Construction site environment monitoring method, system and device based on multi-source fusion and medium

Abstract

This application relates to a method, system, equipment, and medium for construction site environmental monitoring based on multi-source fusion. The method improves data reliability and consistency at the source by spatiotemporally aligning multi-source sensor data and innovatively calculating a real-time data quality factor to guide the weighted fusion of neighboring data. Furthermore, it uses physical constraints to repair missing values ​​and combines spatial interpolation to generate a refined environmental data cube covering the entire area, achieving data enhancement from discrete points to a continuous field. Finally, it extracts multi-dimensional features based on this cube and constructs a dynamic benchmark model combining construction stages and weather conditions for intelligent comparison and early warning. This fundamentally solves the problems of low data quality, information silos, and delayed early warning, achieving a comprehensive improvement in the reliability, panoramic perception capability, and forward-looking early warning of construction site environmental monitoring.

Description

Technical Field

[0001] This invention belongs to the field of data processing, and in particular relates to a method, system, equipment and medium for monitoring construction site environment based on multi-source fusion. Background Technology

[0002] In the current construction of smart construction sites, environmental monitoring is a crucial link in ensuring construction safety and green operation. Traditional methods mainly rely on regular manual inspections and the deployment of a small number of single-function sensors. This approach has significant limitations: data collection is discrete and sparse in both time and space, failing to form a continuous and comprehensive environmental situation profile. A more prominent problem is that with the application of IoT technology, construction sites often deploy sensors from different manufacturers and of different types (such as temperature, humidity, dust, and noise sensors). The data generated by these devices differ in format, frequency, and communication protocols, forming isolated "data silos." This heterogeneity prevents effective correlation and comprehensive analysis of data on a unified spatiotemporal benchmark, constituting a serious "information island." Furthermore, raw sensor data is highly susceptible to complex interference at the construction site (such as mechanical vibration and electromagnetic noise) and the performance drift of the equipment itself, often containing a large amount of noise, outliers, and missing data. Directly using this data for analysis would severely affect the reliability of monitoring results and the accuracy of early warnings.

[0003] Although existing research has attempted to utilize IoT technology for data collection or introduce single data cleaning algorithms, most current solutions focus on specific localized aspects, lacking a systematic, end-to-end solution that integrates data collection, cleaning, and intelligent analysis. This results in current construction site environmental monitoring generally facing challenges such as low data quality, fragmented information, delayed early warnings, and high false alarm rates, making it difficult to meet the urgent needs of modern construction for refined and intelligent environmental management.

[0004] Therefore, the industry urgently needs an innovative method that can fundamentally integrate multi-source heterogeneous data, ensure the intrinsic quality of data, and achieve a seamless transition from data to intelligent decision-making. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, equipment, and medium for construction site environmental monitoring based on multi-source fusion to address the aforementioned technical problems.

[0006] Firstly, this application provides a construction site environmental monitoring method based on multi-source fusion, comprising:

[0007] S1. Obtain the raw data packets output by multiple environmental sensors of different types deployed at different locations on the construction site. Based on the sensor identifiers and timestamps in the raw data packets, perform spatiotemporal alignment processing on the sensor data corresponding to all environmental sensors to generate a spatiotemporally aligned multi-sensor time series matrix.

[0008] S2. Based on the local statistical characteristics and factory accuracy indicators of the data from each sensor in the multi-sensor time series matrix, calculate the real-time data quality factor of each environmental sensor at each moment. Based on the real-time data quality factor, perform weighted fusion on the sensor data of a group of environmental sensors monitoring the same environmental parameter and located in close proximity, generating fused environmental parameter sequences for several spatial regions and their corresponding comprehensive confidence sequences. Among them, the real-time data quality factor characterizes the comprehensive reliability of the current readings of the environmental sensors; the comprehensive confidence sequence is the mean sequence of the real-time data quality factors of all sensors in the environmental sensor cluster.

[0009] S3. Based on the physical constraint relationship between various environmental parameters in the sensor data, perform physical-guided sequence prediction and repair on the missing segments in the fused environmental parameter sequence. Based on the geographic coordinates and comprehensive confidence sequence of each spatial region, use an interpolation algorithm to convert the discrete fused environmental parameter sequence into a regular gridded environmental data cube covering the entire construction site.

[0010] S4. Extract multidimensional environmental state feature vectors from the environmental data cube, construct a dynamic benchmark model based on historical environmental monitoring data of the construction site that is related to the construction stage and external weather conditions, and calculate the deviation of the multidimensional environmental state feature vectors from the dynamic benchmark model at the current moment; when the deviation exceeds the threshold, trigger an environmental anomaly warning.

[0011] Secondly, this application also provides a construction site environmental monitoring system based on multi-source fusion, used to implement the method described in the first aspect, the system comprising:

[0012] The spatiotemporal data synchronization module is used to acquire raw data packets output by multiple environmental sensors of different types deployed at different locations on the construction site. Based on the sensor identifiers and timestamps in the raw data packets, the module performs spatiotemporal alignment processing on the sensor data corresponding to all environmental sensors to generate a spatiotemporally aligned multi-sensor time series matrix.

