Construction environment multi-parameter dynamic threshold self-adaptive early warning method

Abstract

The application relates to the field of construction safety intelligent monitoring, and discloses a construction environment multi-parameter dynamic threshold self-adaptive early warning method, which comprises the following steps: a multi-parameter real-time acquisition step: a plurality of monitoring nodes are arranged in layers at a construction site, a plurality of types of monitoring parameters of the construction environment are synchronously acquired through the monitoring nodes, and the monitoring parameters include at least two of air quality type parameters, physical environment type parameters, safety state type parameters and structure operation type parameters. Through a multi-dimensional data preprocessing mechanism, abnormal data can be accurately removed and missing data can be effectively compensated; invalid data interference caused by transient disturbance of equipment and transient mutation of the environment at the construction site is filtered through short-time fluctuation suppression correction logic; and a joint determination mechanism based on parameter change rate and over-limit duration is used, traditional rough determination logic that alarms as soon as a single instant is over-limit is abandoned, false alarm problems caused by construction condition fluctuation are greatly avoided, and the accuracy of construction environment risk early warning is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring of building construction safety, specifically to an adaptive early warning method for dynamic thresholds of multiple parameters in the construction environment. Background Technology

[0002] Construction sites, including those for buildings, tunnels, underground engineering, deep foundation pits, and open-air high-altitude projects, are characterized by complex working environments, dynamic and changing conditions, high personnel and equipment mobility, significant external weather interference, and diverse risk factors. During construction, various safety hazards are easily generated, such as excessive dust, leaks of toxic and flammable gases, abnormal temperature and humidity, substandard wind speeds, excessive vibration, water accumulation, and insufficient visibility. If these hazards are not identified and warned of in a timely and accurate manner, they can easily lead to safety accidents such as collapses, poisoning, explosions, falls from heights, and dust pollution, seriously threatening the lives of personnel on construction sites and the progress of the project. Therefore, real-time monitoring and intelligent early warning of multi-dimensional parameters of the construction environment are a core and crucial link in ensuring safe construction.

[0003] Currently, most domestic construction environment safety early warning systems adopt a fixed threshold-based warning mechanism. This means that for single monitoring parameters such as dust, gas, wind speed, and vibration, fixed upper and lower thresholds are set according to general industry standards. An alarm is triggered only when the monitored data momentarily exceeds the fixed threshold, thus providing a risk warning at the construction site. While this traditional early warning method is simple in architecture and easy to deploy, and is widely used in various conventional construction sites, it has many inherent technical shortcomings in complex and dynamically changing engineering scenarios, failing to meet the safety management needs of modern intelligent construction.

[0004] Firstly, existing technologies suffer from statically fixed thresholds and poor scenario adaptability. Safety assessment standards at construction sites vary significantly depending on the construction scenario, stage, type of work, and weather conditions. Safety baselines differ drastically across different scenarios, such as tunnel ventilation, foundation pit drainage, open-air high-altitude operations, and underground confined space construction. Traditional fixed thresholds cannot be dynamically adjusted according to on-site conditions, and uniform assessment standards cannot meet the diverse safety management needs of different scenarios, easily leading to scenario mismatch issues. Secondly, existing technologies have a simplistic early warning logic, relying solely on instantaneous exceedances of single parameters to determine risk, without considering the rate of parameter change, duration of exceedances, and the coupling and correlation of multiple parameters. Non-risk factors such as equipment start-up and shutdown, instantaneous environmental disturbances, and short-term sensor fluctuations at the construction site easily trigger frequent false alarms. Furthermore, hidden complex risks caused by the coordinated changes of multiple parameters cannot be effectively identified, resulting in persistently high false alarm and false negative rates.

[0005] Meanwhile, existing traditional early warning systems lack autonomous iterative optimization capabilities. Threshold parameters and risk judgment rules are mostly manually preset, unable to adaptively optimize based on historical monitoring data, early warning records, and on-site accident cases. Long-term use easily leads to outdated judgment standards and decreased adaptability. Furthermore, traditional early warning modes can only achieve simple over-limit alarms, failing to predict risk trends, provide refined hierarchical early warnings, and coordinate equipment responses. The early warning methods are crude, and responses are delayed, making it difficult to achieve proactive prevention and closed-loop management of construction risks, and unable to adapt to the intelligent and refined safety management needs of complex construction sites. Based on these numerous shortcomings of existing technologies, there is an urgent need for an intelligent early warning method for construction environments that can adapt to multiple construction scenarios, dynamically update early warning thresholds, and achieve multi-parameter coupled risk assessment and closed-loop self-learning. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a multi-parameter dynamic threshold adaptive early warning method for construction environments, which solves the problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a multi-parameter dynamic threshold adaptive early warning method for construction environment, comprising the following steps:

[0008] S1. Multi-parameter real-time acquisition steps: Multiple sets of monitoring nodes are deployed in layers at the construction site. Various monitoring parameters of the construction environment are collected synchronously through each monitoring node. The monitoring parameters include at least two of the following: air quality parameters, physical environment parameters, safety status parameters, and structural operation parameters. The monitoring nodes adopt a combination of regional basic deployment and high-risk area densification deployment, including fixed, pole-mounted, and mobile deployment forms. In tunnel and underground long-distance construction scenarios, a series full-coverage deployment along the line is adopted to eliminate monitoring blind spots.

