Construction site construction track monitoring method and system based on GIS

Abstract

This invention discloses a GIS-based method and system for monitoring construction site trajectories, belonging to the field of data processing technology. It includes: dynamically adjusting a preset monitoring sliding window based on the construction risk index of high-altitude steel structure hoisting operations to determine an appropriate monitoring sliding window; configuring a sequence of construction monitoring indicators for high-altitude steel structure hoisting operations; performing anomaly detection on the monitoring data sequence set using the isolated forest algorithm to determine a first set of abnormal trajectory nodes and a first set of abnormal indices; constructing a trajectory prediction plugin based on the construction monitoring indicator sequence to perform anomaly detection on the monitoring data sequence set to determine a second set of abnormal trajectory nodes and a second set of abnormal indices; and fusing the first set of abnormal trajectory nodes, the first set of abnormal indices, the second set of abnormal trajectory nodes, and the second set of abnormal indices to determine the final set of abnormal trajectory nodes, which serves as the result of construction trajectory anomaly monitoring. This invention effectively improves the adaptive capability and detection accuracy of construction trajectory anomaly monitoring.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a GIS-based method and system for monitoring construction site trajectories. Background Technology

[0002] Currently, GIS-based construction site trajectory monitoring technology has been widely used in construction project management. It primarily involves deploying positioning devices to collect spatial location data of construction machinery, hoisting equipment, and workers, and then combining this data with a geographic information system for trajectory visualization and playback analysis. For high-altitude steel structure hoisting operations, existing technologies typically set a fixed monitoring sliding window to periodically collect and store trajectory data, and use threshold judgments or simple statistical methods to identify abnormal behavior.

[0003] However, existing technologies using fixed monitoring sliding windows cannot adaptively adjust to real-time dynamic risks at the construction site, resulting in redundant windows during low-risk periods and insufficient information during high-risk periods, affecting the timeliness and accuracy of anomaly detection. Furthermore, the use of a single anomaly detection strategy lacks comprehensive consideration and dynamic integration of multiple monitoring indicators, which can easily lead to missed detections or false alarms. Summary of the Invention

[0004] This invention provides a GIS-based method and system for monitoring construction site trajectories, aiming to solve the technical problems of insufficient timeliness and accuracy in the monitoring of abnormal construction trajectories in the prior art.

[0005] In view of the above problems, the present invention provides a GIS-based method and system for monitoring construction site trajectories.

[0006] In a first aspect, the present invention provides a GIS-based method for monitoring construction site trajectories, including: The preset monitoring sliding window is dynamically adjusted based on the construction risk index of high-altitude steel structure hoisting operations within a preset time period, and the appropriate monitoring sliding window is determined. Configure a sequence of construction monitoring indicators for high-altitude steel structure hoisting operations; Based on the construction risk index and construction monitoring index sequence, and combined with the isolated forest algorithm, anomaly detection is performed on the monitoring data sequence set within the adaptive monitoring sliding window to determine the first abnormal trajectory node set and the first abnormal index set. Based on the construction monitoring index sequence, a trajectory prediction plugin is constructed by combining the time series prediction model to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window, and to determine the second abnormal trajectory node set and the second abnormal index set. The final abnormal trajectory node set is determined by fusing the first abnormal trajectory node set, the first abnormal index set, the second abnormal trajectory node set, and the second abnormal index set, and is used as the construction trajectory abnormality monitoring result within the adaptive monitoring sliding window.

[0007] Secondly, the present invention provides a GIS-based construction site trajectory monitoring system, comprising: The dynamic window adjustment module is used to dynamically adjust the preset monitoring sliding window according to the construction risk index of the high-altitude steel structure hoisting operation within a preset time period, and to determine the appropriate monitoring sliding window. The indicator sequence configuration module is used to configure the construction monitoring indicator sequence for high-altitude steel structure hoisting operations; The isolated forest detection module is used to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window based on the construction risk index and construction monitoring index sequence, combined with the isolated forest algorithm, to determine the first abnormal trajectory node set and the first abnormal index set. The time-series prediction and detection module is used to construct a trajectory prediction plugin based on the construction monitoring index sequence and the time-series prediction model, and to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window to determine the second abnormal trajectory node set and the second abnormal index set. The anomaly fusion determination module is used to determine the final anomaly trajectory node set by fusing the first anomaly trajectory node set, the first anomaly index set, the second anomaly trajectory node set, and the second anomaly index set, as the construction trajectory anomaly monitoring result within the adaptive monitoring sliding window.

[0008] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a GIS-based method and system for monitoring construction site trajectories. Focusing on the specific operation scenario of high-altitude steel structure hoisting, it first completes the screening and prioritization of multi-dimensional monitoring indicators to construct a standardized monitoring indicator sequence; simultaneously, it introduces multi-factor evaluation to obtain a construction risk index, realizing dynamic adaptive adjustment of the monitoring sliding window. Two anomaly identification paths are designed in parallel: isolated forest layered detection and time-series prediction deviation comparison. The scale of detection indicators is adapted to the risk level, and two types of anomaly indices are calculated separately, with fusion weights dynamically allocated based on the reliability of the indicators. The weighted fusion of the dual detection results outputs the final anomaly trajectory node, effectively compensating for the shortcomings of fixed monitoring modes (weak adaptability), single-algorithm detection accuracy (low accuracy), and one-sided anomaly evaluation, comprehensively improving the real-time performance, comprehensiveness, and accuracy of anomaly identification in high-altitude hoisting construction trajectories. Attached Figure Description

[0009] Figure 1 A flowchart illustrating the GIS-based construction site trajectory monitoring method provided in this embodiment of the invention; Figure 2 A schematic diagram of the structure of a GIS-based construction site trajectory monitoring system provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: The module includes a dynamic window adjustment module 11, an indicator sequence configuration module 12, an isolated forest detection module 13, a time series prediction detection module 14, and an anomaly fusion judgment module 15. Detailed Implementation

[0010] This invention provides a GIS-based method and system for monitoring construction site trajectories, which addresses the technical problems of insufficient timeliness and accuracy in monitoring abnormal construction trajectories in existing technologies.

[0011] Example 1, as Figure 1 As shown, this invention provides a GIS-based method for monitoring construction site trajectories, the method comprising: S100: Dynamically adjust the preset monitoring sliding window based on the construction risk index of high-altitude steel structure hoisting operations within a preset time period, and determine the appropriate monitoring sliding window.

[0012] In this embodiment of the invention, a preset monitoring sliding window is dynamically adjusted based on the construction risk index of high-altitude steel structure hoisting operations within a preset time period to determine an appropriate monitoring sliding window. During the monitoring of the construction trajectory of high-altitude steel structure hoisting operations, the duration of the monitoring sliding window directly determines the amount of historical data covered by each anomaly detection. Existing technologies typically use sliding windows of fixed duration; however, the actual risk level at the construction site changes in real time: if an excessively long window is used during low-risk periods, a large amount of redundant data will be repeatedly processed, increasing the computational burden; if the window is too short during high-risk periods, it will be difficult to capture the process of risk accumulation and evolution, easily leading to missed anomalies. Therefore, it is necessary to dynamically adjust the duration of the sliding window based on the real-time construction risk index to match the current risk level, thereby optimizing the utilization of computational resources while ensuring detection accuracy.

[0013] Step S100 in the method provided in this embodiment of the invention includes: Multiple risk factors for high-altitude steel structure hoisting operations within a preset time period are obtained. After normalization of each risk factor, a weighted evaluation is performed to determine the construction risk index. The multiple risk factors include at least the real-time wind speed level, the ratio of the weight of the hoisted object to the rated load, the continuous working time of the operators, and the number of personnel in the warning zone below the hoisted object. The ratio of the preset benchmark construction risk index to the construction risk index is used as the window duration compensation coefficient, and the product of the window duration compensation coefficient and the preset duration of the monitoring sliding window is used as the adaptive monitoring sliding window.

[0014] First, multiple risk factors for high-altitude steel structure hoisting operations within a preset time period are obtained. After normalizing each risk factor, a weighted evaluation is performed to determine the construction risk index. The multiple risk factors include at least the real-time wind speed level, the ratio of the weight of the hoisted object to the rated load, the continuous working time of the operators, and the number of personnel in the warning zone below the hoisted object.

[0015] The preset time period refers to the time unit for conducting construction risk assessment, such as every 10 minutes, every 5 minutes, or every hoisting cycle. The preset time period can be set according to management needs and is usually consistent with the adjustment frequency of the monitoring sliding window. Risk factors are real-time quantifiable variables that can affect the safety level of high-altitude steel structure hoisting operations. In this embodiment, at least the following are included: real-time wind speed level, the ratio of the hoisted load weight to the rated load, the continuous working time of operators, and the number of personnel in the warning zone below the hoisted load. These risk factors reflect the current hoisting risks from four dimensions: environment, equipment, personnel, and site.

[0016] Normalization refers to mapping risk factors with different dimensions to the same numerical range, such as [0,1], for subsequent weighted summation. For example, a piecewise normalization function can be set according to industry standards. Weighted assessment involves assigning a weight coefficient to each risk factor, reflecting the degree of influence of each risk factor on the overall risk. The normalized factor values ​​are multiplied by their corresponding weights, and then summed to obtain the construction risk index. Weights can be determined through historical accident data analysis or expert scoring.

[0017] Specifically, at the end of each preset time period, real-time values ​​of four risk factors are collected from on-site sensors, video analysis systems, crane control systems, and personnel positioning systems, and normalized according to the following rules: Real-time wind speed level: Level 0 (0-0.2 m / s) is 0, Level 1 (0.3-1.5 m / s) is 0.2, Level 2 (1.6-3.3 m / s) is 0.4, Level 3 (3.4-5.4 m / s) is 0.6, Level 4 ( The weight ratio of the hoisted load to the rated load is 0.8 for speeds between 5.5 and 7.9 m / s, and 1.0 for level 5 and above. The continuous working time of operators is 0 for 0-2 hours, 0.3 for 2-4 hours, 0.6 for 4-6 hours, 0.9 for 6-8 hours, and 1.0 for more than 8 hours. The number of personnel in the warning zone below the hoisted load is 0 for 0 people, 0.4 for 1 person, 0.7 for 2 people, and 1.0 for 3 or more people. Weights are set for each risk factor. The construction risk index is the sum of the products of each normalized value and its corresponding weight, with the result in the range of [0,1]. A larger value indicates a higher overall risk.

[0018] For example, at a steel structure high-altitude hoisting construction site, the preset time period is 10 minutes. During the period from 10:00 AM to 10:10 AM, the following data was collected: real-time wind speed level 4, corresponding to a normalized value of 0.8; load weight 18 tons, rated load 20 tons, ratio 0.9, corresponding to a normalized value of 0.9; operators have been working continuously for 4.5 hours, corresponding to a normalized value of 0.6; there are 2 people in the warning zone below the load, corresponding to a normalized value of 0.7. The weights are: real-time wind speed level 0.3, load weight to rated load ratio 0.3, operator continuous working time 0.2, and number of people in the warning zone below the load 0.2. The construction risk index is calculated as follows: 0.8 × 0.3 + 0.9 × 0.3 + 0.6 × 0.2 + 0.7 × 0.2 = 0.77.

[0019] Secondly, the ratio of the preset benchmark construction risk index to the construction risk index is used as the window duration compensation coefficient, and the product of the window duration compensation coefficient and the preset monitoring sliding window duration is used as the adapted monitoring sliding window. The preset benchmark construction risk index is a pre-set reference risk level, usually taken as a typical risk index value under normal operating conditions, for example, 0.5. The window duration compensation coefficient is used to amplify or reduce the original window duration by a factor. The calculation formula is: Window duration compensation coefficient = Preset benchmark construction risk index / Construction risk index.