[0013] The adaptive data fusion module is used to calculate the real-time data quality factor of each environmental sensor at each moment based on the local statistical characteristics and factory accuracy indicators of the data from each sensor in the multi-sensor time series matrix. Based on the real-time data quality factor, the sensor data of a group of environmental sensors monitoring the same environmental parameter and located in close proximity are weighted and fused to generate fused environmental parameter sequences for several spatial regions and their corresponding comprehensive confidence sequences. Among them, the real-time data quality factor represents the comprehensive reliability of the current readings of the environmental sensors; the comprehensive confidence sequence is the mean sequence of the real-time data quality factors of all sensors in the environmental sensor cluster.

[0014] The data reconstruction module is used to perform physically guided sequence prediction and repair on missing segments in the fused environmental parameter sequence based on the physical constraint relationship between various environmental parameters in the sensor data. Based on the geographic coordinates and comprehensive confidence sequence of each spatial region, it uses an interpolation algorithm to convert the discrete fused environmental parameter sequence into a regular gridded environmental data cube covering the entire construction site.

[0015] The dynamic benchmark early warning module is used to extract multi-dimensional environmental state feature vectors from the environmental data cube, construct a dynamic benchmark model based on historical environmental monitoring data of the construction site and related to the construction stage and external weather conditions, calculate the deviation of the multi-dimensional environmental state feature vectors from the dynamic benchmark model at the current moment; when the deviation exceeds the threshold, an environmental anomaly early warning is triggered.

[0016] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a construction site environmental monitoring method based on multi-source fusion as described in the first aspect.

[0017] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a site environment monitoring method based on multi-source fusion as described in the first aspect.

[0018] The aforementioned construction site environmental monitoring method, system, equipment, and medium based on multi-source fusion unifies the spatiotemporal benchmark of multi-source heterogeneous sensor data through spatiotemporal alignment processing. It then innovatively introduces and calculates a real-time data quality factor to dynamically assess the reliability of each sensor's data. Based on this factor, it weights and fuses neighboring sensor data to improve data reliability and generates a comprehensive confidence sequence characterizing the overall reliability of regional data. Next, it intelligently repairs the fused sequence using physical constraints between environmental parameters and generates a refined environmental data cube covering the entire region through spatial interpolation algorithms, combining geographic coordinates and the confidence sequence, thus transforming discrete point data into continuous surface information. Finally, it extracts multi-dimensional feature vectors from this data cube and constructs a dynamic benchmark model based on construction stages and weather conditions for comparative analysis, achieving intelligent early warning based on the deviation of the overall environmental pattern. This collectively achieves full-chain optimization of the construction site environment from bottom-level data acquisition and midstream fusion processing to upper-level intelligent analysis, significantly improving the quality and reliability of environmental monitoring data, realizing panoramic continuous perception of the construction site environment, and greatly enhancing the accuracy and foresight of anomaly early warning. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a construction site environmental monitoring method based on multi-source fusion provided by the present invention;

[0021] Figure 2 This is a schematic diagram of the process of generating a fusion environmental parameter sequence of several spatial regions and their corresponding comprehensive confidence sequence in an optional embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the structure of a construction site environmental monitoring system based on multi-source fusion, provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0024] refer to Figure 1The document presents a flowchart illustrating a multi-source fusion-based method for monitoring construction site environments, which includes the following steps:

[0025] S1. Obtain the raw data packets output by multiple environmental sensors of different types deployed at different locations on the construction site. Based on the sensor identifiers and timestamps in the raw data packets, perform spatiotemporal alignment processing on the sensor data corresponding to all environmental sensors to generate a spatiotemporally aligned multi-sensor time series matrix.

[0026] Specifically, constructing a network of environmental sensors for construction sites. ,in This refers to the total number of sensors, encompassing various types including temperature and humidity sensors, dust concentration sensors, and noise sensors. Each sensor... Corresponding unique sensor identifier This identifier is pre-stored in the system database and associated with the sensor's physical space coordinates. All sensors are equipped with a high-precision clock module to output timestamps in a uniform format. The timestamps are uniformly calibrated to Coordinated Universal Time (UTC) to eliminate clock drift errors between different devices.

[0027] Extract each sensor from the raw data packet timestamp Corresponding raw monitoring data values A multi-sensor time series matrix is ​​constructed based on a dual index of sensor identifiers and timestamps. The dimension of this matrix is ,in Given the total number of data collection time points, the matrix elements satisfy the formula:

[0028]

[0029] In the above formula, This is a sensor index, with a value range corresponding to all sensors in the sensor deployment set; This is a time index, with a value range corresponding to all data collection time points; The first element in the time series matrix Line 1 The elements of the column correspond to the sensors. timestamp Raw monitoring data values ​​collected The core principle of spatiotemporal alignment processing is to associate spatial locations with sensor identifiers and time nodes with unified timestamps, thereby eliminating the spatial heterogeneity of multi-source sensor data and the asynchronous sampling in the temporal dimension, ensuring the matrix... Any two elements in the same column represent monitoring data from different locations at the same time, and any two elements in the same row represent monitoring data from different times at the same location.