[0009] S2. Data Preprocessing Steps: Perform time synchronization calibration, physical boundary outlier removal, abrupt isolated outlier data screening, multi-sensor consistency verification, and sliding window noise reduction filtering on the collected raw monitoring data. Simultaneously, perform missing value compensation on short-term missing data and output a valid monitoring data sequence that meets the reliability score. Among them, the missing value compensation methods include forward filling, window mean compensation, neighbor node interpolation, and trend extrapolation compensation. During sensor consistency verification, if the data of a single node deviates from the overall trend of the region, the sensor is judged to be faulty. The data of the faulty node is downweighted, removed, or replaced with redundant node data in the same region.

[0010] S3. Construction Scene Status Recognition Steps: Based on the construction area type, work type, spatial environment type, time period of the day, meteorological environment, and historical working condition data, feature matching and recognition are performed to output a unique corresponding scene label. The scene label includes at least tunnel construction scene, deep foundation pit construction scene, underground space construction scene, open-air high-altitude construction scene, and high-noise processing construction scene.

[0011] S4. Initial threshold model loading step: Retrieve the preset initial threshold model of the corresponding scene according to the identified scene label. The initial threshold model configures the basic upper limit threshold, basic lower limit threshold, parameter change rate threshold, and over-limit duration threshold for each type of monitoring parameter. The initial threshold is jointly set based on national construction safety standards, industry specifications, project historical statistical data, and on-site engineering experience.

[0012] S5. Dynamic Threshold Adaptive Update Step: Integrate four dimensions of data: effective monitoring data, full-cycle historical monitoring data, current scene labels, and parameter change trends. Perform multi-factor weighted correction on the initial threshold model to generate a dynamic threshold set adapted to the current working conditions. The dynamic threshold set includes single-parameter dynamic thresholds, multi-parameter composite thresholds, and risk trend thresholds. The dynamic threshold update synchronously executes five types of correction logic: sliding window statistical correction, scene weight correction, trend prediction correction, short-term fluctuation anomaly suppression correction, and historical early warning accident sample iterative correction.

[0013] S6. Multi-parameter coupling risk calculation steps: Compare real-time effective monitoring data with dynamic threshold sets item by item, and combine four dimensions of indicators: single parameter exceedance magnitude, real-time parameter change rate, exceedance duration, and multi-parameter linkage coupling relationship to calculate a weighted comprehensive risk score and match the corresponding risk level; among them, the multi-parameter coupling judgment rules include: when dust concentration increases and wind speed decreases, determine the risk of insufficient ventilation; when temperature and humidity increase and combustible gas exceed the standard, determine the risk of heat and moisture leakage; when noise continuously exceeds the standard and vibration is abnormal, determine the risk of equipment structure; when smoke concentration increases and illuminance drops sharply, determine the risk of fire visibility.

[0014] S7. Tiered early warning output steps: Pre-divide the four early warning levels, match the corresponding early warning level according to the comprehensive risk score and output standardized early warning information; the four levels of early warning include Level 1 alert warning, Level 2 general warning, Level 3 severe warning, and Level 4 emergency warning. The early warning information of different levels simultaneously carries the risk type, the parameters exceeding the limit, the duration of the risk, the location of the on-site area, and the recommended handling measures.

[0015] S8. Early warning closed-loop self-learning correction steps: Feed back all the graded early warning output records, on-site manual risk verification results, equipment linkage handling results, and post-accident review data to the dynamic threshold update module, iteratively optimize the scenario weight coefficient, trend adjustment coefficient, and sensitivity coefficient of each parameter, and synchronously update the initial threshold model and risk judgment coupling rules to form a continuously self-optimizing closed-loop early warning mechanism.

[0016] Preferably, in the S5 dynamic threshold adaptive update step, the dynamic threshold adopts a multi-factor joint update calculation formula:

[0017] X Dynamics = X Basic + T Trend Correction - F Volatility Suppression

[0018] X Scene Correction = SX Dynamics

[0019] In the formula: Xdynamic is the initial dynamic threshold of a single parameter, Xbasic is the basic threshold of the parameter in the scenario, Ttrend correction is the correction amount for the parameter change trend, Ffluctuation suppression is the correction amount for suppressing short-term sudden abnormal fluctuations; Xscenario correction is the final dynamic threshold after scenario weighting, S is the current scenario weight factor, and is the scenario sensitivity coefficient of the corresponding monitoring parameter.

[0020] Among them, the sensitivity coefficients of dust, carbon monoxide, methane, and wind speed parameters are the highest in tunnel construction scenarios; the sensitivity coefficients of wind speed and illuminance parameters are the highest in open-air high-altitude construction scenarios; and the sensitivity coefficients of smoke, oxygen, and toxic gas parameters are the highest in underground space construction scenarios.

[0021] Preferably, in the S5 dynamic threshold adaptive update step, the sliding window statistical correction includes two statistical models: the mean standard deviation method and the quantile method.

[0022] The dynamic threshold calculation formula of the mean standard deviation method is: dynamic upper limit threshold = +k1, dynamic lower limit threshold = -k2, where is the mean of effective data within the sliding window, is the standard deviation of the window data, and k1 and k2 are scene correlation adjustment coefficients.

[0023] The quantile method extracts the high and low quantiles of historical monitoring data as threshold benchmarks, which is specifically adapted to non-normally distributed monitoring data and makes up for the adaptation defects of the mean-standard deviation method.

[0024] Preferably, in the S6 multi-parameter coupled risk calculation step, a logic for classifying and determining the duration of exceeding the limit is set:

[0025] When a parameter momentarily exceeds the dynamic threshold, but the duration of the exceedance is less than the first duration threshold t1, only a minor anomaly is marked, and the warning level is not increased.

[0026] When the duration of exceeding the limit is between t1 and the second duration threshold t2, the warning level is upgraded to level one.