[0020] The preset monitoring sliding window is the default, unadjusted standard sliding window duration, such as 5 or 10 minutes. This value can be set based on experience under normal operating conditions. The adapted monitoring sliding window refers to the actual sliding window duration used for anomaly detection in the current cycle after dynamic adjustment by the risk index. The calculation formula is: Adapted monitoring sliding window = Window duration compensation coefficient × Preset monitoring sliding window. The higher the construction risk index, the smaller the window duration compensation coefficient, and the shorter the adapted monitoring sliding window; the lower the construction risk index, the larger the window duration compensation coefficient, and the longer the adapted monitoring sliding window.

[0021] Specifically, a preset baseline construction risk index is obtained. This value can be calculated by multiplying the average normalized value of risk factors during historical normal operation phases by a weight, or it can be set empirically. In this embodiment, the preset baseline construction risk index is 0.5. The actual construction risk index within the current time period is read, and the window duration compensation coefficient is calculated. The preset duration of the monitoring sliding window is obtained, for example, set to 600 seconds, and the adaptive monitoring sliding window duration is calculated. To prevent the loss of key historical information due to an excessively short window, a lower limit of 300 seconds and an upper limit of 1200 seconds can be set. The calculated adaptive monitoring sliding window duration is applied to subsequent anomaly detection steps within the current time period. At the end of each new preset time period, S100 is re-executed to dynamically update the adaptive window for the next period.

[0022] For example, the preset baseline construction risk index is 0.5, and the preset monitoring sliding window duration is 600 seconds. The actual construction risk index is 0.77. Therefore, the compensation coefficient is approximately 0.5 / 0.77, and the adapted monitoring sliding window duration is approximately 600 × 0.649, which is 389 seconds. This means that after the current 10-minute cycle, a sliding window of approximately 389 seconds will be used to detect anomalies in subsequent monitoring data sequences. Compared to a fixed 600-second window, the window is shortened by about 35%, allowing for a faster response to current trajectory changes during high-risk periods and preventing outdated data from diluting abnormal signals. If the construction risk index drops to 0.3 in another cycle, the compensation coefficient becomes approximately 0.5 / 0.3, which is 1.667, and the adapted window becomes approximately 600 × 1.667, which is 1000 seconds. The extended window allows for the accumulation of more normal trajectory data for model training or baseline comparison, while reducing the computation frequency.

[0023] In this embodiment of the invention, a dynamic correlation between the sliding window duration and the real-time construction risk index is achieved. The window is automatically shortened to improve sensitivity during high-risk periods and automatically extended to reduce computational redundancy during low-risk periods, overcoming the shortcomings of fixed windows. Simultaneously, a weighted evaluation is performed on four typical risk factors: wind speed, load, personnel fatigue, and personnel in the warning zone. This makes the window adjustment more closely reflect the actual safety situation and avoids the bias of a single indicator. While ensuring anomaly detection performance, it effectively reduces the repetitive processing of irrelevant historical data and lowers the load on the backend server in long-term continuous monitoring scenarios. The adaptation window is recalculated every preset time period, with a moderate adjustment frequency that avoids frequent window jitter while tracking short- to medium-term risk trends.

[0024] S200: A sequence of construction monitoring indicators for high-altitude steel structure hoisting operations.

[0025] In this embodiment of the invention, a construction monitoring index sequence for high-altitude steel structure hoisting operations is configured. High-altitude steel structure hoisting operations involve multiple high-risk aspects, including machinery operation, high-altitude edge handling, and heavy object lifting, resulting in complex and diverse types of operational risks. In conventional trajectory monitoring, monitoring parameters are randomly selected without a fixed screening range, and there is no risk level classification or sequential order for various monitoring parameters. The degree of accident hazard and trigger probability corresponding to different monitoring parameters vary significantly. Randomly mixing monitoring parameters will cause confusion in the subsequent abnormal indicator screening logic and unreasonable weight allocation, reducing the targeted nature of trajectory anomaly identification. Therefore, it is necessary to define a specific monitoring index range adapted to the hoisting scenario, quantify the risk level of the indicators based on safety accident characteristics, and arrange them in an orderly manner to form a standardized index sequence, providing a standardized data foundation for subsequent two-layer anomaly detection and weighted index screening.

[0026] Step S200 in the method provided in this embodiment of the invention includes: A set of construction monitoring indicators for high-altitude steel structure hoisting operations is configured, wherein the set of construction monitoring indicators includes the clear distance between the hoisted object and the obstacle, the horizontal movement speed of the high-altitude workers, the number of times the hoisting path intersects with the trajectory of the personnel below, the abnormality of the boom rotation-amplitude coordination, the alignment deviation of the hoisted object placement point, and the deviation of the total hoisting cycle time. The monitoring indicators in the construction monitoring indicator set are sorted from largest to smallest according to their impact to generate a construction monitoring indicator sequence. The impact of each monitoring indicator is determined based on the severity of the safety accident corresponding to each monitoring indicator and the probability of the safety accident occurring.

[0027] First, a set of construction monitoring indicators is configured for high-altitude steel structure hoisting operations. This set includes the clear distance between the hoisted object and obstacles, the horizontal movement speed of personnel working at height, the number of intersections between the hoisting path and the trajectory of personnel below, the anomaly degree of boom rotation-amplitude coordination, the alignment deviation of the hoisted object's placement point, and the deviation of the total hoisting cycle time. The construction monitoring indicator set refers to a quantifiable and targeted set of fixed monitoring parameters selected for high-altitude steel structure hoisting operations, combined with the work process and typical safety hazards. It covers control dimensions such as spatial safety, personnel behavior, mechanical coordination, operational accuracy, and operational timeliness, and is used to comprehensively capture safety risks related to the hoisting trajectory.

[0028] Among them, the clear distance between the hoisted object and the obstacle refers to the straight-line distance between the center of gravity of the hoisted object and the nearest point of the surrounding obstacle, such as scaffolding, building exterior walls, other hoisting equipment, and temporary components, during the hoisting process. It is used to determine whether there is a risk of collision with the hoisted object. The horizontal movement speed of the workers at height refers to the real-time movement rate of the workers on the aerial work platform or component in the horizontal direction. It is used to determine whether the workers' work behavior is stable and whether there is a risk of instability and fall from height. The number of times the hoisting path intersects with the trajectory of the personnel below refers to the number of times the trajectory of the hoisted object from the lifting point to the landing point overlaps with the trajectory of the workers on the ground within a single preset monitoring period. It is used to determine whether there is a risk of personnel collision with the hoisting path. The abnormality degree of the boom slewing-luffing coordination refers to the degree of deviation between the actual action coordination and the standard coordination logic during the horizontal slewing and luffing of the crane boom. The value ranges from [0,1]. The larger the value, the greater the coordination deviation and the higher the risk of mechanical overturning. The alignment deviation of the hoisted object's placement point refers to the offset distance between the coordinates of the center of gravity of the hoisted object when it is actually placed and the coordinates of the preset target placement point. It is used to judge the hoisting alignment accuracy and avoid component installation errors or falls due to alignment deviations. The total hoisting cycle time deviation refers to the difference between the actual time consumed in a single complete hoisting cycle (lifting, transferring, placing, and resetting) and the preset standard hoisting cycle time. A positive number indicates that the time is too long, which can easily lead to personnel fatigue and equipment overload. A negative number indicates that the time is too short, which can easily lead to hasty operation and increased risk of error.

[0029] All indicators are collected and calculated in real time using existing monitoring equipment at the construction site. The specific collection and calculation methods are as follows: The clear distance between the hoisted object and obstacles is directly detected by a laser rangefinder sensor mounted on the hoisted object, outputting the straight-line distance value in real time; the horizontal movement speed of high-altitude workers is calculated by continuously collecting position coordinates through personnel positioning terminals and calculating the movement speed based on the horizontal coordinate displacement per unit time; the number of intersections between the hoisting path and the trajectory of personnel below is calculated by defining the personnel activity range through on-site video monitoring and comparing the overlap with the equipment positioning trajectory, accumulating the number of trajectory intersections within the monitoring period; the anomaly degree of boom rotation-amplitude coordination is calculated by collecting real-time operating parameters through angle sensors and telescopic length sensors mounted on the crane, comparing and calculating with standard coordination matching rules to obtain a dimensionless anomaly degree value; the alignment deviation of the hoisted object placement point is calculated by collecting the actual placement coordinates through a GPS positioning module, retrieving the coordinates of a preset standard point, and calculating the actual offset distance through the coordinate difference; the total time deviation of the hoisting cycle is calculated by recording the actual duration of a single hoisting process through a timing module, calculating the difference with the preset standard operation time, and generating time deviation data in real time.

[0030] Specifically, when configuring the construction monitoring indicator set, the entire process of high-altitude steel structure hoisting operations is comprehensively reviewed. This process covers all stages, including hoisting preparation, lifting of the load, aerial transfer, load placement, and equipment reset. Typical safety hazards such as collisions, falls, overturning, installation errors, and personnel fatigue are accurately identified at each stage. Based on the mechanisms of various safety hazards, monitoring parameters with quantifiable characteristics and real-time data acquisition are selected one by one, ensuring that each type of safety hazard has a corresponding monitoring indicator for targeted control. Invalid parameters that are unrelated to the hoisting trajectory, cannot be quantified, or have excessively high collection costs are eliminated, leaving six core monitoring parameters to form a dedicated construction monitoring indicator set. The collection frequency of all indicators is uniformly defined, and the collection rhythm is consistent with the preset time cycle, for example, using 10 minutes as a complete collection unit. The quantitative judgment standards and reasonable value ranges for each indicator are simultaneously clarified, ensuring a unified format for multi-dimensional monitoring data and effective horizontal comparison.

[0031] For example, at a steel structure factory hoisting site, the main on-site operations involve routine hoisting of steel columns and beams. Six monitoring indicators were determined based on the on-site working environment and risk types, and real-time data was acquired strictly according to the aforementioned collection and calculation methods. The laser rangefinder measured a net distance of 2.8m between the hoisted object and surrounding obstacles; the personnel positioning terminal calculated the horizontal movement speed of the personnel working at height to be 0.6m / s; trajectory comparison statistics showed that the hoisting path intersected with the personnel trajectory below once within a 10-minute cycle; the boom operation parameters, converted using collaborative rules, yielded a collaborative anomaly degree of 0.2; GPS coordinate difference calculations showed an alignment deviation of 35mm at the hoisted object's placement point; and a comparison of the actual working time with the standard time yielded a total hoisting cycle time deviation of 20s. These six standardized collected and calculated parameters together constitute the construction monitoring indicator set for the current working period at this site, providing complete basic data for subsequent indicator influence ranking and trajectory anomaly detection.

[0032] Secondly, the monitoring indicators in the construction monitoring indicator set are sorted from largest to smallest according to their impact, generating a construction monitoring indicator sequence. The impact of each monitoring indicator is determined based on the severity of the safety accident corresponding to that indicator and the probability of such an accident. The impact of a monitoring indicator is a numerical value used to quantify the overall impact of a single monitoring indicator on the safety of hoisting operations. The value ranges from [0,1]. A higher value indicates a higher overall risk of a safety accident caused by that indicator, and therefore requires priority attention. The severity of a safety accident refers to the level of harm that a single indicator might cause, such as the number of casualties, the extent of equipment damage, or the duration of construction stoppage. The probability of a safety accident refers to the likelihood of a safety accident occurring when a single indicator is abnormal. Both factors jointly determine the impact of the indicator.