[0030] S2. Based on the local statistical characteristics and factory accuracy indicators of the data from each sensor in the multi-sensor time series matrix, calculate the real-time data quality factor of each environmental sensor at each moment. Based on the real-time data quality factor, perform weighted fusion on the sensor data of a group of environmental sensors that monitor the same environmental parameter and are spatially adjacent, generating a fused environmental parameter sequence for several spatial regions and its corresponding comprehensive confidence sequence. Among them, the real-time data quality factor characterizes the comprehensive reliability of the current reading of the environmental sensor; the comprehensive confidence sequence is the mean sequence of the real-time data quality factors of all sensors in the environmental sensor cluster.

[0031] Specifically, the real-time data quality factor characterizes the overall reliability of the current readings of environmental sensors. Its calculation requires consideration of both the local statistical characteristics of the sensor data and the factory accuracy specifications. The calculation formula is as follows:

[0032]

[0033] In the above formula, For sensors In time The real-time data quality factor has a value range of [value range missing]. The larger the value, the higher the reliability of the sensor data at the corresponding moment; For sensors In time The mean value of the data within the corresponding sliding window. The length of the sliding window is reasonably set according to the sensor sampling frequency. Its specific value needs to be adapted to the sampling characteristics of different types of sensors to ensure the validity of the statistical results. The standard deviation of the data within the sliding window is used to reflect the dispersion of the data within the window; The deviation between the current data value and the window mean reflects the deviation of the current data from a local stationary state. For sensors The maximum permissible error at the factory is determined by the technical manual provided by the sensor manufacturer and represents the upper limit of the sensor's measurement accuracy under standard operating conditions. For sensors The factory zero-point drift coefficient characterizes the output offset of the sensor when there is no input, and reflects the inherent error characteristics of the sensor itself. , , The weighting coefficient is used to balance the impact of different error factors on data quality assessment. Its specific value can be determined by regression analysis of historical monitoring data of the construction site to ensure the consistency between the quality factor calculation results and the reliability of the actual data.

[0034] When generating the fused environmental parameter sequence and its corresponding comprehensive confidence sequence, the sensor set is first grouped. The grouping rule is that the sensors monitor the same environmental parameter and are spatially adjacent. The spatial proximity determination condition is set by the spatial Euclidean distance between the sensors. The distance threshold needs to be reasonably adjusted in combination with the site scale and sensor deployment density to ensure that the sensors in the group can cover the same local spatial area and that the monitoring data have spatial correlation. After grouping, several sensor groups are obtained. Each sensor group corresponds to a specific spatial area within the site, and all sensors in the group monitor the same environmental parameter.

[0035] For each sensor group, calculate its time... Fusion environment parameter values The calculation formula is:

[0036]

[0037] In the above formula, For the first A group of sensors in time The fusion environment parameter values; For the first A group of sensors, Indicates sensor Belongs to the One sensor group; For sensors In time Real-time data quality factor; For sensors In time The original monitoring data values. This calculation method assigns weights to different sensor data through quality factors, giving higher weight to sensor data with higher reliability in the fusion result, thereby reducing the interference of low-quality data on the fusion result.

[0038] Simultaneously calculate the time for each sensor group Overall confidence level This parameter is the average of the real-time data quality factors of all sensors within the group, and the calculation formula is:

[0039]

[0040] In the above formula, For the first A group of sensors in time The overall confidence level; For the first The number of sensors in a sensor group; For the group of sensors In time The real-time data quality factor, based on continuous values ​​of the time index, can generate fusion environmental parameter sequences and comprehensive confidence sequences for each sensor group, providing high-quality basic data for subsequent data processing.

[0041] S3. Based on the physical constraint relationship between various environmental parameters in the sensor data, perform physical-guided sequence prediction and repair on the missing segments in the fused environmental parameter sequence. Based on the geographic coordinates and comprehensive confidence sequence of each spatial region, use an interpolation algorithm to convert the discrete fused environmental parameter sequence into a regular gridded environmental data cube covering the entire construction site.

[0042] Specifically, based on the physical constraints between various environmental parameters in the sensor data, physical-guided sequence prediction is used to repair missing segments in the fused environmental parameter sequence. First, missing segments in the fused environmental parameter sequence are identified, defined as those where the fused environmental parameter value is empty within a continuous time index interval. For repairing these missing segments, a physical-guided sequence prediction method is employed. The core principle is to construct a prediction model using the inherent physical constraints between environmental parameters. Different environmental parameters have different physical constraints; for example, dust concentration is negatively correlated with wind speed and air humidity, while noise is positively correlated with the operating status of construction machinery.