[0027] When the duration of exceeding the limit exceeds t2, the alert level is directly upgraded to Level 3 or higher.

[0028] When the rate of change of the monitored parameter per unit time exceeds the preset rate of change threshold, the time threshold is shortened directly to raise the warning level in advance and achieve early risk prediction.

[0029] Preferably, in the S7 graded early warning output step, the four-level early warning corresponds to specific triggering conditions and on-site handling logic:

[0030] Level 1 alert: The triggering conditions are that the parameter is close to the dynamic threshold but does not exceed the limit, a single parameter fluctuates slightly for a short time, or two or more parameters are slightly coupled abnormally; the handling method is to only pop up a notification on the monitoring platform, push a reminder message to the on-duty personnel, and record it as an on-site observation event;

[0031] Level 2 General Warning: The triggering conditions are a single parameter stably exceeding the limit, the exceeding time reaching t1, and multiple parameters simultaneously showing an upward trend in risk; the handling method is to trigger local on-site audio and visual prompts, push warning information to the construction supervisor, and prompt the site to strengthen manual inspections and optimize and adjust the operation methods;

[0032] Level 3 Severe Warning: The triggering conditions are that key safety parameters continue to exceed the limits, multiple parameters are coupled to determine medium to high risk, and the comprehensive risk score reaches the severe threshold; the handling methods are to activate the on-site high-decibel audible and visual alarm, broadcast the risk in the construction area, automatically link ventilation, sprinkler and smoke exhaust equipment, restrict personnel from entering the risk area and report to the project's overall management platform at the same time.

[0033] Level 4 Emergency Warning: The triggering conditions are serious exceedance of high-risk safety parameters, identification of major risks such as fire, explosion, oxygen deficiency, large-area water accumulation, and deformation of the building envelope; the response is to activate the highest level of audible and visual alarms across the entire area, coordinate the cutting off of non-essential power supplies in the area, turn on emergency lighting, broadcast personnel evacuation, push emergency alarms to all project management personnel and safety officers, and force the push of work stoppage and evacuation orders.

[0034] Preferably, the S7 graded early warning output step further includes a multi-level linkage control process, and the linkage equipment includes audible and visual alarms, fan exhaust devices, fog cannon spray devices, smoke exhaust equipment, area power outage devices, emergency lighting, area access control gates, and surveillance cameras.

[0035] The equipment linkage execution follows a hierarchical execution logic of first providing information prompts, then automatic intervention, and finally mandatory control, taking into account both construction safety and operational continuity. When any level of warning is triggered, the camera is automatically turned to the risk area, and monitoring data before and after the warning is collected in a preset time window, which is then uploaded to the platform for archiving and risk tracing analysis.

[0036] Preferably, the S1 multi-parameter real-time acquisition step adopts an adaptive mechanism of layered sampling and layered uploading;

[0037] The data sampling period is divided into three modes: high frequency, medium frequency, and event trigger. Dust, combustible and toxic gases, smoke, and wind speed are sampled at a fixed high frequency period. Temperature, humidity, illuminance, and water accumulation are sampled at a fixed medium frequency period. Noise and vibration are sampled at a high frequency during high-risk operation periods and the sampling frequency is automatically reduced during stable operation periods.

[0038] Adaptive switching of data upload cycle: Under normal operating conditions, edge nodes adopt a batch delayed upload mode to reduce device power consumption and network load; when monitoring data triggers abnormal or early warning conditions, the fixed upload cycle limit is broken, and event data and on-site status data are uploaded in real time; when the network is interrupted, the monitoring node locally caches the original monitoring data, and automatically re-uploads all cached data after the network is restored to ensure data integrity.

[0039] Preferably, in the S2 data preprocessing step, the abnormal data removal includes five judgment criteria. If any one of these criteria is met, the data is judged as invalid abnormal data and is removed, downgraded, or replaced:

[0040] First, the monitored value exceeds the sensor's preset physical range boundary; second, the value change amplitude between adjacent sampling times exceeds the preset allowable change threshold; third, the value of a single sampling point deviates from the continuous and stable data sequence before and after; fourth, the monitoring trend of multiple sensors in the same area is consistent, with only a single node's data deviating significantly; fifth, the sensor continuously outputs fixed values, no changing values, or equipment fault error codes.

[0041] Meanwhile, a data credibility score is calculated for each sampled data. The scoring dimensions include data integrity, temporal stability, consistency with sensors in the same area, and sensor equipment operating status. Data with a credibility score lower than the preset value will not participate in the dynamic threshold update calculation and will only be used as on-site auxiliary reference data.

[0042] Preferably, each construction scenario label corresponds to a differentiated initial threshold model and risk control logic:

[0043] In tunnel construction scenarios, the focus should be on tightening the thresholds for dust, carbon monoxide, methane, and minimum ventilation velocity, and prioritizing the identification of combined risks such as insufficient ventilation, gas accumulation, and dust diffusion.

[0044] Deep foundation pit construction scenario: refine the threshold for water accumulation depth classification, tighten the vibration threshold around the retaining structure, and coordinate drainage equipment to control the risk of water accumulation and slope instability;

[0045] Open-air high-altitude construction scenarios: Strengthen the control of wind speed and illuminance thresholds, automatically tighten early warning standards under strong wind and low illuminance conditions, and restrict hoisting and high-altitude high-risk operations;

[0046] Underground space construction scenario: Significantly reduce the warning thresholds for smoke, combustible gas, and oxygen deficiency, and improve the response priority for fire, poisoning, and oxygen deficiency risks.