[0033] Based on the enterprise employee injury and death accident classification standards, the severity of safety accidents is divided into four levels and assigned values: Extremely High (death or multiple serious injuries), High (single serious injury), Medium (minor injuries or property damage of 10,000-100,000 yuan), and Low (no injuries, only work stoppage), with values ​​of 4, 3, 2, and 1 respectively. The probability of safety accidents is scored by safety experts based on historical construction data, with a value range of 0-1. The formula for calculating the impact of monitoring indicators is: Impact of monitoring indicators = Severity score of safety accident × Probability of safety accident occurrence.

[0034] For example, firstly, the severity and probability of occurrence of each of the six indicators in the construction monitoring index are assessed, and the original impact is calculated using the product method: Horizontal movement speed of high-altitude workers: extremely high severity, probability of occurrence 0.3, 4×0.3=1.2; Clear distance between the hoisted object and the obstacle: extremely high severity, probability of occurrence 0.25, 4×0.25=1.0; Number of intersections between the hoisting path and the trajectory of the personnel below: high severity, probability of occurrence 0.3, 3×0.3=0.9; Abnormality of boom rotation-amplitude coordination: high severity, probability of occurrence 0.2, 3×0.2=0.6; Alignment deviation of the hoisted object's placement point: medium severity, probability of occurrence 0.1, 2×0.1=0.2; Deviation of total hoisting cycle time: low severity, probability of occurrence 0.05, 1×0.05=0.05. The impact is normalized as follows: Total impact = 1.2 + 1.0 + 0.9 + 0.6 + 0.2 + 0.05 = 3.95. The normalized impact weights are: horizontal movement speed of high-altitude workers = 1.2 / 3.95 ≈ 0.304, net distance between the hoisted object and the obstacle = 1.0 / 3.95 ≈ 0.253, number of intersections between the hoisting path and the trajectory of the personnel below = 0.9 / 3.95 ≈ 0.228, abnormality of boom rotation-amplitude coordination = 0.6 / 3.95 ≈ 0.152, alignment deviation of the hoisted object placement point = 0.2 / 3.95 ≈ 0.050, and deviation of total hoisting cycle time = 0.05 / 3.95 ≈ 0.013.

[0035] Secondly, based on the product of the original impact degrees from largest to smallest, a fixed sequence of construction monitoring indicators is generated: horizontal movement speed of high-altitude workers, clear distance between the suspended object and obstacles, number of intersections between the hoisting path and the trajectory of personnel below, anomaly degree of boom rotation-amplitude coordination, alignment deviation of the suspended object's placement point, and deviation of the total hoisting cycle time. If the impact degrees are the same, they are sorted from largest to smallest according to the severity score of the safety accident. This sequence will be directly used for indicator screening and weighted calculation in subsequent anomaly detection, prioritizing the detection of indicators with high impact degrees.

[0036] In this embodiment of the invention, six specific monitoring indicators for high-altitude steel structure hoisting are selected to form a closed set of construction monitoring indicators. This avoids interference from irrelevant indicators, ensuring the relevance and practicality of the monitoring data. Furthermore, the data collection sources and quantification standards for each indicator are clearly defined, guaranteeing data collectability and comparability. Weights are determined through historical accident data analysis and expert scoring, combining the severity and probability of safety accidents to quantify the influence of indicators. This makes the indicator ranking more closely reflect the actual safety situation, avoiding the one-sidedness of single-dimensional evaluation. The generated construction monitoring indicator sequence provides a clear standardized basis for subsequent risk index-based indicator selection, dual-algorithm anomaly detection, and weighted fusion of anomaly indices. This enables priority detection of high-risk indicators and reasonable control of low-risk indicators, effectively improving the targeting and accuracy of subsequent construction trajectory anomaly detection while reducing the computational burden caused by invalid detections.

[0037] S300: Based on the construction risk index and construction monitoring index sequence, and combined with the isolated forest algorithm, perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window to determine the first abnormal trajectory node set and the first abnormal index set.

[0038] In this embodiment of the invention, based on the construction risk index and the construction monitoring indicator sequence, an anomaly detection is performed on the monitoring data sequence set within the adapted monitoring sliding window using the isolated forest algorithm to determine the first abnormal trajectory node set and the first abnormal index set. The isolated forest algorithm has advantages such as fast detection speed and sensitivity to abnormal data. However, when directly applied to the trajectory monitoring of high-altitude steel structure hoisting, it fails to optimize detection parameters based on the on-site risk level, easily leading to problems such as delayed detection response during high-risk periods and wasted computing power during low-risk periods. Furthermore, it does not optimize the detection dimension based on the influence of monitoring indicators, resulting in insufficient targeting of anomaly identification. Therefore, it is necessary to dynamically select the number of monitoring indicators based on the construction risk index, build an adapted isolated forest model based on the selected indicators, calculate the anomaly score through path length, quantify the degree of anomaly, and finally determine the first abnormal trajectory node set and the first abnormal index set, providing accurate preliminary detection data for subsequent anomaly result fusion.

[0039] Step S300 in the method provided in this embodiment of the invention includes: The ratio of the construction risk index to the preset benchmark construction risk index is multiplied by the number of benchmark monitoring indicators selected and rounded down to obtain the current number of monitoring indicators selected. The number of benchmark monitoring indicators selected is less than or equal to half of the total number of construction monitoring indicators, and the current number of monitoring indicators selected is not less than 2. Based on the current number of monitoring indicators selected, the construction monitoring indicator sequence is filtered to obtain the first monitoring indicator set; The monitoring data sequence set within the adaptive monitoring sliding window is filtered based on the first monitoring index set to obtain the first monitoring data sequence set, wherein the monitoring data sequence set includes monitoring data of several construction trajectory nodes within the adaptive monitoring sliding window; Based on the first monitoring indicator set, K monitoring data are randomly selected from each monitoring indicator data in the first monitoring data sequence set to build an isolated tree, generating multiple first isolated trees; Based on the multiple first isolated trees, anomaly judgment is performed on the monitoring data in the first monitoring data sequence set to determine the first abnormal trajectory node set and the first abnormal index set.

[0040] First, the ratio of the construction risk index to the preset benchmark construction risk index is multiplied by the number of benchmark monitoring indicators selected and rounded down to obtain the current number of monitoring indicators selected. The number of benchmark monitoring indicators selected is less than or equal to half of the total number of construction monitoring indicators, and the current number of monitoring indicators selected is not less than 2. The number of benchmark monitoring indicators selected is a pre-set fixed number of indicators used for reference calculations, set in conjunction with the total number of construction monitoring indicators, and must meet the requirement of not exceeding half of the total number of construction monitoring indicators. The current number of monitoring indicators selected is the actual number of indicators used for isolated forest detection after dynamic adjustment based on the real-time construction risk index. The value changes with the construction risk index; the higher the risk, the fewer indicators are selected to improve the detection response speed.

[0041] The formula for calculating the current number of monitoring indicators is: Current number of monitoring indicators = Construction risk index / Preset benchmark construction risk index × Number of benchmark monitoring indicators selected, rounded down. A higher construction risk index results in a larger ratio of the construction risk index to the preset benchmark construction risk index, leading to a larger number of current monitoring indicators. This allows for increased monitoring indicators and more detailed detection during high-risk periods, while simplifying indicators and reducing computational costs during low-risk periods. If the calculated result is less than 2, it is forced to be 2 to ensure at least two detection dimensions, avoiding the bias of single-indicator detection.

[0042] The preset benchmark construction risk index is a fixed reference value for steel structure high-altitude hoisting operations under standard, compliant, and stable operating conditions without high-risk hazards. This value is determined based on historical risk factor sample data from long-term normal hoisting operations on site. Various risk factors are collected from multiple stable operating cycles, and after uniformly applying the same normalization rules and weighted calculation logic, the average value of the steady-state interval is taken as a fixed value. At the same time, it is calibrated with the assistance of safety management standards in the construction hoisting industry, and the value is fixed in the range of 0-1, serving as a unified reference standard for risk multiplier conversion throughout the entire process.

[0043] For example, the total number of construction monitoring indicators is 6, the number of baseline monitoring indicators is ≤3, the number of selected baseline monitoring indicators is 3, and the preset baseline construction risk index is 0.5. S100 has a known construction risk index of 0.77. According to the calculation formula: 0.77 / 0.5×3=4.62, rounded down, we get the current number of selected monitoring indicators as 4.

[0044] Secondly, based on the current number of selected monitoring indicators, the construction monitoring indicator sequence is filtered to obtain the first monitoring indicator set. The first monitoring indicator set is a set of specialized indicators extracted from the construction monitoring indicator sequence for anomaly detection using the Isolation Forest algorithm. The selection logic follows the inherent sorting rules of the indicator sequence, selecting indicators in descending order of influence to ensure that high-risk indicators are included in the detection scope first.

[0045] For example, the current number of monitoring indicators is calculated to be 4. The first four indicators are selected in order to obtain the first monitoring indicator set: horizontal movement speed of high-altitude workers, clear distance between the hoisted object and the obstacle, number of times the hoisting path intersects with the trajectory of the personnel below, and the degree of abnormality in the coordination of boom rotation and amplitude change.

[0046] Next, the monitoring data sequence set within the adaptive monitoring sliding window is filtered based on the first monitoring indicator set to obtain the first monitoring data sequence set. This first monitoring data sequence set includes monitoring data from several construction trajectory nodes within the adaptive monitoring sliding window. The monitoring data sequence set is a comprehensive collection of all original monitoring data from all construction trajectory nodes within the time frame of the adaptive monitoring sliding window. Each construction trajectory node corresponds to the collection parameters of all six indicators at the same time. The first monitoring data sequence set is a simplified subset of data, retaining only the monitoring data corresponding to the indicators in the first monitoring indicator set.

[0047] Specifically, all construction trajectory nodes within the adaptive monitoring sliding window are traversed, and the monitoring categories contained in the first monitoring indicator set are matched one by one. The corresponding real-time monitoring data are extracted, other non-selected indicator data are removed, and the data are arranged in chronological order to form a first monitoring data sequence set that is time-continuous and dimension-matched.

[0048] For example, the duration of the adaptive monitoring sliding window is 389 seconds, and 10 continuous construction trajectory nodes are collected evenly within the window. The original monitoring data sequence set contains six complete indicator data for each node. Based on four selected indicators, only the measured data of the horizontal movement speed of the high-altitude workers, the clear distance between the hoisted object and the obstacle, the number of intersections between the hoisting path and the trajectory of the personnel below, and the anomaly degree of the boom rotation-amplitude coordination are retained, and the first monitoring data sequence set is obtained. For example, the data for node 1 is as follows: 0.6 m / s, 2.8 m, 0 times, 0.2; nodes 2 to 10 similarly contain the above four data.

[0049] Subsequently, based on the first set of monitoring indicators, K monitoring data points are randomly selected from each monitoring indicator data point in the first set of monitoring data sequences to construct an isolated tree, generating multiple first isolated trees. K is a pre-set sample data volume for a single isolated tree, and the sample data volume needs to stably adapt to the total number of trajectory nodes within the window.

[0050] The isolated tree is a binary tree structure, using K randomly selected sets of monitoring data as tree nodes, and is constructed hierarchically through random threshold partitioning. The first isolated tree is a dedicated tree structure adapted to the detection indicators. Multiple first isolated trees together form an isolated forest, reducing the random error of a single model. When building the isolated tree, monitoring data randomly selected from each monitoring indicator is strictly used as the basic nodes of the isolated tree, and the tree is split level by level until the termination condition is reached.

[0051] Specifically, the initial number of isolated trees is set to 10. For each indicator included in the first monitoring indicator set, K data points are randomly selected without repetition from the corresponding dimension data of the first monitoring data sequence set. Based on the selected data nodes, a random partitioning threshold is set, and binary splitting is completed layer by layer. After the single tree structure is constructed, the sampling and building process is repeated cyclically to generate multiple first isolated trees in batches, which are then combined to form a complete isolated forest model.