[0043] Taking the repair of missing segments of dust concentration parameters as an example, the prediction model is constructed as follows:

[0044]

[0045] In the above formula, For the first Each sensor group corresponds to the dust concentration parameter over time. The predicted values ​​are used to fill in the missing data at that moment; These are the fused values ​​of wind speed parameters corresponding to the same sensor group; These are the fused values ​​of air humidity parameters corresponding to the same sensor group. , , These are the model coefficients, whose values ​​can be determined through multiple linear regression analysis of historical monitoring data at the construction site to ensure the good fit between the predicted values ​​and the actual data. Using this model, the missing values ​​of dust concentration parameters can be predicted based on the wind speed and humidity parameters corresponding to the missing segments, thus completing the repair of the fused environmental parameter sequence and ensuring the continuity and integrity of the sequence.

[0046] After the repair is completed, based on the geographic coordinates and comprehensive confidence scores of each spatial region, the discrete fused environmental parameter sequence is transformed into a regular gridded environmental data cube covering the entire construction site using an interpolation algorithm. First, a regular grid covering the entire construction site is constructed. The grid resolution needs to be reasonably set according to the site monitoring accuracy requirements, and the grid's coordinate range covers the entire construction site area to ensure no monitoring blind spots. The Kriging interpolation algorithm is used for gridding processing. This algorithm can fully consider the spatial correlation and comprehensive confidence scores of the data, improving the accuracy of the interpolation results.

[0047] The core of Kriging interpolation is constructing a mutation function. The calculation formula is:

[0048]

[0049] In the above formula, The variability function is used to describe the spatial or temporal variability of monitoring data; The time interval or spatial distance between two sampling points reflects the degree of correlation between the sampling points; The interval is The number of sampling point pairs; For the first The sensor group in the first The fused environmental parameter values ​​of each sampling point; For the first The sensor group is in contact with the first The interval between sampling points is The environmental parameter values ​​of the sampling points are fused. After obtaining the interpolation model by fitting the variogram function, the environmental parameter values ​​of each grid node at each time step are calculated, and finally a regular gridded environmental data cube covering the entire construction site is generated. .

[0050] The dimensions of this data cube are ,in This represents the number of nodes in the horizontal direction of the grid. This represents the number of nodes in the grid along the ordinate direction. For time points, the data cube elements Characterizing grid nodes In time The environmental parameter values ​​are used to convert discrete data into continuous data across the entire domain, providing comprehensive data support for subsequent environmental status analysis.

[0051] S4. Extract multidimensional environmental state feature vectors from the environmental data cube, construct a dynamic benchmark model based on historical environmental monitoring data of the construction site that is related to the construction stage and external weather conditions, and calculate the deviation of the multidimensional environmental state feature vectors from the dynamic benchmark model at the current moment; when the deviation exceeds the threshold, trigger an environmental anomaly warning.

[0052] Specifically, multi-dimensional environmental state feature vectors are extracted from the environmental data cube. The dimensions of the feature vectors are determined according to the types of environmental parameters, covering the statistical characteristics of various key environmental parameters, such as average temperature and humidity, peak dust concentration, and effective noise levels. The feature vectors are calculated by statistically analyzing the environmental parameter values ​​of each grid node in the data cube and integrating them to obtain environmental state characteristics across the entire domain, comprehensively reflecting the overall condition of the current construction site environment.

[0053] A dynamic benchmark model is constructed based on historical environmental monitoring data of construction sites, relating it to construction stages and external weather conditions. This dynamic benchmark model can adapt to differences in environmental states under different construction stages and weather conditions, improving the accuracy of anomaly detection. The construction method involves collecting historical monitoring data from the construction site and classifying it according to construction stages and weather conditions. Construction stages include earthwork excavation, main structure construction, and decoration and finishing, while weather conditions include sunny, cloudy, rainy, and windy days. Statistical analysis is performed on the multidimensional environmental state feature vectors of each category to obtain the benchmark feature vectors and their fluctuation ranges for the corresponding categories, serving as a reference for environmental state assessment.

[0054] After completing the construction of the dynamic baseline model, calculate the deviation of the multidimensional environmental state feature vector at the current moment from the dynamic baseline model. The calculation formula is:

[0055]

[0056] In the above formula, This represents the deviation of the current feature vector from the dynamic baseline model, used to quantify the degree of difference between the current environmental state and the baseline state. This is the multidimensional environmental state feature vector extracted at the current moment; For a dynamic benchmark model, where These are the parameters for the current construction phase. These are the current external weather condition parameters; It is the L2 norm, calculated as follows: ,in For any vector, For vectors The Each dimension element For vectors The number of dimensions.

[0057] A pre-set deviation threshold is established, determined through analysis of historical anomaly data from the construction site, ensuring accurate differentiation between normal fluctuations and abnormal states. When the calculated deviation exceeds the threshold, the system determines that the current construction site environment is abnormal and triggers an environmental anomaly warning. The warning information can be pushed to relevant management personnel through various channels such as SMS, audible and visual alarms, and the construction site management platform, enabling timely control measures to be taken.