[0047] Preferably, the iterative optimization logic of the S8 early warning closed-loop self-learning correction step includes:

[0048] In response to historical false alarm events, the scene weights and parameter sensitivity coefficients were optimized, and the short-term fluctuation suppression correction was increased to reduce invalid warnings caused by instantaneous environmental disturbances.

[0049] For historical accident underreporting samples, the dynamic threshold of the corresponding risk parameters is lowered and the threshold for judging the time of exceeding the limit is shortened to improve the sensitivity of high-risk identification;

[0050] Based on the real risk events verified by manual verification, the multi-parameter coupled risk judgment rules are iteratively optimized, and the initial threshold model benchmark values ​​under various construction scenarios are updated to achieve continuous adaptive optimization of dynamic thresholds and risk judgment rules according to on-site working conditions.

[0051] This invention provides a multi-parameter dynamic threshold adaptive early warning method for construction environment. It has the following beneficial effects:

[0052] 1. Compared to existing fixed threshold early warning technologies, this invention effectively solves the industry pain points of high false alarm rates and weak anti-interference capabilities in traditional construction environment early warning methods. This invention achieves accurate removal of abnormal data and effective compensation for missing data through a multi-dimensional data preprocessing mechanism. Combined with short-term fluctuation suppression and correction logic, it filters out invalid data interference caused by transient disturbances in construction site equipment and instantaneous environmental changes. Furthermore, relying on a joint judgment mechanism based on parameter change rate and duration of exceeding limits, it abandons the crude judgment logic of alarming based solely on a single instantaneous exceedance, greatly avoiding false alarms caused by fluctuations in construction conditions. This significantly improves the accuracy and stability of construction environment risk early warning, providing reliable data support for on-site safety management.

[0053] 2. This invention possesses a strong adaptive capability to complex construction scenarios, overcoming the technical bottlenecks of traditional early warning technologies, such as fixed thresholds and poor scenario versatility. This invention can automatically identify differentiated construction scenarios such as tunnels, deep foundation pits, underground spaces, and open-air high-altitude environments, and matches specific initial threshold models, parameter sensitivity coefficients, and risk control logic to different scenarios. Simultaneously, through scenario weight correction, historical data statistical correction, and trend prediction correction, it achieves dynamic adaptive iteration of thresholds, accurately adapting to safety assessment standards of different construction stages, operation types, meteorological environments, and regional working conditions. This effectively solves the problem that fixed thresholds cannot adapt to complex and ever-changing construction sites, comprehensively covering the safety early warning needs of various high-risk construction scenarios.

[0054] 3. This invention constructs a multi-parameter coupled risk assessment and closed-loop self-learning optimization system, realizing a technological upgrade of construction risk from single-point over-limit alarms to trend prediction and composite risk early warning. This invention breaks through the limitations of traditional single-parameter independent judgment, identifying hidden composite safety risks through a multi-parameter linkage coupling mechanism, and achieving early risk prediction by combining risk change trends, effectively reducing the risk of missed reports. Simultaneously, relying on feedback data from the entire process of early warning records, manual verification, and on-site handling, it continuously optimizes the threshold model and risk judgment rules. Combined with a tiered early warning and equipment linkage control mechanism, it forms a complete closed loop of data collection – risk assessment – ​​tiered early warning – automatic handling – model optimization, continuously improving the system's early warning capabilities, significantly enhancing the level of intelligent safety management at construction sites, and ensuring safe and controllable construction throughout the entire process. Attached Figure Description

[0055] Figure 1 This is an overall flowchart of the present invention;

[0056] Figure 2 This is a flowchart of the core early warning logic of the present invention. Detailed Implementation

[0057] The present invention provides embodiments, which are further described in detail below with reference to specific examples. However, the implementation of the present invention is not limited thereto. The present invention aims to protect a multi-parameter dynamic threshold adaptive early warning method for construction environments. Without departing from the core principles of the technical solution of the present invention, those skilled in the art can make equivalent substitutions and optimizations for parameters, scenarios, and equipment types, all of which fall within the scope of protection of the present invention. The present invention achieves accurate, adaptive, and proactive intelligent early warning of environmental risks in complex construction scenarios such as building construction, tunnel construction, underground spaces, deep foundation pits, and open-air high-altitude environments through a complete technical chain including real-time multi-parameter acquisition, data preprocessing, construction scenario identification, dynamic threshold adaptive updating, multi-parameter coupled risk calculation, hierarchical early warning output, and closed-loop self-learning correction.

[0058] Please see the appendix Figure 1 - Appendix Figure 2The overall implementation process of this invention is as follows: First, multi-source sensor monitoring nodes are deployed in layers according to the type of construction area, using a method of denser deployment in key areas and uniform deployment in regular areas to complete real-time collection of multiple parameters such as dust, temperature and humidity, wind speed, vibration, toxic and combustible gases, smoke, illuminance, and water accumulation. Then, the raw data is synchronized in time, outlier removal, noise filtering, missing value compensation, and sensor consistency verification to obtain a highly reliable and effective data sequence. Further, the construction scene is accurately identified by combining working conditions, time periods, and environmental characteristics, and the initial threshold model of the corresponding scene is retrieved. Relying on sliding window statistics, scene weights, trend prediction, fluctuation suppression, and historical learning multiple correction logics, the real-time dynamic thresholds of each monitoring parameter are dynamically updated. The comprehensive risk level is calculated by combining the parameter over-limit amplitude, rate of change, duration, and multi-parameter coupling correlation, and a four-level graded early warning signal is output. The system is then linked with on-site sound and light, ventilation, sprinkler, power outage, broadcast, video monitoring, and other equipment to complete automated handling. Finally, the early warning results, manual verification, and on-site handling data are fed back to the threshold model to achieve iterative optimization of algorithm rules and threshold parameters, forming a closed-loop self-learning mechanism.