[0052] For example, K=5 is set, and time-series data of the four monitoring indicators are randomly sampled sequentially. Taking the boom slewing-amplitude coordination anomaly degree as an example, 5 sets of values ​​are randomly selected from 10 sets of data as nodes to build isolated trees; simultaneously, the same sampling and tree construction operations are performed on three types of data: spatial clearance, alignment deviation, and trajectory intersection count. The four indicators are cross-fused to complete the construction of a single isolated tree. The construction process is executed ten times in a loop to generate 10 structurally independent first isolated trees, which are used for subsequent anomaly determination of the entire trajectory nodes.

[0053] Furthermore, based on the plurality of first isolated trees, anomaly judgment is performed on the monitoring data in the first monitoring data sequence set to determine the first abnormal trajectory node set and the first abnormal index set.

[0054] Specifically, based on the plurality of first isolated trees, anomaly detection is performed on the monitoring data in the first monitoring data sequence set to determine the first abnormal trajectory node set and the first abnormal index set, including: For each construction trajectory node in the first monitoring data sequence set, the average path length of the current construction trajectory node in the multiple first isolated trees is calculated, and the anomaly score of the current construction trajectory node is calculated based on the average path length. When the abnormal score is greater than the preset first abnormal score threshold, the current construction trajectory node is determined as the first abnormal trajectory node and is included in the first abnormal trajectory node set. For each first abnormal trajectory node in the first abnormal trajectory node set, the abnormality coefficient of each monitoring indicator in the first monitoring indicator set corresponding to the current first abnormal trajectory node is obtained, wherein the abnormality coefficient is determined based on the degree to which the monitoring data of the monitoring indicator at the construction trajectory node deviates from the normal range. For each first abnormal trajectory node in the first abnormal trajectory node set, the abnormality coefficients of each monitoring indicator corresponding to the current first abnormal trajectory node are weighted and summed according to the influence weight of the monitoring indicator in the construction monitoring indicator sequence to calculate the first abnormal index of the current first abnormal trajectory node. The first abnormal indices of all first abnormal trajectory nodes are then aggregated to obtain the first abnormal index set.

[0055] First, for each construction trajectory node in the first monitoring data sequence set, the average path length of the current construction trajectory node in the multiple first isolated trees is calculated, and the anomaly score of the current construction trajectory node is calculated based on the average path length. Path length refers to the number of traversal levels of the trajectory node data from the root node to the leaf node of the isolated tree. Normal working condition data is concentrated and has a longer path length, while abnormal data has a higher dispersion and a shorter path length. The average path length is the arithmetic mean of the path lengths of a single trajectory node in all first isolated trees. The anomaly score is obtained by converting the average path length, and the value converges to the interval between 0 and 1. The higher the score, the stronger the trajectory anomaly risk.

[0056] Specifically, the process iterates through all construction trajectory nodes in the first monitoring data sequence set, queries the path length of each individual node in each of the first isolated trees, sums these values, and divides the sum by the total number of isolated trees to calculate the average path length. This average path length is then substituted into the isolated forest standard anomaly score calculation formula, combined with the expected path length parameter of the total number of trajectory nodes within the window, to quantitatively calculate the independent anomaly score for each node.

[0057] For example, there are 10 isolated trees and 10 trajectory nodes. Taking node 8 as an example, traversing all isolated trees yields path lengths of 1, 2, 1, 2, 1, 1, 2, 1, 2, 1, resulting in an average path length of 1.5. Substituting these into the standard anomaly score calculation formula for isolated forests, the anomaly score for this node is 0.73. The average path lengths of the remaining regular trajectory nodes remain between 2.3 and 3.0, with anomaly scores generally below 0.65.

[0058] Secondly, when the anomaly score exceeds a preset first anomaly score threshold, the current construction trajectory node is identified as a first anomaly trajectory node and added to the first anomaly trajectory node set. The first anomaly score threshold is a pre-defined anomaly determination threshold, calibrated in conjunction with historical monitoring data of steel structure hoisting. The first anomaly trajectory node set is used to uniformly collect all initially screened anomaly points, achieving centralized marking and independent management of anomaly trajectories.

[0059] Specifically, the abnormal scores of each construction trajectory node are compared one by one with a preset first abnormal score threshold. If the abnormal score exceeds the threshold, it is determined to be the first abnormal trajectory node and included in the corresponding set; if the abnormal score does not exceed the threshold, it is determined to be a normal node and is not marked or stored. For example, if the preset first abnormal score threshold is 0.7, after screening all nodes, only node 8 has an abnormal score of 0.73, which exceeds the threshold value, and is determined to be the first abnormal trajectory node. The first abnormal trajectory node set uniquely contains node 8.

[0060] Next, for each first abnormal trajectory node in the first abnormal trajectory node set, the abnormality coefficient of each monitoring indicator in the first monitoring indicator set corresponding to the current first abnormal trajectory node is obtained. The abnormality coefficient is determined based on the degree to which the monitoring data of the monitoring indicator deviates from the normal range at the construction trajectory node. The abnormality coefficient is a value between 0 and 1, quantifying the degree of deviation of the parameter; the greater the deviation of the data from the reasonable range, the higher the abnormality coefficient value. The normal value range and extreme abnormal threshold of each indicator are pre-set according to the hoisting construction specifications, serving as a unified calculation standard for the abnormality coefficient.

[0061] Specifically, the standard normal range and extreme anomaly boundary of all indicators in the first monitoring indicator set are retrieved, and the real-time monitoring parameters of the first abnormal trajectory node are extracted and compared with the parameter deviation status. The anomaly coefficient is calculated using a 0-1 interval linear conversion method. When the monitoring data is within the normal range, the anomaly coefficient is 0; when the measured value is greater than the upper limit of the normal range, the anomaly coefficient = (measured value - upper limit of normal range) / (extreme anomaly value - upper limit of normal range); when the measured value is less than the lower limit of the normal range, the anomaly coefficient = (lower limit of normal range - measured value) / (lower limit of normal range - extreme anomaly value). When the calculation result exceeds 1, it is uniformly set to 1.

[0062] For example, the control ranges for the four indicators in the first monitoring indicator set are as follows: the normal range for the horizontal movement speed of high-altitude workers is 0-0.8 m / s, with the extreme abnormal value set at 2.0 m / s; the normal lower limit for the clear distance between the suspended object and the obstacle is 0.5 m, with no upper limit constraint; the reasonable threshold for the number of times the hoisting path intersects with the trajectory of the personnel below is ≤1 time, with the extreme abnormal value set at 5 times; the normal range for the abnormality of the boom rotation-amplitude coordination is 0-0.4, with the extreme abnormal value set at 1.0. The measured data for node 8 are as follows: horizontal movement speed 1.5 m / s, clear distance 1.5 m, number of trajectory intersections 3 times, and coordination abnormality 0.8. The anomaly coefficients were calculated linearly for each item: the measured horizontal movement speed of the high-altitude worker exceeded the normal upper limit, with an anomaly coefficient of (1.5-0.8) / (2.0-0.8)≈0.58; the measured net distance between the suspended object and the obstacle met the specifications, with an anomaly coefficient of 0; the number of intersections between the hoisting path and the trajectory of the personnel below exceeded the reasonable threshold, with an anomaly coefficient of (3-1) / (5-1)=0.5; and the measured anomaly of the boom rotation-amplitude coordination exceeded the normal upper limit, with an anomaly coefficient of (0.8-0.4) / (1.0-0.4)≈0.67. The final summation yielded an anomaly coefficient combination of 0.58, 0, 0.5, and 0.67 for the four indicators corresponding to this first abnormal trajectory node.

[0063] Finally, for each first abnormal trajectory node in the first abnormal trajectory node set, the abnormality coefficients of each monitoring indicator corresponding to the current first abnormal trajectory node are weighted and summed according to the influence weight of the monitoring indicator in the construction monitoring indicator sequence to calculate the first abnormality index of the current first abnormal trajectory node. The first abnormality indices of all first abnormal trajectory nodes are then aggregated to obtain the first abnormality index set. The first abnormality index is a comprehensive abnormality quantification value that integrates the deviation of multiple indicators, and its value is in the range of 0-1. The weights of each indicator are directly taken from the predetermined influence values ​​within the sequence, with higher influence indicators having a higher weight. The weighted summation can objectively reflect the comprehensive risk level of the node. The first abnormality index set is bound one-to-one with the first abnormal trajectory node set, providing a quantitative basis for the subsequent fusion of multiple algorithm results.

[0064] Specifically, the impact parameters corresponding to each indicator in the first monitoring indicator set are matched, and the anomaly coefficient of each indicator is multiplied by its corresponding impact value in sequence. All weighted results are summed to obtain the first anomaly index for a single anomaly node. The calculation results of all anomaly nodes are summarized and integrated to form a structured first anomaly index set.

[0065] For example, the influence of the four selected indicators is as follows: horizontal movement speed of high-altitude workers (0.304), clear distance between the suspended object and obstacle (0.253), number of intersections between the hoisting path and the trajectory of personnel below (0.228), and anomaly degree of boom rotation-amplitude coordination (0.152). Combining the anomaly coefficient combination of node 8 (0.58, 0, 0.5, 0.67), the first anomaly index = 0.58×0.304 + 0×0.253 + 0.5×0.228 + 0.67×0.152 ≈ 0.392. Currently, only a single anomalous node exists, corresponding to a first anomaly index set of 0.392.

[0066] In this embodiment of the invention, the number of monitoring indicators is dynamically adjusted by combining the ratio of the construction risk index to the benchmark risk index. This strictly adheres to the constraints on the number of benchmark indicators while relaxing the current indicator quantity limits. Under high-risk conditions, the detection dimensions are automatically expanded to achieve multi-dimensional, refined screening. A first set of monitoring indicators and a first set of monitoring data sequences are constructed based on the influence ranking of the indicators, ensuring that the detection logic aligns with the on-site risk distribution patterns. Multiple isolated trees are used for collaborative discrimination, and the average path length is used to quantify the degree of anomaly. The anomaly coefficient is calculated by combining the indicator deviation, and a quantitative anomaly index is obtained by weighting based on the influence weight, accurately completing the graded judgment of trajectory node anomalies. This approach leverages the advantages of the isolated forest algorithm in efficiently identifying discrete anomaly data while also achieving adaptive parameter optimization in the context of hoisting operations, effectively improving the comprehensiveness and quantitative accuracy of trajectory anomaly identification during high-risk operations.

[0067] S400: Based on the construction monitoring index sequence, a trajectory prediction plugin is constructed in conjunction with the time series prediction model to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window, and to determine the second abnormal trajectory node set and the second abnormal index set.

[0068] In this embodiment of the invention, a trajectory prediction plugin is constructed based on the construction monitoring index sequence and combined with a time-series prediction model. This plugin performs anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window to determine the second abnormal trajectory node set and the second abnormal index set. Isolated forests can only determine the static data dispersion and cannot identify gradual anomalies, temporal misalignments, and trend deviations in the continuous evolution of the trajectory time sequence. The hoisting operation trajectory has continuous temporal correlation; compliance of an index at a single moment does not mean that the operation time sequence conforms to the standard procedure. Therefore, a multi-branch time-series prediction structure is built based on a long short-term memory network. Historical trajectory data is used to learn the normal change patterns of each index. The predicted value of the next node is deduced from the previous time-series data. The deviation between the measured data and the predicted data is compared to achieve time-series anomaly identification. Finally, the second abnormal trajectory node set and the second abnormal index set are determined, forming a complementary verification with the S300 static detection results.