[0058] The aforementioned construction site environmental monitoring method based on multi-source fusion unifies the spatiotemporal benchmark of multi-source heterogeneous sensor data through spatiotemporal alignment processing. It then innovatively introduces and calculates a real-time data quality factor to dynamically evaluate the reliability of each sensor's data. Based on this factor, it weights and fuses neighboring sensor data to improve data reliability and generates a comprehensive confidence sequence characterizing the overall reliability of regional data. Next, it intelligently repairs the fused sequence using physical constraints between environmental parameters and generates a refined environmental data cube covering the entire region through spatial interpolation algorithms, combining geographic coordinates and the confidence sequence, thus transforming discrete point data into continuous surface information. Finally, it extracts multi-dimensional feature vectors from this data cube and constructs a dynamic benchmark model based on construction stages and weather conditions for comparative analysis, achieving intelligent early warning based on the deviation of the overall environmental pattern. This collectively achieves full-chain optimization of the construction site environment from bottom-level data acquisition and midstream fusion processing to upper-level intelligent analysis, significantly improving the quality and reliability of environmental monitoring data, realizing panoramic continuous perception of the construction site environment, and greatly enhancing the accuracy and foresight of anomaly early warning.

[0059] refer to Figure 2 In one optional embodiment, based on the local statistical characteristics and factory accuracy indicators of the sensor data in the multi-sensor time series matrix, the real-time data quality factor of each environmental sensor at each moment is calculated; based on the real-time data quality factor, the sensor data of a group of environmental sensors monitoring the same environmental parameter and located in close proximity are weighted and fused to generate a fused environmental parameter sequence for several spatial regions and its corresponding comprehensive confidence sequence, including the following steps:

[0060] S11. For the readings of environmental sensor i at time t in the multi-sensor time series matrix, calculate the corresponding historical mean and historical standard deviation based on the historical data of the environmental sensor within the sliding time window.

[0061] Specifically, the sliding time window setting needs to be adapted to the sensor sampling frequency to ensure that the historical data within the window can cover a sufficient sample size to guarantee the reliability of the statistical results, while also capturing recent trends in data changes, avoiding statistical characteristics lagging behind actual data fluctuations due to an excessively large window. The window length is calibrated through the sensor's sampling period, including only continuous historical sampling data before time t, excluding current readings and future data, to prevent the introduction of data redundancy or prediction bias.

[0062] During the calculation, all valid historical readings of sensor i within the sliding time window are first extracted, and obvious outliers that have been identified are removed (this is only a preliminary screening and does not affect subsequent quality assessment). Then, the historical mean and historical standard deviation are obtained through statistical calculations. The historical mean is used to characterize the stationary central value of the data within the window, reflecting the typical reading range of the sensor under normal operating conditions; the historical standard deviation is used to quantify the dispersion of the data within the window, reflecting the stability of the sensor readings. Together, they constitute the core indicators of the local statistical characteristics of the data, providing a basis for subsequent assessment of abnormal deviations.

[0063] S12. Combining the historical mean, historical standard deviation, and factory accuracy indicators of environmental sensor i, the real-time data quality factor of environmental sensor i at time t is calculated by fusing the anomaly deviation evaluation term, the data continuity evaluation term, and the inherent accuracy weight term; wherein, the formula for calculating the real-time data quality factor is:

[0064]

[0065] in, Indicates environmental sensors At any moment Real-time data quality factor; Indicates environmental sensors At any moment The original readings; and These represent the sensors. Historical mean and historical standard deviation; This is an adjustable coefficient used to control the sensitivity of anomaly detection; Indicates environmental sensors Data missing rate within the recent time window; This is the preset missing rate threshold; Indicates environmental sensors The factory precision specifications; This indicates the total number of environmental sensors in the environmental sensor cluster; , , Let be the weight coefficient, and satisfy... .

[0066] Specifically, this quality factor comprehensively reflects the reliability of the sensor's current readings. The three evaluation items construct an evaluation system from three dimensions: data deviation characteristics, transmission stability, and inherent equipment accuracy. By balancing the influence of each dimension through weight allocation, a standardized quantitative quality index is finally formed.

[0067] Let be the real-time data quality factor of environmental sensor i at time t, with a value range of [0,1]. The higher the value, the stronger the reliability of the sensor reading at that time, and it can be directly used as the weight basis for subsequent data fusion. The raw readings collected by environmental sensor i at time t are extracted directly from the multi-sensor time series matrix without any preprocessing. and These are the historical mean and historical standard deviation of sensor i within the sliding time window, respectively, which are the calculation results of step S11 and are the core parameters reflecting the local statistical characteristics of the data.

[0068] The adjustable coefficient is used to control the sensitivity of anomaly detection. Its value needs to be calibrated in conjunction with the interference environment at the construction site. By adjusting this coefficient, it can be adapted to the anomaly identification needs under different construction scenarios. The larger the coefficient value, the lower the anomaly detection sensitivity, and the less likely it is to judge normal fluctuations as anomalies; the smaller the coefficient value, the higher the sensitivity, and the more likely it is to capture subtle abnormal deviations. The data missing rate of environmental sensor i within a recent time window is calculated as the ratio of the number of missing data points within the window to the total number of sampling points in the window. It is used to evaluate the continuity of data transmission of the sensor. The higher the missing rate, the worse the stability of the sensor's operation and the lower the reliability of the corresponding data.