[0059] Example 1: Implementation of a Tunnel Construction Scenario

[0060] For high-risk construction scenarios involving enclosed tunnels with poor ventilation, easy gas accumulation, and difficulty in dust dispersion, this invention prioritizes the deployment of dust sensors, wind speed sensors, carbon monoxide sensors, methane sensors, temperature and humidity sensors, and smoke sensors. Monitoring nodes are deployed in sections along the tunnel face, construction work section, and tunnel exit. A denser deployment method is used in high-risk areas at the tunnel face to completely eliminate monitoring blind spots. Upon system startup, it automatically identifies the current working condition as a tunnel construction scenario, loads a tunnel-specific initial threshold model, and sets stringent basic thresholds for dust concentration, combustible and toxic gas concentration, and minimum ventilation wind speed, while correspondingly increasing the scenario sensitivity coefficients for gas, dust, and wind speed parameters. During construction, the system frequently collects various environmental parameters and performs data preprocessing, eliminating abnormal interference data caused by short-term sensor vibrations and equipment start-up and shutdown. When the system detects a continuous increase in dust concentration inside the tunnel while the ventilation speed shows a continuous downward trend, it automatically triggers a multi-parameter coupled risk assessment to identify the combined risks of insufficient tunnel ventilation and dust accumulation. It then dynamically lowers the wind speed warning threshold and increases the dust risk weight in real time, tightening the warning assessment criteria in advance. If the duration of dust exceeding the limit reaches the Level 1 assessment threshold, the system outputs a Level 2 general warning, triggering on-site audible and visual alerts and reminding management personnel to strengthen ventilation inspections. If dust continues to exceed the standard and is accompanied by abnormal increases in carbon monoxide and methane concentrations or a slight increase in smoke parameters, the system determines it as a high-risk combined risk and directly upgrades it to a Level 3 severe warning. It automatically activates the tunnel exhaust equipment and dust suppression spray system, while simultaneously broadcasting risk warnings on-site and restricting personnel from entering high-risk areas at the tunnel face. In extreme cases where gas concentrations severely exceed limits and smoke rapidly increases, the system immediately triggers a Level 4 emergency warning, executing regional power outages, emergency lighting activation, and a full evacuation broadcast, achieving rapid risk management. In this embodiment, the system can dynamically adjust the threshold based on the tunnel construction progress, ventilation equipment operation status, and number of workers inside the tunnel. At the same time, it records each early warning event and the result of manual handling, continuously optimizes the parameter sensitivity and coupling judgment rules in the tunnel scenario, and continuously reduces the probability of false alarms and missed alarms.

[0061] Example 2: Implementation of Deep Foundation Pit Construction Scenario

[0062] Deep foundation pit construction is characterized by its low-lying location prone to water accumulation, susceptibility to vibration affecting the retaining structure, and a surge in environmental risks during rainy weather. This embodiment deploys water accumulation sensors, vibration sensors, temperature and humidity sensors, noise sensors, and gas monitoring sensors at the bottom of the foundation pit, its slopes, surrounding retaining structures, and at the pit entrances and exits. After automatically identifying the deep foundation pit construction scenario, the system loads a foundation pit-specific initial threshold model, focusing on tightening the warning thresholds for water accumulation depth and retaining structure vibration. The system dynamically adjusts the risk assessment criteria based on rainfall, soil moisture, and foundation pit drainage conditions. During daily construction, the system continuously monitors the foundation pit's water accumulation status and surrounding vibration data, filtering out abnormal and distorted data from single nodes through multi-node data consistency verification. When encountering rainfall, if the system detects a gradual increase in foundation pit water depth and persistently high ambient humidity, it automatically increases the water accumulation risk weight and dynamically adjusts the water accumulation warning threshold. If continuous vibration caused by construction machinery operations around the foundation pit is simultaneously detected, the system uses multi-parameter coupled judgment logic to identify the combined risk of slope instability caused by the superimposed vibration disturbance from foundation pit water infiltration. Based on the duration of excessive water accumulation and the rate of vibration change, the system outputs general and severe warning levels in stages, triggering on-site inspection reminders and automatically activating the foundation pit drainage equipment. When high-risk conditions occur, such as rapid water spread and a sharp increase in vibration amplitude, the system triggers a level four emergency warning, sends a work stoppage order, prohibits all construction work within the foundation pit, and reminds personnel to evacuate quickly to prevent foundation pit collapse and flooding accidents. Furthermore, this embodiment can iteratively optimize the threshold model based on historical rainy season water accumulation data and vibration risk events to improve the ability to predict construction risks in foundation pits during the rainy season.

[0063] Example 3: Implementation of an Open-Air High-Altitude Construction Scenario