[0069] Step S400 in the method provided in this embodiment of the invention includes: Based on the construction monitoring index sequence, a trajectory prediction plugin is constructed by combining a time-series prediction model, wherein the constructed trajectory prediction plugin includes multiple trajectory prediction branches; The monitoring data sequence set is divided according to the type of construction monitoring indicator to obtain multiple monitoring data sequences; Using the multiple trajectory prediction branches, anomaly detection is performed on the multiple monitoring data sequences respectively, and a second abnormal trajectory node set and a second abnormal index set are determined based on the anomaly detection results.

[0070] First, based on the construction monitoring index sequence, a trajectory prediction plugin is constructed in conjunction with a time-series prediction model. The trajectory prediction plugin includes multiple trajectory prediction branches.

[0071] The trajectory prediction plugin, constructed based on the construction monitoring index sequence and combined with a time-series prediction model, includes: Configure a data sliding selection window, wherein the data sliding selection window includes J continuous construction trajectory nodes; Randomly select any construction monitoring indicator from the construction monitoring indicator sequence as the first construction monitoring indicator; Constrained by the data sliding selection window and the first construction monitoring index, based on the historical construction trajectory monitoring log of the steel structure high-altitude hoisting operation, a sample first monitoring data sequence set is collected, and the historical first monitoring data of the next adjacent node of different sample first monitoring data sequences is used as the sample first prediction data to obtain the sample first prediction dataset; Using the first monitoring data sequence set and the first prediction dataset of the sample as training data, the long short-term memory network is trained until convergence, generating the first trajectory prediction branch, and then training multiple trajectory prediction branches for multiple construction monitoring indicators in sequence.

[0072] First, a data sliding selection window is configured, comprising J consecutive construction trajectory nodes. The data sliding selection window serves as the basic sampling unit for time series prediction, using a fixed number of consecutive construction trajectory nodes as input samples, and iteratively sliding forward to complete point-by-point prediction. J is a pre-set number of window nodes, with a fixed value maintained throughout the process. The window size balances temporal correlation and computational efficiency. Due to limitations of the sliding prediction mechanism, the first J initial construction trajectory nodes fully covered by the data sliding selection window have no preceding time series samples for prediction and are uniformly and defaulted to normal nodes, excluded from the time series anomaly screening scope.

[0073] Specifically, based on the total number of nodes within the sliding window of this adaptation monitoring, the number of consecutive nodes included in the data sliding selection window is reasonably set, and the window size is fixed and no longer dynamically changed; J consecutive trajectory nodes are selected in chronological order to form a set of time-series input samples, and the window slides backward one node at a time to generate new input samples one by one, so as to realize rolling prediction of time-series data throughout the entire time period; fixed rule constraints are applied, and the first J initial nodes are directly determined as normal nodes because they lack the pre-sliding sampling conditions.

[0074] For example, the adaptation monitors a total of 10 continuous construction trajectory nodes within the sliding window, numbered sequentially from node 1 to node 10. The data sliding selection window J=5, meaning that every 5 consecutive nodes constitute a set of time-series input samples. Nodes 1 to 5 are the initial coverage nodes, and since there is no preceding time-series data to complete the prediction and deduction, they are all judged as normal nodes by default. Starting from node 6, single-node prediction is completed sequentially using the data from the preceding 5 nodes, and time-series anomaly detection is carried out.

[0075] Secondly, any construction monitoring indicator is randomly selected from the construction monitoring indicator sequence as the first construction monitoring indicator. This first construction monitoring indicator serves as the first training object in the multi-branch prediction training, acting as the baseline branch for building the time-series model and unifying the basic structure, training parameters, and convergence conditions of the Long Short-Term Memory network. By keeping the construction monitoring indicator sequence fixed and arbitrarily selecting a single indicator for initial training, the structural homogeneity and parameter uniformity of the multi-branch model are ensured, avoiding computational biases caused by differences in the prediction model architecture for different indicators.

[0076] Specifically, based on the fixed construction monitoring index sequence generated by S200, one monitoring index is randomly selected from the sequence and designated as the first construction monitoring index. Based on the time-series data variation characteristics and distribution patterns of this index, a fixed topology and training hyperparameters of the Long Short-Term Memory (LSTM) network are uniformly set, serving as a universal template for the prediction branches corresponding to all monitoring indicators. The specific structural configuration of the LSTM network is as follows: two LSTM hidden layers are set, with 64 neurons in the first hidden layer and 32 neurons in the second hidden layer. Both hidden layers use the ReLU activation function, and the output layer is a fully connected layer using a linear activation function. Simultaneously, global training hyperparameters are fixed: the learning rate is set to 0.001, the maximum number of iterations is set to 100, the batch size is set to 16, and the mean squared error (MSE) is used as the loss function for training the LSTM network. The entire structure and parameters are standardized and reused in the training process of all subsequent index branches, ensuring a unified network architecture and consistent computational logic for the multi-trajectory prediction branches.

[0077] For example, the construction monitoring index sequence, ordered by impact, is as follows: horizontal movement speed of high-altitude workers, clear distance between the suspended object and obstacles, number of intersections between the hoisting path and the trajectory of personnel below, abnormality of boom rotation-amplitude coordination, alignment deviation of the suspended object's placement point, and deviation of the total hoisting cycle time. The clear distance between the suspended object and obstacles is randomly selected as the first construction monitoring index. Based on the data distribution characteristics and temporal variation patterns of this index, the basic training parameters for the long short-term memory network are set.

[0078] Furthermore, constrained by the aforementioned data sliding selection window and the first construction monitoring index, and based on the historical construction trajectory monitoring logs of steel structure high-altitude hoisting operations, a sample first monitoring data sequence set is collected. The historical first monitoring data of the next adjacent node of different sample first monitoring data sequences is used as the sample first prediction data, resulting in the sample first prediction dataset. The historical construction trajectory monitoring log is a long-term accumulated time-series database of all indicators for normal hoisting operations, recording the standard time-series changes of each indicator under different working conditions. The sample first monitoring data sequence set is a set of historical continuous time-series samples truncated according to the data sliding selection window size. The sample first prediction data is the actual measured data corresponding to the next moment of each set of historical time-series samples, serving as label data for supervised model training, thus achieving supervised time-series learning.

[0079] Specifically, historical monitoring logs of similar steel structure high-altitude hoisting operations are retrieved, and historical time-series data that are stable, compliant, and without safety anomalies are selected. According to the window size of J nodes, continuous time-series segments are batched to form the first monitoring data sequence set of the sample. For each historical time-series segment, the measured index data of the next adjacent node in the time dimension are extracted, and integrated and summarized into the first prediction dataset of the sample to complete the pairing and construction of model training samples and labels.

[0080] For example, based on the factory's past 30 days of normal hoisting operation logs, time series data without violations or abnormalities are selected, and 1000 sets of continuous 5-node historical time series segments are extracted according to the J=5 window specification to form the first monitoring data sequence set of the samples used in this training; the real net distance data of the 6th node after each 5-node segment is extracted simultaneously and uniformly collected into the first prediction dataset of the samples to form a complete training sample pair.

[0081] Finally, using the first monitoring data sequence set and the first prediction dataset as training data, the Long Short-Term Memory (LSTM) network is trained until convergence, generating the first trajectory prediction branch. Multiple trajectory prediction branches for various construction monitoring indicators are then trained sequentially. The LSM network is a time-series prediction model adapted to fit time-series data, effectively learning the time-series change trends, periodic patterns, and linkage characteristics of the indicators. The first trajectory prediction branch is a dedicated prediction model trained based on the first construction monitoring indicator. Multiple trajectory prediction branches correspond one-to-one with each construction monitoring indicator, with each indicator independently equipped with a dedicated training model to fit its own data change characteristics, ensuring the prediction accuracy of each indicator.

[0082] Specifically, the sample data corresponding to the first construction monitoring indicator is input into the Long Short-Term Memory (LSTM) network. Iterative optimization is performed based on a pre-defined, unified network structure and training parameters: the mean squared error (MSE) is used as the LSM network loss function, and iterative training is performed with a learning rate of 0.001, a batch size of 16, and a maximum number of iterations of 100. The model weights and biases are continuously updated to minimize the prediction error. The convergence condition for the LSM network is set as follows: after 10 consecutive iterations of training, the loss does not decrease significantly, and the loss value is below 0.001. Once this condition is met, the LSM network is considered to have converged, and the first trajectory prediction branch is generated. Subsequently, all remaining indicators in the construction monitoring indicator sequence are selected in rotation, strictly reusing the same network structure, activation function, learning rate, batch size, loss function, and convergence condition. The sample collection, model training, and convergence determination process are repeated, generating an independent trajectory prediction branch corresponding to each monitoring indicator. All prediction branches are then uniformly combined and embedded into the trajectory prediction plugin.

[0083] For example, using the net distance between the suspended object and the obstacle as the first construction monitoring indicator, after completing the training of sample data and meeting the convergence condition, the first trajectory prediction branch is generated. Keeping the two LSTM hidden layers (64 and 32 neurons), ReLU activation function, linear output layer, learning rate of 0.001, batch size of 16, maximum number of iterations of 100, mean square error loss function and convergence judgment condition completely unchanged, five indicators are trained independently in sequence: horizontal movement speed of high-altitude workers, number of intersections between the hoisting path and the trajectory of the personnel below, anomaly degree of boom rotation-amplitude coordination, alignment deviation of the suspended object's placement point, and deviation of the total hoisting cycle time. Finally, six independent trajectory prediction branches corresponding one-to-one with the monitoring indicators are generated, which are used for the time series numerical prediction of the corresponding indicators.

[0084] Secondly, the monitoring data sequence set is divided according to the type of construction monitoring indicators, resulting in multiple monitoring data sequences. After division by indicator type, the full-node time-series data of a single indicator independently form a monitoring data sequence, achieving multi-dimensional data decoupling and facilitating independent input and calculation for each trajectory prediction branch. For example, by traversing all construction trajectory nodes in the current period, the data is extracted according to the categories of six monitoring indicators, and then organized into six time-series data sequences of equal length. Each sequence completely records all measured values ​​of a single indicator from node 1 to node 10, with the data time sequence remaining consistent.

[0085] Next, using the multiple trajectory prediction branches, anomaly detection is performed on the multiple monitoring data sequences respectively, and the second abnormal trajectory node set and the second abnormal index set are determined based on the anomaly detection results.

[0086] Specifically, the multiple trajectory prediction branches are used to perform anomaly detection on the multiple monitoring data sequences, and the second abnormal trajectory node set and the second abnormal index set are determined based on the anomaly detection results, including: Based on the data sliding selection window, according to the data sequence, the multiple trajectory prediction branches are used to predict the indicators of multiple construction trajectory nodes in the multiple monitoring data sequences to obtain multiple second prediction data sequences. The initial construction trajectory nodes covered by the data sliding selection window are assumed to be normal nodes. The data deviations of the multiple monitoring data sequences and the multiple second prediction data sequences are calculated respectively. The data deviations are then weighted and summed according to the influence weights of the monitoring indicators in the construction monitoring indicator sequences to obtain the second deviation index sequence. Construction trajectory nodes whose second deviation index is greater than the preset deviation index threshold are marked as second abnormal trajectory nodes, and the second deviation index corresponding to the second abnormal trajectory node is used as the second abnormal index to obtain the second abnormal trajectory node set and the second abnormal index set.

