk level is determined to be High Risk, the system determines that the current situation is at a critical point and immediately triggers the highest priority emergency blocking signal. The signal executes two control commands in parallel: one is sent to the smart sound and light alarm installed on site via the Internet of Things interface, triggering a high-decibel alarm and flashing red light to physically warn workers to stop construction immediately; the other is sent to the dispatch and command center, which automatically generates an emergency work order and plans the optimal route to dispatch the nearest professional repair or law enforcement personnel to the site for handling.

[0067] Furthermore, to overcome the misjudgment problem of computer vision models in complex environments, this embodiment also includes an adaptive iterative process for model parameters. The system periodically collects sample data that has been manually reviewed and confirmed as misjudgments, constructing a negative sample set. This negative sample set is used to incrementally train the behavior recognition model, fine-tuning the weight parameters of the feature extraction network. Simultaneously, the system analyzes the correlation between the scores of each evaluation dimension and the actual occurrence of accidents, using regression analysis to dynamically adjust the weights of each item in the quantitative evaluation model. The system incorporates a deduction coefficient, thereby enabling the evaluation results to continuously approximate the true risk distribution and achieve self-evolution and improved accuracy.

[0068] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A third-party intelligent safety evaluation method for construction operations, characterized in that, Includes the following steps: Collect multi-source heterogeneous data from third-party construction sites and establish a unified data input vector; The pre-trained computer vision model is used to extract features from the multi-source heterogeneous data to identify construction behavior characteristics and violation status. Based on the aforementioned construction behavior characteristics and violation status, and combined with a pre-set multi-dimensional quantitative evaluation model, the overall safety score of the third-party construction operation is calculated. The construction risk level is determined based on the overall safety score, and the corresponding graded control strategy is implemented according to the construction risk level.

2. The method according to claim 1, characterized in that, The multi-source heterogeneous data includes visual data, structured data, and dynamic data; The visual data is a sequence of keyframes from a video stream collected by intelligent monitoring equipment. The structured data includes construction permit information and pipeline basic attributes. The construction permit information includes the construction unit, the scope of the permit, and the permit period. The pipeline basic attributes include the pipeline diameter, the pipeline pressure, and the pipeline burial depth. The dynamic data includes the real-time GPS coordinates of the construction machinery and the location information of on-site personnel.

3. The method according to claim 2, characterized in that, The establishment of a unified data input vector specifically includes: using the construction permit number as the primary key to associate the visual data, the structured data, and the dynamic data; By using a GIS system to map the pixel coordinates in the visual data to a geographic coordinate system, the visual data is spatially aligned with the pipeline protection zone.

4. The method according to claim 3, characterized in that, The identification of construction behavior characteristics and violation status, including the identification of regional compliance, includes: determining the pipeline centerline and calculating the Euclidean distance between the identified construction object and the pipeline centerline; The area attribute of the construction object is determined based on the Euclidean distance. The area attribute includes a no-excavation zone, a restricted excavation zone, and a safe zone. The cumulative time length during which the construction object falls into the no-excavation zone or the restricted excavation zone is calculated.

5. The method according to claim 3, characterized in that, The identification of construction behavior characteristics and violation status includes the identification of operational standardization and protection effectiveness: comparing the excavation depth of the construction equipment with the burial depth of the pipeline. When the excavation depth is greater than the burial depth of the pipeline, it is determined to be an excavation depth violation. The distance between the hot work site and the pipeline is detected. If the distance is less than a preset safety threshold, it is determined to be a hot work violation. The system detects the clarity of warning signs at the testing site and the on-duty status of monitoring personnel, generating characteristic values ​​for the protective facilities.

6. The method according to claim 1, characterized in that, The multi-dimensional quantitative evaluation model is calculated using a linear weighting method, and the total safety score is the sum of the scores of each evaluation dimension. The evaluation dimensions include at least: regional compliance, equipment operation specifications, completeness of protective measures, license matching, and historical compliance records.

7. The method according to claim 6, characterized in that, The calculation rules for the scores of each evaluation dimension are as follows: The regional compliance dimension is calculated by deducting points based on the cumulative duration of the construction object's intrusion into the prohibited excavation zone and the deduction coefficient per unit time. The equipment operation standard dimension is: deduction points are calculated based on the cumulative number of violations and the deduction coefficient for each violation; The integrity dimension of the protective measures includes the compliance rate of warning signs and the on-duty rate of monitoring personnel. If the standards are not met, the corresponding points will be deducted. The permitted matching degree dimension: calculates the proportion of the actual construction scope exceeding the permitted scope, and deducts points based on the proportion and the tiered penalty coefficient; The historical compliance record dimension: obtain the number of historical violations by the construction unit, and calculate the deduction points based on the number of historical violations.

8. The method according to claim 7, characterized in that, In the calculation rules, for a dimension that needs to be deducted points, if the calculation result after deduction is less than zero, then the score for that dimension is zero.

9. The method according to claim 1, characterized in that, The implementation of the corresponding graded control strategy based on the construction risk level includes: A preset hierarchical mapping function is used to map the overall safety score into four levels: excellent, qualified, risky, and high-risk. When the construction risk level is classified as "risk", an early warning signal is generated and sent to the relevant responsible person. When the construction risk level is high, an emergency blocking signal is triggered, the on-site audible and visual alarm is activated, and a dispatch command is generated.

10. A system for use in a third-party intelligent safety evaluation system for construction operations as described in any one of claims 1 to 9, characterized in that, include: The data acquisition and processing module is used to collect multi-source heterogeneous data from third-party construction sites and establish a unified data input vector. The behavioral intelligence analysis module is used to extract features from the multi-source heterogeneous data using a pre-trained computer vision model, and to identify construction behavior characteristics and violation status. The quantitative evaluation calculation module is used to calculate the overall safety score of the third-party construction operation based on the construction behavior characteristics and violation status, combined with a preset multi-dimensional quantitative evaluation model. The risk classification and control module is used to determine the construction risk level based on the overall safety score, and to execute the corresponding classification and control strategy based on the construction risk level.

Images