[0064] Open-air high-altitude construction is greatly affected by meteorological conditions; changes in wind speed, sunlight, and temperature directly determine the safety of hoisting and high-altitude operations. This embodiment strategically deploys wind speed sensors, illuminance sensors, temperature sensors, and noise sensors around the high-altitude work platform, hoisting area, and construction tower crane. After identifying the open-air high-altitude construction scenario, the system prioritizes and enhances the monitoring weight and sensitivity of wind speed, illuminance, and ambient temperature parameters to adapt to the variable weather conditions of open-air scenarios. The system employs a layered sampling mechanism, automatically increasing the sampling frequency during periods of strong winds, high temperatures, and nighttime construction to accurately capture sudden trends in environmental parameters. When the system detects that the on-site wind speed is gradually approaching the safe operating threshold or that the rate of wind speed change is consistently high, the system lowers the warning threshold in advance through a trend prediction and correction mechanism, achieving early risk warning. If there are short-term gust fluctuations, the system filters out instantaneous interference through duration determination and fluctuation suppression mechanisms, only recording anomalies without triggering high-level warnings, effectively avoiding false alarms. During nighttime construction, the system dynamically adjusts warning standards based on illuminance parameters. When illuminance remains below the safe operating threshold, it outputs a warning and triggers supplemental lighting on-site. In hot weather, the system dynamically increases the heat risk level based on ambient temperature and work hours, reminding workers to take precautions against heatstroke and stagger work hours. In severe conditions such as persistent strong winds or low illuminance before heavy rain, the system outputs a high-level warning and pushes control recommendations to suspend hoisting and stop high-altitude operations, avoiding the risks of falls from heights and hoisting overturning. This embodiment can continuously optimize the dynamic thresholds for outdoor scenarios based on seasonal weather changes and historical strong wind warning data, significantly improving weather adaptability and warning accuracy.

[0065] Example 4: Implementation of Underground Space Construction Scenarios

[0066] Underground, enclosed spaces are characterized by poor air circulation, low oxygen levels, rapid fire spread, and the accumulation of toxic gases, making them extremely high-risk construction scenarios. This embodiment comprehensively deploys smoke sensors, oxygen sensors, combustible and toxic gas sensors, temperature and humidity sensors, and water accumulation monitoring equipment. After identifying the underground space scenario, the system defaults to stringent safety warning standards, significantly lowering the warning thresholds for smoke, gas, and oxygen deficiency, and increasing the response priority of these parameters. During construction, the system monitors the gas concentration and air quality in the space in real time, identifying hidden risks through multi-parameter coupled logic: when a slight increase in smoke is accompanied by a slight decrease in oxygen, a suspected fire risk can be identified and a warning issued in advance, detecting potential hazards earlier than traditional single-threshold alarms; when a slow leak of toxic gas is detected and its concentration continues to rise, the system dynamically tightens the threshold based on the enclosed space characteristics, activating ventilation equipment in advance. In the event of high-risk situations such as severely excessive gas levels, rapid smoke spread, or a sharp drop in oxygen concentration, the system immediately triggers an emergency warning, linking power outages, smoke extraction, emergency lighting, and personnel evacuation systems to comprehensively ensure the safety of underground space construction. At the same time, relying on historical fire and gas leak simulation data, the judgment rules are continuously iterated and optimized to achieve the ultimate prevention and control of risks in underground confined spaces.

[0067] In the aforementioned embodiments, this invention uniformly adopts a data credibility scoring mechanism, a multi-factor dynamic threshold update formula, a joint judgment logic of duration and rate of change, and a hierarchical linkage handling strategy. Different scenarios achieve differentiated adaptation only through scenario weight factors, parameter sensitivity coefficients, and basic threshold models. The overall technical solution is universal, stable, and portable. Simultaneously, through a closed-loop self-learning mechanism for early warning feedback, the threshold models, coupling rules, and correction coefficients for each scenario can be continuously iterated and optimized according to the actual on-site working conditions. This results in a continuous reduction in false alarm and false negative rates during long-term system operation, and a continuous improvement in risk prediction capabilities and scenario adaptability. It effectively solves the shortcomings of traditional fixed threshold early warning technologies, such as poor adaptability to working conditions, single judgment, and lack of self-optimization capabilities, comprehensively meeting the intelligent and refined safety early warning and control needs of various complex construction sites.