[0087] First, based on the data sliding selection window, and according to the data sequence, multiple trajectory prediction branches are used to predict indicators for multiple construction trajectory nodes in the multiple monitoring data sequences, resulting in multiple second prediction data sequences. The initial construction trajectory nodes covered by the data sliding selection window are assumed to be normal nodes. The second prediction data sequences are a collection of predicted values ​​for all nodes obtained through prior time-series data extrapolation, corresponding one-to-one with the measured monitoring data sequences. During the sliding prediction process, J consecutive historical node data points are used as input to output the predicted indicator value for the next node, and the entire extrapolation process is completed by sliding window by window. The first J initial nodes have no prior input conditions, do not generate prediction values, are directly marked as normal nodes, and do not participate in deviation calculations.

[0088] Specifically, for each monitoring data sequence, the corresponding dedicated trajectory prediction branch is called, and the sampling is performed from front to back in chronological order. The initial J node data is used as the first input sample to predict the index value of the (J+1)th node. The data is then slid backward one node at a time, and all subsequent nodes are predicted in a loop. The results are then aggregated to generate the second prediction data sequence for each index. If there are no prediction results for the initial J nodes, no deviation comparison is performed, and they are uniformly identified as normal nodes.

[0089] For example, J=5, nodes 1 to 5 are the initial window coverage nodes, which are normal nodes by default; the predicted values ​​of each indicator of node 6 are predicted by nodes 1-5 in sequence, the predicted values ​​of each indicator of node 7 are predicted by nodes 2-6, the predicted values ​​of each indicator of node 8 are predicted by nodes 3-7, the predicted values ​​of each indicator of node 9 are predicted by nodes 4-8, and the predicted values ​​of each indicator of node 10 are predicted by nodes 5-9. Finally, the second predicted data sequence corresponding to the six indicators is generated, which completely covers the predicted values ​​of nodes 6 to 10.

[0090] Next, the data deviations of the multiple monitoring data sequences and the multiple second predicted data sequences are calculated separately. The data deviations are then weighted and summed according to the influence weights of the monitoring indicators in the construction monitoring indicator sequences to obtain the second deviation index sequence. Data deviation represents the degree of deviation between the measured value and the predicted value at the same trajectory node for a single indicator. It is uniformly converted to the 0-1 range using a linear normalization method; the larger the deviation, the higher the deviation value. The limit fluctuation range of each indicator is preset based on the equipment range or the maximum and minimum values ​​of historical data, for example, the limit range of horizontal movement speed is 0~2.5m / s. The influence weights are based on the indicator influence values ​​obtained from S200 quantization, and the weights remain fixed. The second deviation index is the weighted fusion value of the deviations of multiple indicators at a single time-series node, with a value in the 0-1 range. A higher index indicates a higher degree of anomaly in the time-series trajectory at that node. The second deviation index sequence corresponds one-to-one with the construction trajectory node time sequence.

[0091] The data deviation is calculated as follows: for the same indicator at the same node, the absolute difference between the measured value and the predicted value is taken, and a linear normalization conversion is performed in combination with the extreme fluctuation range of the indicator, converging to the 0-1 interval; the second deviation index is equal to the sum of the normalized data deviation of each indicator and the corresponding influence weight in sequence.

[0092] Specifically, the measured data and predicted data are compared one by one for each indicator and one by one to calculate the normalized data deviation; the fixed influence weights of the six indicators are retrieved, and the deviations of all indicators for a single node are weighted and fused to obtain the second deviation index of that node; all predictable nodes are calculated in chronological order and integrated to generate a time-continuous second deviation index sequence.

[0093] For example, the normalized influence weights of each indicator are as follows: horizontal movement speed of high-altitude workers 0.304, clear distance between the suspended object and obstacle 0.253, number of intersections between the hoisting path and the trajectory of personnel below 0.228, anomaly degree of boom rotation-amplitude coordination 0.152, alignment deviation of the suspended object placement point 0.050, and deviation of total hoisting cycle time 0.013. Taking node 8 as an example, after normalization and conversion, the data deviations between the measured data and the time-series predicted data of each indicator are 0.55, 0.05, 0.52, 0.60, 0.08, and 0.03, respectively; the second deviation index = 0.55×0.304+0.05×0.253+0.52×0.228+0.60×0.152+0.08×0.050+0.03×0.013≈0.394. The deviation index calculations for nodes 6, 7, 9, and 10 are completed synchronously. Initially, nodes 1 to 5 have no deviation index, which is uniformly recorded as 0, and finally a complete second deviation index sequence is formed.

[0094] Finally, construction trajectory nodes whose second deviation index exceeds the preset deviation index threshold are marked as second abnormal trajectory nodes, and the second deviation index corresponding to the second abnormal trajectory nodes is used as the second abnormal index, resulting in a set of second abnormal trajectory nodes and a set of second abnormal indices. The preset deviation index threshold is a pre-defined critical value for judging time-series anomalies, ranging from 0 to 1, and is set in conjunction with the deviation distribution statistics of historical normal trajectories to distinguish between time-series normal nodes and time-series abnormal nodes. Second abnormal trajectory nodes are trajectory points whose time-series evolution trend and index change patterns deviate from the standard operating mode. The second abnormal index is directly taken from the second deviation index of the corresponding node to quantitatively characterize the severity of time-series anomalies.

[0095] Specifically, a preset deviation index threshold is retrieved, and the second deviation index of each node from node 6 to node 10 is compared with the threshold. If the second deviation index is greater than the threshold, it is determined to be a second abnormal trajectory node and included in the set management. If it is less than or equal to the threshold, it is determined to be a time-series normal node. All abnormal nodes and their corresponding deviation indices are collected to form a one-to-one set of second abnormal trajectory nodes and a set of second abnormal indices.

[0096] For example, the preset deviation index threshold is set to 0.3. Node 8 has a second deviation index of 0.394 > 0.3, and is therefore identified as the second abnormal trajectory node. The deviation indices of the remaining predictable nodes are all below the threshold, and the temporal changes conform to the standard pattern. Thus, it is determined that the second abnormal trajectory node set includes node 8, the second abnormal index set corresponds to the recorded value of 0.394, and the initial nodes 1 to 5 are all determined to be normal nodes according to the rules and are not included in the abnormal range.

[0097] In this embodiment of the invention, a multi-branch temporal prediction structure is built based on a Long Short-Term Memory (LSTM) network, which aligns with the continuous temporal characteristics of hoisting trajectories and overcomes the shortcomings of static anomaly detection in identifying trend deviations and temporal anomalies. Scrolling prediction is achieved through a fixed-size sliding window, clearly defining the normal judgment rules for initial window nodes, resulting in clear boundary judgments and strong reproducibility. By combining the influence weights of various indicators and fusing multi-dimensional temporal deviations, a standardized second deviation index is quantified, achieving precise quantification of the degree of temporal anomalies. The multi-branch independent prediction architecture can adapt to the differentiated change patterns of different indicators, effectively improving the comprehensiveness of trajectory temporal anomaly identification and providing a complete temporal detection basis for subsequent global anomaly index fusion and comprehensive trajectory risk assessment.

[0098] S500: The final abnormal trajectory node set is determined by fusing the first abnormal trajectory node set, the first abnormal index set, the second abnormal trajectory node set, and the second abnormal index set, and is used as the construction trajectory abnormality monitoring result within the adaptive monitoring sliding window.

[0099] In this embodiment of the invention, a final abnormal trajectory node set is determined by fusing the first abnormal trajectory node set, the first abnormal index set, the second abnormal trajectory node set, and the second abnormal index set, and serves as the construction trajectory anomaly monitoring result within the adaptive monitoring sliding window. An adaptive weighted fusion of multi-source anomaly results is performed by combining the first abnormal trajectory node set and the first abnormal index set output by the isolated forest algorithm, and the second abnormal trajectory node set and the second abnormal index set output by the time-series prediction model. Single detection methods have limitations; the reliability of static anomaly detection varies with the number of monitoring indicators involved in the calculation. The fewer the indicators selected, the lower the reliability of the static detection result. It is necessary to dynamically reduce their weight ratio and increase the weight of the time-series prediction result. By fusing the two types of anomaly indices through differentiated weight allocation, a comprehensive anomaly index is calculated. The final abnormal trajectory node determination is completed based on a unified threshold, and a complete construction trajectory anomaly monitoring result within the adaptive monitoring sliding window is output.

[0100] Step S500 in the method provided in this embodiment of the invention includes: The first confidence level is obtained by summing the influence weights of the first monitoring indicator set. The product of the first confidence level and the preset first weight is used as the first adaptation weight, and the second adaptation weight is obtained by subtracting the first adaptation weight from 1. Based on the affiliation of each construction trajectory node in the first abnormal trajectory node set and the second abnormal trajectory node set, the corresponding first abnormal index is multiplied by the first adaptive weight, and / or the corresponding second abnormal index is multiplied by the second adaptive weight to obtain the comprehensive abnormal index of the construction trajectory node. When the comprehensive anomaly index is greater than the preset comprehensive anomaly index threshold, the corresponding construction trajectory node is determined as the final abnormal trajectory node and included in the final abnormal trajectory node set.

[0101] First, the influence weights of the first monitoring indicator set are summed to obtain the first confidence level. The current monitoring indicators are all indicators within the first monitoring indicator set selected by S300. The first confidence level is used to quantify the reliability of the static anomaly detection results in isolated forests. The fewer the number of indicators in the first monitoring indicator set, the more limited the indicator coverage dimensions, and the lower the overall first confidence level, indicating a decrease in the reference value of the static detection results. The influence weights of each indicator directly adopt the fixed influence values ​​from the construction monitoring indicator sequence to ensure consistency in the weighting system.

[0102] Specifically, the influence values ​​corresponding to all monitoring indicators covered by the current first monitoring indicator set are extracted, and then summed item by item. The sum is the first confidence level used in this fusion calculation, providing a basic parameter for the subsequent conversion of adaptive fusion weights. For example, if the current first monitoring indicator set has 4 monitoring indicators with corresponding influence weights of 0.304, 0.253, 0.228, and 0.152 respectively, the first confidence level = 0.304 + 0.253 + 0.228 + 0.152 = 0.937.

[0103] Secondly, the product of the first confidence level and the preset first weight is used as the first adaptation weight, and the second adaptation weight is obtained by subtracting the first adaptation weight from 1. The preset first weight is a pre-set fixed benchmark weight, which serves as the basic proportion of the static detection results, for example, 0.6. The first adaptation weight is the dynamically corrected fusion weight of the isolated forest results, which changes positively with the first confidence level. The second adaptation weight is the fusion weight of the time series prediction model results, and the sum of the two is always equal to 1, satisfying the weight normalization constraint. The fewer the number of monitoring indicators involved in the detection, the lower the first confidence level, the smaller the first adaptation weight, and the larger the second adaptation weight, thereby increasing the weight proportion of the time series prediction results in low-confidence scenarios.

[0104] Specifically, a preset first weight value is retrieved, and the first confidence level is multiplied by the preset first weight to calculate the adapted first weight. Using 1 as a baseline, the adapted first weight is subtracted, and the adapted second weight is calculated in reverse, thus completing the adaptive configuration of the dual-model fusion weights. For example, if the preset first weight is 0.6 and the first confidence level is 0.937, the adapted first weight = 0.937 × 0.6 = 0.5622, and the adapted second weight = 1 − 0.5622 = 0.4378, thereby completing the dynamic weight allocation of the two types of anomaly detection results.

[0105] Next, based on the affiliation of each construction trajectory node in the first abnormal trajectory node set and the second abnormal trajectory node set, the corresponding first abnormal index is multiplied by the first adaptive weight, and / or the corresponding second abnormal index is multiplied by the second adaptive weight to obtain the comprehensive abnormal index of the construction trajectory node.