[0069] A preset missing rate threshold is used to define the acceptable range for sensor data continuity. When the threshold is exceeded, the value of the data continuity assessment item decreases, thereby lowering the overall quality factor. The factory accuracy specification of environmental sensor i is determined by the technical manual provided by the sensor manufacturer. It characterizes the inherent measurement error of the sensor under standard operating conditions; the smaller the error, the better. The smaller the value, the higher the inherent accuracy of the sensor. This represents the total number of sensors in the environmental sensor cluster, ensuring that the statistical range is consistent with the cluster range of subsequent fusion.

[0070] , , These are the weighting coefficients for the three evaluation items, and they strictly satisfy... The specific values ​​are calibrated through regression analysis of historical monitoring data from the construction site, and the weight allocation ratio is adjusted according to the monitoring needs of different environmental parameters. Typically, the weight of the abnormal deviation assessment item is... The highest percentage is due to the fact that real-time data deviation has the most direct impact on monitoring results; data continuity assessment item and inherent precision weighting term Fine-tune the settings according to the sensor deployment scenario. If the sensor deployment environment has strong interference, the settings can be appropriately increased. The percentage can be appropriately increased if there are significant differences in sensor models. Percentage.

[0071] The three evaluation terms in the formula each have their own emphasis: the first term, the abnormal deviation evaluation term, adopts an exponential function form, such that when and The larger the deviation, the faster the value of this item decays, which can quickly reduce the weight of abnormal readings; the second item, data continuity assessment, is calculated by the ratio of missing rate to threshold. When the missing rate is below the threshold, the value of this item is close to 1. After the missing rate exceeds the threshold, the value decreases linearly as the missing rate increases, which intuitively reflects the stability of data transmission; the third item, inherent accuracy weight, is normalized to convert the factory accuracy of each sensor into a relative weight, ensuring that sensors with higher accuracy receive a higher proportion in the quality assessment.

[0072] S13. For an environmental sensor cluster whose spatial distance is less than a set threshold and monitors the same environmental parameter, calculate the fused environmental parameter value of the corresponding environmental sensor cluster at time t based on the real-time data quality factor of each environmental sensor in the cluster and its raw reading at time t; wherein, the calculation formula for the fused environmental parameter value is:

[0073]

[0074] in, Represents an environmental sensor cluster At any moment The fusion environment parameter values; This refers to a collection of environmental sensors. For traversing the set Index of all sensors in the system; Indicates environmental sensors At any moment Real-time data quality factor; Indicates environmental sensors At any moment The original readings.

[0075] Specifically, the division of sensor clusters requires clarifying two core conditions: first, the consistency of monitoring parameters, ensuring that all sensors in the cluster collect data for the same environmental parameter, and avoiding the fusion of heterogeneous data across parameters; second, spatial distance constraints, defining the cluster range by setting spatial distance thresholds, ensuring that sensors in the cluster cover the same local spatial area, and that the collected data has spatial correlation, reflecting the overall state of environmental parameters in that area.

[0076] The spatial distance threshold needs to be set in conjunction with the site size, sensor deployment density, and spatial diffusion characteristics of environmental parameters to ensure that the correlation of sensor readings within the same cluster meets the fusion requirements, while avoiding excessively large clusters that could lead to decreased data representativeness. After partitioning, each cluster corresponds to a specific spatial area within the site, and the number of sensors within the cluster is dynamically adjusted according to the deployment density to ensure that each area has sufficient sensor samples to support fusion calculations.

[0077] The fusion of environmental parameter values ​​is calculated using a weighted average algorithm. The core logic is to give higher weight to sensor readings with higher real-time data quality factors in the fusion result, thereby suppressing the interference of low-quality data on the fusion accuracy.

[0078] This represents the fused environmental parameter value of the environmental sensor cluster S at time t. This value accurately characterizes the actual state of environmental parameters in the corresponding spatial region of the cluster at time t, and is more reliable than single sensor readings. S is the set of environmental sensor clusters, including all sensors that meet the conditions of consistent monitoring parameters and spatial distance constraints. The set size is determined by spatial coordinate verification after the sensor deployment is completed. If the sensor positions are adjusted, the cluster division results need to be recalibrated.

[0079] The real-time data quality factor of sensor i at time t, i.e. the calculation result of step S12, serves as the core weight basis for weighted fusion. The raw reading of sensor i at time t is matched with the corresponding quality factor to ensure that the weight allocation of each data point accurately corresponds to its reliability. In the formula, the numerator is the sum of the products of all sensor readings in the cluster and their corresponding quality factors, and the denominator is the sum of the quality factors of all sensors in the cluster. This weighted calculation can maximize the retention of the reference value of high-quality data and reduce the interference of low-quality data.

[0080] S14. Take the average of the real-time data quality factors of all environmental sensors in the environmental sensor cluster as the comprehensive confidence level of the fused environmental parameter value at time t; traverse all time points and all environmental sensor clusters to generate the fused environmental parameter sequence and its corresponding comprehensive confidence level sequence for each spatial region.