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

Claims

1. A multi-parameter dynamic threshold adaptive early warning method for construction environment, characterized in that, Includes the following steps: S1. Multi-parameter real-time acquisition steps: Multiple sets of monitoring nodes are deployed in layers at the construction site. Various monitoring parameters of the construction environment are collected synchronously through each monitoring node. The monitoring parameters include at least two of the following: air quality parameters, physical environment parameters, safety status parameters, and structural operation parameters. The monitoring nodes adopt a combination of regional basic deployment and high-risk area densification deployment, including fixed, pole-mounted, and mobile deployment forms. In tunnel and underground long-distance construction scenarios, a series full-coverage deployment along the line is adopted to eliminate monitoring blind spots. S2. Data Preprocessing Steps: Perform time synchronization calibration, physical boundary outlier removal, abrupt isolated outlier data screening, multi-sensor consistency verification, and sliding window noise reduction filtering on the collected raw monitoring data. Simultaneously, perform missing value compensation on short-term missing data and output a valid monitoring data sequence that meets the reliability score. Among them, the missing value compensation methods include forward filling, window mean compensation, neighbor node interpolation, and trend extrapolation compensation. During sensor consistency verification, if the data of a single node deviates from the overall trend of the region, the sensor is judged to be faulty. The data of the faulty node is downweighted, removed, or replaced with redundant node data in the same region. S3. Construction Scene Status Recognition Steps: Based on the construction area type, work type, spatial environment type, time period of the day, meteorological environment, and historical working condition data, feature matching and recognition are performed to output a unique corresponding scene label. The scene label includes at least tunnel construction scene, deep foundation pit construction scene, underground space construction scene, open-air high-altitude construction scene, and high-noise processing construction scene. S4. Initial threshold model loading step: Retrieve the preset initial threshold model of the corresponding scene according to the identified scene label. The initial threshold model configures the basic upper limit threshold, basic lower limit threshold, parameter change rate threshold, and over-limit duration threshold for each type of monitoring parameter. The initial threshold is jointly set based on national construction safety standards, industry specifications, project historical statistical data, and on-site engineering experience. S5. Dynamic Threshold Adaptive Update Step: Integrate four dimensions of data: effective monitoring data, full-cycle historical monitoring data, current scene labels, and parameter change trends. Perform multi-factor weighted correction on the initial threshold model to generate a dynamic threshold set adapted to the current working conditions. The dynamic threshold set includes single-parameter dynamic thresholds, multi-parameter composite thresholds, and risk trend thresholds. The dynamic threshold update synchronously executes five types of correction logic: sliding window statistical correction, scene weight correction, trend prediction correction, short-term fluctuation anomaly suppression correction, and historical early warning accident sample iterative correction. S6. Multi-parameter coupling risk calculation steps: Compare real-time effective monitoring data with dynamic threshold sets item by item, and combine four dimensions of indicators: single parameter exceedance magnitude, real-time parameter change rate, exceedance duration, and multi-parameter linkage coupling relationship to calculate a weighted comprehensive risk score and match the corresponding risk level; among them, the multi-parameter coupling judgment rules include: when dust concentration increases and wind speed decreases, determine the risk of insufficient ventilation; when temperature and humidity increase and combustible gas exceed the standard, determine the risk of heat and moisture leakage; when noise continuously exceeds the standard and vibration is abnormal, determine the risk of equipment structure; when smoke concentration increases and illuminance drops sharply, determine the risk of fire visibility. S7. Tiered early warning output steps: Pre-divide the four early warning levels, match the corresponding early warning level according to the comprehensive risk score and output standardized early warning information; the four levels of early warning include Level 1 alert warning, Level 2 general warning, Level 3 severe warning, and Level 4 emergency warning. The early warning information of different levels simultaneously carries the risk type, the parameters exceeding the limit, the duration of the risk, the location of the on-site area, and the recommended handling measures. S8. Early warning closed-loop self-learning correction steps: Feed back all the graded early warning output records, on-site manual risk verification results, equipment linkage handling results, and post-accident review data to the dynamic threshold update module, iteratively optimize the scenario weight coefficient, trend adjustment coefficient, and sensitivity coefficient of each parameter, and synchronously update the initial threshold model and risk judgment coupling rules to form a continuously self-optimizing closed-loop early warning mechanism.

2. The adaptive early warning method for multi-parameter dynamic thresholds in construction environment according to claim 1, characterized in that, In the S5 dynamic threshold adaptive update step, the dynamic threshold adopts a multi-factor joint update calculation formula: X Dynamics = X Basic + T Trend Correction - F Volatility Suppression X Scene Correction = SX Dynamics In the formula: Xdynamic is the initial dynamic threshold of a single parameter, Xbasic is the basic threshold of the parameter in the scenario, Ttrend correction is the correction amount for the parameter change trend, Ffluctuation suppression is the correction amount for suppressing short-term sudden abnormal fluctuations; Xscenario correction is the final dynamic threshold after scenario weighting, S is the current scenario weight factor, and is the scenario sensitivity coefficient of the corresponding monitoring parameter. Among them, the sensitivity coefficients of dust, carbon monoxide, methane, and wind speed parameters are the highest in tunnel construction scenarios; the sensitivity coefficients of wind speed and illuminance parameters are the highest in open-air high-altitude construction scenarios; and the sensitivity coefficients of smoke, oxygen, and toxic gas parameters are the highest in underground space construction scenarios.

3. The adaptive early warning method for multi-parameter dynamic thresholds in construction environment according to claim 1, characterized in that, In the S5 dynamic threshold adaptive update step, the sliding window statistical correction includes two statistical models: the mean standard deviation method and the quantile method. The dynamic threshold calculation formula of the mean standard deviation method is: dynamic upper limit threshold = +k1, dynamic lower limit threshold = -k2, where is the mean of effective data within the sliding window, is the standard deviation of the window data, and k1 and k2 are scene correlation adjustment coefficients. The quantile method extracts the high and low quantiles of historical monitoring data as threshold benchmarks, which is specifically adapted to non-normally distributed monitoring data and makes up for the adaptation defects of the mean-standard deviation method.

4. The adaptive early warning method for multi-parameter dynamic thresholds in construction environment according to claim 1, characterized in that, In the S6 multi-parameter coupled risk calculation step, a logic for classifying and determining the duration of exceeding limits is set: When a parameter momentarily exceeds the dynamic threshold, but the duration of the exceedance is less than the first duration threshold t1, only a minor anomaly is marked, and the warning level is not increased. When the duration of exceeding the limit is between t1 and the second duration threshold t2, the warning level is upgraded to level one. When the duration of exceeding the limit exceeds t2, the alert level is directly upgraded to Level 3 or higher. When the rate of change of the monitored parameter per unit time exceeds the preset rate of change threshold, the time threshold is shortened directly to raise the warning level in advance and achieve early risk prediction.