[0106] Specifically, for each construction trajectory node, when the construction trajectory node exists only in the first abnormal trajectory node set, the first abnormal index corresponding to the construction trajectory node is multiplied by the first adaptive weight to obtain the comprehensive abnormal index of the construction trajectory node; when the construction trajectory node exists only in the second abnormal trajectory node set, the second abnormal index corresponding to the construction trajectory node is multiplied by the second adaptive weight to obtain the comprehensive abnormal index of the construction trajectory node; when the construction trajectory node exists in both the first abnormal trajectory node set and the second abnormal trajectory node set, the first abnormal index corresponding to the construction trajectory node is multiplied by the first adaptive weight to obtain a first weighted value, the second abnormal index corresponding to the construction trajectory node is multiplied by the second adaptive weight to obtain a second weighted value, and the first weighted value and the second weighted value are added to obtain the comprehensive abnormal index of the construction trajectory node. For example, node 8 belongs to both the first abnormal trajectory node set and the second abnormal trajectory node set. The first abnormality index is 0.392, the second abnormality index is 0.394, the first adaptation weight is 0.5622, and the second adaptation weight is 0.4378. The comprehensive abnormality index of node 8 is approximately 0.393 (0.392 × 0.5622 + 0.394 × 0.4378). The other normal nodes have no abnormality index, and their comprehensive abnormality index is uniformly recorded as 0.

[0107] Finally, when the comprehensive anomaly index exceeds the preset comprehensive anomaly index threshold, the corresponding construction trajectory node is determined as the final abnormal trajectory node and included in the final abnormal trajectory node set. The preset comprehensive anomaly index threshold is a globally unified anomaly judgment threshold, ranging from 0 to 1, and is calibrated by combining on-site hoisting operation safety management standards with historical fusion data. The final abnormal trajectory node set is a quasi-anomaly set after dual-model fusion verification, serving as the final output result of this sliding window trajectory anomaly monitoring.

[0108] Specifically, the preset comprehensive anomaly index threshold is retrieved, and the comprehensive anomaly index of all construction trajectory nodes is compared one by one. If the comprehensive anomaly index exceeds the threshold, it is determined to be a final abnormal trajectory node and uniformly collected into the final abnormal trajectory node set. If the value does not exceed the threshold, it is determined to be a normal trajectory node. Finally, the collected abnormal node set is output to complete the monitoring of anomalies in the entire construction trajectory.

[0109] For example, in accordance with the safety management requirements for high-altitude steel structure hoisting operations, a preset comprehensive anomaly index threshold of 0.25 is established. The comprehensive anomaly index of each node is compared one by one. Node 8 has a comprehensive anomaly index of 0.393 > 0.25, and is therefore identified as an abnormal trajectory node; the comprehensive anomaly indices of nodes 1-7 and 9-10 are both 0 < 0.25, and are therefore identified as normal trajectory nodes, not included in the anomaly set. The final set of abnormal trajectory nodes is determined to be node 8, which represents the final trajectory anomaly monitoring result for high-altitude steel structure hoisting operations within the adaptive monitoring sliding window (389 seconds). Those skilled in the art can use this result to promptly identify potential hazards in hoisting operations, adjust construction procedures, and accurately implement safety management measures.

[0110] In this embodiment of the invention, the adaptive fusion of anomaly results from two models effectively solves the problems of false detection and missed detection caused by the limitations of a single detection model. It achieves dynamic adaptation of fusion weights, using the first confidence level to correlate and adapt the first and second weights. The fewer the number of monitoring indicators, the lower the credibility of static detection, and the higher the weight ratio of time-series prediction, ensuring accurate matching between weight allocation and detection credibility, and improving the rationality of the fusion results. A scenario-specific fusion strategy is adopted, designing exclusive fusion formulas for different situations where nodes exist only in a single anomaly set or simultaneously in two anomaly sets. This avoids the one-sidedness of single-model results while achieving complementary advantages between the two models, accurately quantifying the comprehensive anomaly degree of each node. The final judgment is completed through a unified comprehensive anomaly threshold. The output final anomaly trajectory node set undergoes static and time-series dual-model verification, taking into account the identification needs of both static discrete anomalies and time-series trend anomalies. This effectively improves the accuracy and reliability of anomaly monitoring in high-altitude steel structure hoisting, providing clear anomaly location information for on-site safety management and reducing safety risks in high-altitude hoisting operations.

[0111] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides a GIS-based method and system for monitoring construction site trajectories. First, it scientifically ranks monitoring indicators based on risk weights. Then, it dynamically adjusts the detection dimensions in conjunction with real-time construction risks, conducting bidirectional anomaly mining from two dimensions: data discrete distribution and temporal evolution patterns. Simultaneously, it dynamically allocates fusion weights based on the coverage of the detection indicators, compensating for the identification defects and applicability limitations of single detection algorithms. This allows for accurate identification of both sudden discrete anomalies and gradual trend anomalies, achieving adaptive detection under different operational risk conditions. It effectively reduces the probability of missed or false anomalies, comprehensively enhancing the dynamic monitoring capability of steel structure high-altitude hoisting operations, and providing reliable technical support for on-site construction risk early warning, operational standard rectification, and routine safety management.

[0112] Example 2, as Figure 2As shown, this invention provides a GIS-based construction site trajectory monitoring system, the system comprising: The dynamic window adjustment module 11 is used to dynamically adjust the preset monitoring sliding window according to the construction risk index of the high-altitude hoisting operation of the steel structure within a preset time period, and to determine the appropriate monitoring sliding window. The indicator sequence configuration module 12 is used to configure the construction monitoring indicator sequence for high-altitude steel structure hoisting operations; The isolated forest detection module 13 is used to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window based on the construction risk index and construction monitoring index sequence, combined with the isolated forest algorithm, to determine the first abnormal trajectory node set and the first abnormal index set; The time series prediction and detection module 14 is used to construct a trajectory prediction plugin based on the construction monitoring index sequence and the time series prediction model, and to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window to determine the second abnormal trajectory node set and the second abnormal index set. The anomaly fusion determination module 15 is used to determine the final anomaly trajectory node set based on the fusion of the first anomaly trajectory node set, the first anomaly index set, the second anomaly trajectory node set, and the second anomaly index set, as the construction trajectory anomaly monitoring result within the adaptive monitoring sliding window.

[0113] In one embodiment, the dynamic window adjustment module 11 is further configured to: Multiple risk factors for high-altitude steel structure hoisting operations within a preset time period are obtained. After normalization of each risk factor, a weighted evaluation is performed to determine the construction risk index. The multiple risk factors include at least the real-time wind speed level, the ratio of the weight of the hoisted object to the rated load, the continuous working time of the operators, and the number of personnel in the warning zone below the hoisted object. The ratio of the preset benchmark construction risk index to the construction risk index is used as the window duration compensation coefficient, and the product of the window duration compensation coefficient and the preset duration of the monitoring sliding window is used as the adaptive monitoring sliding window.

[0114] In one embodiment, the indicator sequence configuration module 12 is further configured to: A set of construction monitoring indicators for high-altitude steel structure hoisting operations is configured, wherein the set of construction monitoring indicators includes the clear distance between the hoisted object and the obstacle, the horizontal movement speed of the high-altitude workers, the number of times the hoisting path intersects with the trajectory of the personnel below, the abnormality of the boom rotation-amplitude coordination, the alignment deviation of the hoisted object placement point, and the deviation of the total hoisting cycle time. The monitoring indicators in the construction monitoring indicator set are sorted from largest to smallest according to their impact to generate a construction monitoring indicator sequence. The impact of each monitoring indicator is determined based on the severity of the safety accident corresponding to each monitoring indicator and the probability of the safety accident occurring.

[0115] In one embodiment, the isolated forest detection module 13 is further used for: The ratio of the construction risk index to the preset benchmark construction risk index is multiplied by the number of benchmark monitoring indicators selected and rounded down to obtain the current number of monitoring indicators selected. The number of benchmark monitoring indicators selected is less than or equal to half of the total number of construction monitoring indicators, and the current number of monitoring indicators selected is not less than 2. Based on the current number of monitoring indicators selected, the construction monitoring indicator sequence is filtered to obtain the first monitoring indicator set; The monitoring data sequence set within the adaptive monitoring sliding window is filtered based on the first monitoring index set to obtain the first monitoring data sequence set, wherein the monitoring data sequence set includes monitoring data of several construction trajectory nodes within the adaptive monitoring sliding window; Based on the first monitoring indicator set, K monitoring data are randomly selected from each monitoring indicator data in the first monitoring data sequence set to build an isolated tree, generating multiple first isolated trees; Based on the multiple first isolated trees, anomaly judgment is performed on the monitoring data in the first monitoring data sequence set to determine the first abnormal trajectory node set and the first abnormal index set.

[0116] Specifically, based on the plurality of first isolated trees, anomaly detection is performed on the monitoring data in the first monitoring data sequence set to determine the first abnormal trajectory node set and the first abnormal index set, including: For each construction trajectory node in the first monitoring data sequence set, the average path length of the current construction trajectory node in the multiple first isolated trees is calculated, and the anomaly score of the current construction trajectory node is calculated based on the average path length. When the abnormal score is greater than the preset first abnormal score threshold, the current construction trajectory node is determined as the first abnormal trajectory node and is included in the first abnormal trajectory node set. For each first abnormal trajectory node in the first abnormal trajectory node set, the abnormality coefficient of each monitoring indicator in the first monitoring indicator set corresponding to the current first abnormal trajectory node is obtained, wherein the abnormality coefficient is determined based on the degree to which the monitoring data of the monitoring indicator at the construction trajectory node deviates from the normal range. For each first abnormal trajectory node in the first abnormal trajectory node set, the abnormality coefficients of each monitoring indicator corresponding to the current first abnormal trajectory node are weighted and summed according to the influence weight of the monitoring indicator in the construction monitoring indicator sequence to calculate the first abnormal index of the current first abnormal trajectory node. The first abnormal indices of all first abnormal trajectory nodes are then aggregated to obtain the first abnormal index set.

[0117] In one embodiment, the time-series prediction detection module 14 is further configured to: Based on the construction monitoring index sequence, a trajectory prediction plugin is constructed by combining a time-series prediction model, wherein the constructed trajectory prediction plugin includes multiple trajectory prediction branches; The monitoring data sequence set is divided according to the type of construction monitoring indicator to obtain multiple monitoring data sequences; Using the multiple trajectory prediction branches, anomaly detection is performed on the multiple monitoring data sequences respectively, and a second abnormal trajectory node set and a second abnormal index set are determined based on the anomaly detection results.

[0118] The trajectory prediction plugin, constructed based on the construction monitoring index sequence and combined with a time-series prediction model, includes: Configure a data sliding selection window, wherein the data sliding selection window includes J continuous construction trajectory nodes; Randomly select any construction monitoring indicator from the construction monitoring indicator sequence as the first construction monitoring indicator; Constrained by the data sliding selection window and the first construction monitoring index, based on the historical construction trajectory monitoring log of the steel structure high-altitude hoisting operation, a sample first monitoring data sequence set is collected, and the historical first monitoring data of the next adjacent node of different sample first monitoring data sequences is used as the sample first prediction data to obtain the sample first prediction dataset; Using the first monitoring data sequence set and the first prediction dataset of the sample as training data, the long short-term memory network is trained until convergence, generating the first trajectory prediction branch, and then training multiple trajectory prediction branches for multiple construction monitoring indicators in sequence.