[0081] Specifically, the overall confidence score is used to quantify the credibility of the fusion result. Its value directly reflects the overall working status and data quality level of the sensors in the cluster. If the quality factors of most sensors in the cluster are high, the overall confidence score will be close to a high value, indicating that the fusion result is highly reliable. If the quality factors of multiple sensors are low, the overall confidence score will be reduced accordingly, providing a reliable reference for subsequent data processing.

[0082] The calculation of the overall confidence level requires traversing all sensors within the cluster. The values ​​at each time point are obtained by arithmetic averaging to ensure a comprehensive reflection of the overall data quality of the cluster. Then, all time points and all environmental sensor clusters are traversed. For each cluster, the fused environmental parameter values ​​at each time point are paired with corresponding comprehensive confidence scores to generate fused environmental parameter sequences and comprehensive confidence score sequences for each spatial region. The fused environmental parameter sequence is a set of fused values ​​for the same cluster at consecutive time points, reflecting the temporal trend of environmental parameters in that region. The comprehensive confidence score sequence synchronously corresponds to the reliability of the fused results at each time point, providing data quality references for subsequent processing steps such as missing data repair and global gridding, ensuring the reliability of subsequent analysis results.

[0083] In one optional embodiment, based on the geographic coordinates and comprehensive confidence sequence of each spatial region, an interpolation algorithm is used to convert the discrete fused environmental parameter sequence into a regular gridded environmental data cube covering the entire construction site, including the following steps:

[0084] S21. Obtain the geographic coordinates of the center point of each spatial region, as well as the corresponding fusion environment parameter values ​​and their comprehensive confidence levels at each time.

[0085] Specifically, each spatial region corresponds to an environmental sensor cluster. The geographic coordinates of the center point of each cluster are calculated by weighting the geographic coordinates of all sensors within the cluster. The weighting is linked to the real-time data quality factor of the sensors; sensors with higher quality factors contribute more to the center point coordinates, ensuring that the center point accurately represents the spatial location characteristics of the region. The fused environmental parameter values ​​and the overall confidence score both originate from the sequence generated in step S14. The fused environmental parameter values ​​characterize the actual state of environmental parameters in the corresponding region at a given time, while the overall confidence score quantifies the reliability of the fused value. These two values ​​are correlated one-to-one, providing foundational data with confidence-labeled values ​​for subsequent interpolation calculations. During the acquisition process, data consistency verification is required to ensure that the geographic coordinates, fused environmental parameter values, and overall confidence scores of each spatial region are completely synchronized over time, avoiding data misalignment or missing correlations.

[0086] S22. Based on the spatial distance of geographic coordinates, the time difference between adjacent times, and the comprehensive confidence level, a spatiotemporal variogram model incorporating confidence level weights is constructed. The spatiotemporal variogram model quantifies the correlation of data in the spatiotemporal dimension; the comprehensive confidence level is used to adjust the correlation measure between data points, with data points having higher comprehensive confidence levels receiving greater weight in the spatiotemporal variogram model calculation. The expression for the spatiotemporal variogram model is:

[0087]

[0088] in, Represents known data points and The spatiotemporal variation function values ​​between; and Representing data points respectively and The overall confidence level; Representing data points and The spatial Euclidean distance between them; Representing data points and The absolute time difference between corresponding timestamps; For spatial characteristic range parameters, The time-feature range parameter represents the critical distance at which spatial correlation essentially disappears, while the time-feature range parameter represents the critical time span at which time correlation essentially disappears. and Let be the weighting coefficients for the spatial and temporal components, and satisfy . ;

[0089] The spatial characteristic range parameter and the temporal characteristic range parameter are obtained as follows: high-quality data points with historical comprehensive confidence scores higher than the preset confidence threshold are selected from the historical fusion environment parameter sequence and its corresponding historical comprehensive confidence scores; based on the spatial distance and time difference of the high-quality data points, the spatial characteristic range parameter and the temporal characteristic range parameter are determined by fitting the empirical variogram curve.

[0090] Specifically, the core function of the spatiotemporal variogram model is to quantify the correlation of data in both spatiotemporal dimensions, breaking the limitation of traditional models that only consider correlation in a single dimension. At the same time, by adjusting the correlation measurement weights between data points through comprehensive confidence, the model enables high-confidence data points to dominate the correlation calculation, thereby increasing the model's reliance on high-quality data.

[0091] The spatiotemporal variability function value between known data point i and known data point j is used. The larger the value, the weaker the correlation between the two data points in the spatiotemporal dimension, and vice versa. It is the core indicator for measuring the degree of data correlation in subsequent interpolation operations. and These are the combined confidence scores for data points i and j, respectively, directly taken from the combined confidence score sequence obtained in step S21. They are used to adjust the correlation metric weights between the two points. Calculate the collaborative effect of the two-point confidence levels, and then... By implementing weight calibration, the coefficient of this term is closer to 0.5 for data points with higher overall confidence, and the weakening effect on the variogram value is more obvious, thereby strengthening the correlation measurement weight between high-confidence data points.