5. The adaptive early warning method for multi-parameter dynamic thresholds in construction environment according to claim 1, characterized in that, In the S7 graded early warning output steps, the fourth-level early warning corresponds to specific triggering conditions and on-site handling logic: Level 1 warning: The triggering conditions are that the parameter is close to the dynamic threshold but does not exceed the limit, a single parameter fluctuates slightly for a short time, or two or more parameters are slightly coupled abnormally. The handling method is to only issue a pop-up notification on the monitoring platform and push a reminder message to the on-duty personnel, and record it as an on-site observation event; Level 2 General Warning: The triggering conditions are a single parameter stably exceeding the limit, the exceeding time reaching t1, and multiple parameters simultaneously showing an upward trend in risk; The response method involves triggering local audio-visual alerts on-site, sending early warning information to the construction supervisor, and prompting them to strengthen on-site manual inspections and optimize and adjust work methods. Level 3 Severe Warning: The triggering conditions are that key safety parameters continue to exceed the limits, multiple parameters are coupled to determine medium to high risk, and the comprehensive risk score reaches the severe threshold; the handling methods are to activate the on-site high-decibel audible and visual alarm, broadcast the risk in the construction area, automatically link ventilation, sprinkler and smoke exhaust equipment, restrict personnel from entering the risk area and report to the project's overall management platform at the same time. Level 4 Emergency Warning: The triggering conditions are serious exceedance of high-risk safety parameters, identification of major risks such as fire, explosion, oxygen deficiency, large-area water accumulation, and deformation of the building envelope; the response is to activate the highest level of audible and visual alarms across the entire area, coordinate the cutting off of non-essential power supplies in the area, turn on emergency lighting, broadcast personnel evacuation, push emergency alarms to all project management personnel and safety officers, and force the push of work stoppage and evacuation orders.

6. The adaptive early warning method for multi-parameter dynamic thresholds in construction environment according to claim 1, characterized in that, The S7 graded early warning output step also includes a multi-level linkage control process, and the linkage equipment includes audible and visual alarms, fan exhaust devices, fog cannon spray devices, smoke exhaust equipment, area power cut-off devices, emergency lighting, area access control gates, and surveillance cameras. The equipment linkage execution follows a hierarchical execution logic of first providing information prompts, then automatic intervention, and finally mandatory control, taking into account both construction safety and operational continuity. When any level of warning is triggered, the camera is automatically turned to the risk area, and monitoring data before and after the warning is collected in a preset time window, which is then uploaded to the platform for archiving and risk tracing analysis.

7. The adaptive early warning method for multi-parameter dynamic thresholds in construction environment according to claim 1, characterized in that, The S1 multi-parameter real-time acquisition step adopts an adaptive mechanism of layered sampling and layered uploading; The data sampling period is divided into three modes: high frequency, medium frequency, and event trigger. Dust, combustible and toxic gases, smoke, and wind speed are sampled at a fixed high frequency period. Temperature, humidity, illuminance, and water accumulation are sampled at a fixed medium frequency period. Noise and vibration are sampled at a high frequency during high-risk operation periods and the sampling frequency is automatically reduced during stable operation periods. Adaptive switching of data upload cycle: Under normal operating conditions, edge nodes adopt batch delayed upload mode to reduce device power consumption and network load; When monitoring data triggers abnormal or early warning conditions, the fixed upload cycle limit is broken, and event data and on-site status data are uploaded in real time. When the network is interrupted, the monitoring node caches the original monitoring data locally, and automatically re-uploads all cached data after the network is restored to ensure data integrity.

8. The adaptive early warning method for multi-parameter dynamic thresholds in construction environment according to claim 1, characterized in that, In the S2 data preprocessing step, the outlier data removal includes five criteria. If any one of these criteria is met, the data is determined to be invalid outlier and will be removed, downweighted, or replaced. First, the monitored value exceeds the sensor's preset physical range boundary; second, the value change amplitude between adjacent sampling times exceeds the preset allowable change threshold; third, the value of a single sampling point deviates from the continuous and stable data sequence before and after; fourth, the monitoring trend of multiple sensors in the same area is consistent, with only a single node's data deviating significantly; fifth, the sensor continuously outputs fixed values, no changing values, or equipment fault error codes. Meanwhile, a data credibility score is calculated for each sampled data. The scoring dimensions include data integrity, temporal stability, consistency with sensors in the same area, and sensor equipment operating status. Data with a credibility score lower than the preset value will not participate in the dynamic threshold update calculation and will only be used as on-site auxiliary reference data.

9. The adaptive early warning method for multi-parameter dynamic thresholds in construction environment according to claim 1, characterized in that, Differentiated initial threshold models and risk management logic corresponding to each construction scenario label: In tunnel construction scenarios, the focus should be on tightening the thresholds for dust, carbon monoxide, methane, and minimum ventilation velocity, and prioritizing the identification of combined risks such as insufficient ventilation, gas accumulation, and dust diffusion. Deep foundation pit construction scenario: refine the threshold for water accumulation depth classification, tighten the vibration threshold around the retaining structure, and coordinate drainage equipment to control the risk of water accumulation and slope instability; Open-air high-altitude construction scenarios: Strengthen the control of wind speed and illuminance thresholds, automatically tighten early warning standards under strong wind and low illuminance conditions, and restrict hoisting and high-altitude high-risk operations; Underground space construction scenario: Significantly reduce the warning thresholds for smoke, combustible gas, and oxygen deficiency, and improve the response priority for fire, poisoning, and oxygen deficiency risks.

10. The adaptive early warning method for multi-parameter dynamic thresholds in construction environment according to claim 1, characterized in that, The iterative optimization logic of the S8 early warning closed-loop self-learning correction step includes: In response to historical false alarm events, the scene weights and parameter sensitivity coefficients were optimized, and the short-term fluctuation suppression correction was increased to reduce invalid warnings caused by instantaneous environmental disturbances. For historical accident underreporting samples, the dynamic threshold of the corresponding risk parameters is lowered and the threshold for judging the time of exceeding the limit is shortened to improve the sensitivity of high-risk identification; Based on the real risk events verified by manual verification, the multi-parameter coupled risk judgment rules are iteratively optimized, and the initial threshold model benchmark values ​​under various construction scenarios are updated to achieve continuous adaptive optimization of dynamic thresholds and risk judgment rules according to on-site working conditions.

Images