[0119] Specifically, the multiple trajectory prediction branches are used to perform anomaly detection on the multiple monitoring data sequences, and the second abnormal trajectory node set and the second abnormal index set are determined based on the anomaly detection results, including: Based on the data sliding selection window, according to the data sequence, the multiple trajectory prediction branches are used to predict the indicators of multiple construction trajectory nodes in the multiple monitoring data sequences to obtain multiple second prediction data sequences. The initial construction trajectory nodes covered by the data sliding selection window are assumed to be normal nodes. The data deviations of the multiple monitoring data sequences and the multiple second prediction data sequences are calculated respectively. The data deviations are then weighted and summed according to the influence weights of the monitoring indicators in the construction monitoring indicator sequences to obtain the second deviation index sequence. Construction trajectory nodes whose second deviation index is greater than the preset deviation index threshold are marked as second abnormal trajectory nodes, and the second deviation index corresponding to the second abnormal trajectory node is used as the second abnormal index to obtain the second abnormal trajectory node set and the second abnormal index set.

[0120] In one embodiment, the anomaly fusion determination module 15 is further configured to: The first confidence level is obtained by summing the influence weights of the first monitoring indicator set. The product of the first confidence level and the preset first weight is used as the first adaptation weight, and the second adaptation weight is obtained by subtracting the first adaptation weight from 1. Based on the affiliation of each construction trajectory node in the first abnormal trajectory node set and the second abnormal trajectory node set, the corresponding first abnormal index is multiplied by the first adaptive weight, and / or the corresponding second abnormal index is multiplied by the second adaptive weight to obtain the comprehensive abnormal index of the construction trajectory node. When the comprehensive anomaly index is greater than the preset comprehensive anomaly index threshold, the corresponding construction trajectory node is determined as the final abnormal trajectory node and included in the final abnormal trajectory node set.

[0121] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A GIS-based method for monitoring construction site trajectories, characterized in that, The method includes: The preset monitoring sliding window is dynamically adjusted based on the construction risk index of high-altitude steel structure hoisting operations within a preset time period, and the appropriate monitoring sliding window is determined. Configure a sequence of construction monitoring indicators for high-altitude steel structure hoisting operations; Based on the construction risk index and construction monitoring index sequence, and combined with the isolated forest algorithm, anomaly detection is performed on the monitoring data sequence set within the adaptive monitoring sliding window to determine the first abnormal trajectory node set and the first abnormal index set. Based on the construction monitoring index sequence, a trajectory prediction plugin is constructed by combining the time series prediction model to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window, and to determine the second abnormal trajectory node set and the second abnormal index set. The final abnormal trajectory node set is determined by fusing the first abnormal trajectory node set, the first abnormal index set, the second abnormal trajectory node set, and the second abnormal index set, and is used as the construction trajectory abnormality monitoring result within the adaptive monitoring sliding window.

2. The GIS-based construction site trajectory monitoring method according to claim 1, characterized in that, The preset monitoring sliding window is dynamically adjusted based on the construction risk index of high-altitude steel structure hoisting operations within a preset time period, and the appropriate monitoring sliding window is determined, including: Multiple risk factors for high-altitude steel structure hoisting operations within a preset time period are obtained. After normalization of each risk factor, a weighted evaluation is performed to determine the construction risk index. The multiple risk factors include at least the real-time wind speed level, the ratio of the weight of the hoisted object to the rated load, the continuous working time of the operators, and the number of personnel in the warning zone below the hoisted object. The ratio of the preset benchmark construction risk index to the construction risk index is used as the window duration compensation coefficient, and the product of the window duration compensation coefficient and the preset duration of the monitoring sliding window is used as the adaptive monitoring sliding window.

3. The GIS-based construction site trajectory monitoring method according to claim 1, characterized in that, Configure a series of construction monitoring indicators for high-altitude steel structure hoisting operations, including: A set of construction monitoring indicators for high-altitude steel structure hoisting operations is configured, wherein the set of construction monitoring indicators includes the clear distance between the hoisted object and the obstacle, the horizontal movement speed of the high-altitude workers, the number of times the hoisting path intersects with the trajectory of the personnel below, the abnormality of the boom rotation-amplitude coordination, the alignment deviation of the hoisted object placement point, and the deviation of the total hoisting cycle time. The monitoring indicators in the construction monitoring indicator set are sorted from largest to smallest according to their impact to generate a construction monitoring indicator sequence. The impact of each monitoring indicator is determined based on the severity of the safety accident corresponding to each monitoring indicator and the probability of the safety accident occurring.

4. The GIS-based construction site trajectory monitoring method according to claim 1, characterized in that, Based on the construction risk index and construction monitoring index sequence, and combined with the isolated forest algorithm, anomaly detection is performed on the monitoring data sequence set within the adaptive monitoring sliding window to determine the first abnormal trajectory node set and the first abnormal index set, including: The ratio of the construction risk index to the preset benchmark construction risk index is multiplied by the number of benchmark monitoring indicators selected and rounded down to obtain the current number of monitoring indicators selected. The number of benchmark monitoring indicators selected is less than or equal to half of the total number of construction monitoring indicators, and the current number of monitoring indicators selected is not less than 2. Based on the current number of monitoring indicators selected, the construction monitoring indicator sequence is filtered to obtain the first monitoring indicator set; The monitoring data sequence set within the adaptive monitoring sliding window is filtered based on the first monitoring index set to obtain the first monitoring data sequence set, wherein the monitoring data sequence set includes monitoring data of several construction trajectory nodes within the adaptive monitoring sliding window; Based on the first monitoring indicator set, K monitoring data are randomly selected from each monitoring indicator data in the first monitoring data sequence set to build an isolated tree, generating multiple first isolated trees; Based on the multiple first isolated trees, anomaly judgment is performed on the monitoring data in the first monitoring data sequence set to determine the first abnormal trajectory node set and the first abnormal index set.

5. The GIS-based construction site trajectory monitoring method according to claim 4, characterized in that, Based on the multiple first isolated trees, anomaly detection is performed on the monitoring data in the first monitoring data sequence set to determine the first abnormal trajectory node set and the first abnormal index set, including: For each construction trajectory node in the first monitoring data sequence set, the average path length of the current construction trajectory node in the multiple first isolated trees is calculated, and the anomaly score of the current construction trajectory node is calculated based on the average path length. When the abnormal score is greater than the preset first abnormal score threshold, the current construction trajectory node is determined as the first abnormal trajectory node and is included in the first abnormal trajectory node set. For each first abnormal trajectory node in the first abnormal trajectory node set, the abnormality coefficient of each monitoring indicator in the first monitoring indicator set corresponding to the current first abnormal trajectory node is obtained, wherein the abnormality coefficient is determined based on the degree to which the monitoring data of the monitoring indicator at the construction trajectory node deviates from the normal range. For each first abnormal trajectory node in the first abnormal trajectory node set, the abnormality coefficients of each monitoring indicator corresponding to the current first abnormal trajectory node are weighted and summed according to the influence weight of the monitoring indicator in the construction monitoring indicator sequence to calculate the first abnormal index of the current first abnormal trajectory node. The first abnormal indices of all first abnormal trajectory nodes are then aggregated to obtain the first abnormal index set.

6. The GIS-based construction site trajectory monitoring method according to claim 1, characterized in that, Based on the construction monitoring index sequence, a trajectory prediction plugin is constructed using a time-series prediction model to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window, determining a second set of abnormal trajectory nodes and a second set of abnormal indices, including: Based on the construction monitoring index sequence, a trajectory prediction plugin is constructed by combining a time-series prediction model, wherein the constructed trajectory prediction plugin includes multiple trajectory prediction branches; The monitoring data sequence set is divided according to the type of construction monitoring indicator to obtain multiple monitoring data sequences; Using the multiple trajectory prediction branches, anomaly detection is performed on the multiple monitoring data sequences respectively, and a second abnormal trajectory node set and a second abnormal index set are determined based on the anomaly detection results.

7. The GIS-based construction site trajectory monitoring method according to claim 6, characterized in that, Based on the aforementioned construction monitoring index sequence, a trajectory prediction plugin is constructed using a time-series prediction model, including: Configure a data sliding selection window, wherein the data sliding selection window includes J continuous construction trajectory nodes; Randomly select any construction monitoring indicator from the construction monitoring indicator sequence as the first construction monitoring indicator; Constrained by the data sliding selection window and the first construction monitoring index, based on the historical construction trajectory monitoring log of the steel structure high-altitude hoisting operation, a sample first monitoring data sequence set is collected, and the historical first monitoring data of the next adjacent node of different sample first monitoring data sequences is used as the sample first prediction data to obtain the sample first prediction dataset; Using the first monitoring data sequence set and the first prediction dataset of the sample as training data, the long short-term memory network is trained until convergence, generating the first trajectory prediction branch, and then training multiple trajectory prediction branches for multiple construction monitoring indicators in sequence.

8. The GIS-based construction site trajectory monitoring method according to claim 7, characterized in that, Using the multiple trajectory prediction branches, anomaly detection is performed on the multiple monitoring data sequences respectively. Based on the anomaly detection results, a second set of abnormal trajectory nodes and a second set of abnormal indices are determined, including: Based on the data sliding selection window, according to the data sequence, the multiple trajectory prediction branches are used to predict the indicators of multiple construction trajectory nodes in the multiple monitoring data sequences to obtain multiple second prediction data sequences. The initial construction trajectory nodes covered by the data sliding selection window are assumed to be normal nodes. The data deviations of the multiple monitoring data sequences and the multiple second prediction data sequences are calculated respectively. The data deviations are then weighted and summed according to the influence weights of the monitoring indicators in the construction monitoring indicator sequences to obtain the second deviation index sequence. Construction trajectory nodes whose second deviation index is greater than the preset deviation index threshold are marked as second abnormal trajectory nodes, and the second deviation index corresponding to the second abnormal trajectory node is used as the second abnormal index to obtain the second abnormal trajectory node set and the second abnormal index set.

9. The GIS-based construction site trajectory monitoring method according to claim 1, characterized in that, The final abnormal trajectory node set is determined by fusing the first abnormal trajectory node set, the first abnormal index set, the second abnormal trajectory node set, and the second abnormal index set, including: The first confidence level is obtained by summing the influence weights of the first monitoring indicator set. The product of the first confidence level and the preset first weight is used as the first adaptation weight, and the second adaptation weight is obtained by subtracting the first adaptation weight from 1. Based on the affiliation of each construction trajectory node in the first abnormal trajectory node set and the second abnormal trajectory node set, the corresponding first abnormal index is multiplied by the first adaptive weight, and / or the corresponding second abnormal index is multiplied by the second adaptive weight to obtain the comprehensive abnormal index of the construction trajectory node. When the comprehensive anomaly index is greater than the preset comprehensive anomaly index threshold, the corresponding construction trajectory node is determined as the final abnormal trajectory node and included in the final abnormal trajectory node set.

10. A GIS-based construction site trajectory monitoring system, characterized in that, For implementing the GIS-based construction site trajectory monitoring method according to any one of claims 1-9, the system comprises: The dynamic window adjustment module is used to dynamically adjust the preset monitoring sliding window according to the construction risk index of the high-altitude steel structure hoisting operation within a preset time period, and to determine the appropriate monitoring sliding window. The indicator sequence configuration module is used to configure the construction monitoring indicator sequence for high-altitude steel structure hoisting operations; The isolated forest detection module is used to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window based on the construction risk index and construction monitoring index sequence, combined with the isolated forest algorithm, to determine the first abnormal trajectory node set and the first abnormal index set. The time-series prediction and detection module is used to construct a trajectory prediction plugin based on the construction monitoring index sequence and the time-series prediction model, and to perform anomaly detection on the monitoring data sequence set within the adaptive monitoring sliding window to determine the second abnormal trajectory node set and the second abnormal index set. The anomaly fusion determination module is used to determine the final anomaly trajectory node set by fusing the first anomaly trajectory node set, the first anomaly index set, the second anomaly trajectory node set, and the second anomaly index set, as the construction trajectory anomaly monitoring result within the adaptive monitoring sliding window.

Images