[0092] The spatial Euclidean distance between data point i and data point j is calculated using the geographic coordinates of the two points. It reflects the physical distance between the two points in the spatial dimension. The larger the distance, the weaker the spatial correlation usually is. The absolute time difference between the timestamps corresponding to data point i and data point j reflects the interval between the two points in the time dimension. The larger the interval, the weaker the time correlation usually is. This is a spatial characteristic range parameter, representing the critical distance at which spatial correlation essentially disappears. When the Euclidean distance between two points exceeds this parameter, it can be considered that the two data points have no significant correlation in the spatial dimension and do not need to be included in the scope of mutual influence. The time range parameter represents the critical time span at which time correlation essentially disappears. When the time difference between two points exceeds this parameter, the two data points can be considered to have no significant correlation in the time dimension.

[0093] and These are the weighting coefficients for the spatial and temporal components, respectively, and they strictly satisfy... This is used to balance the effects of spatial and temporal correlations on the variogram value. The value is calibrated using historical monitoring data from the construction site. If the spatial distribution of environmental parameters differs significantly, it can improve... Percentage; if environmental parameters fluctuate drastically over time, it can be increased. Percentage.

[0094] Among them, spatial characteristic range parameters and time characteristic range parameters The acquisition process is as follows: First, from the historical fusion environment parameter sequence and its corresponding historical comprehensive confidence, high-quality data points with historical comprehensive confidence scores higher than the preset confidence threshold are selected, and low-confidence data points are removed to avoid interfering with the accuracy of parameter calculation; then, based on these high-quality data points, the variance values ​​corresponding to different spatial distances and different time differences are statistically analyzed, and empirical variance function curves are plotted; finally, spatial characteristic range parameters and temporal characteristic range parameters are determined through curve fitting. During the fitting process, the critical value when the curve tends to stabilize is used as the range parameter to ensure that the parameters can accurately reflect the critical condition of the disappearance of correlation.

[0095] S23. To meet the requirements of the interpolation rule grid points at the target time, establish the corresponding Kriging equation system based on the spatiotemporal variogram model; the goal of the Kriging equation system is to minimize the variance of the interpolation estimate and satisfy the unbiasedness condition.

[0096] Specifically, the core advantage of the Kriging interpolation algorithm is that it can minimize the variance of the interpolation estimate while satisfying the unbiasedness condition, thereby maximizing the accuracy of the interpolation result. The construction of the equation system is the core link to achieve this goal. Its essence is to establish the correlation equation between the known data points and the points to be interpolated by using the spatiotemporal variability function value.

[0097] During the construction process, the target time and spatial coordinates of the grid points to be interpolated are first determined. Then, known data points within the spatiotemporal neighborhood of the grid point are selected, with the selection range based on the spatial characteristic range parameter. and time characteristic range parameters To define the boundaries, only known data points with significant spatiotemporal correlation to the points to be interpolated are included, reducing the interference of irrelevant data on the interpolation results. Then, based on the selected known data points, the spatiotemporal variogram values ​​among the known data points are calculated using the spatiotemporal variogram model constructed earlier. And the spatiotemporal variability function values ​​between known data points and the regular grid points to be interpolated. These values ​​are used as the core inputs to the system of equations. By combining unbiased constraints and the minimum variance objective, the corresponding Kriging equations are established.

[0098] S24. Solve the Kriging equations to obtain the known point weight coefficients for each known point used to calculate the environmental parameter values ​​of the regular grid points to be interpolated; where the Kriging equations are expressed as:

[0099]

[0100] in, Indicates the first The known point weight coefficients of the known data points to be interpolated into the grid points of the interpolation rule; Represents a known point With known points The spatiotemporal variation function values ​​between; Represents a known point With the grid points of the interpolation rule The spatiotemporal variation function values ​​between; For Lagrange multipliers; This represents the total number of known points involved in the interpolation calculation.

[0101] Specifically, the Kriging equations achieve a balance between unbiasedness constraints and the minimum variance objective by introducing Lagrange multipliers.

[0102] The known point weight coefficient for the j-th known data point is the known point of the interpolation rule grid point. It represents the degree of influence of the known data point on the interpolation result. The larger the weight coefficient, the stronger the spatiotemporal correlation between the corresponding known data point and the interpolation point, and the higher the contribution to the interpolation result. The spatiotemporal variability function value between known point i and known point j is the result calculated based on the spatiotemporal variability function model in step S23, reflecting the degree of spatiotemporal correlation between the known points. The spatiotemporal variability function value between the known point i and the interpolation point 0 of the regular grid is also calculated through the spatiotemporal variability function model, reflecting the spatiotemporal correlation between the known point and the interpolation point.

[0103] The Lagrange multiplier is designed to satisfy the unbiased constraint of Kriging interpolation, ensuring that the expected value of the interpolation result is consistent with the actual environmental parameter value of the interpolation point, thus fundamentally avoiding the impact of systematic bias on interpolation accuracy. This multiplier is not a fixed value but is dynamically adapted based on the interpolation scenario. Its value is closely related to the spatiotemporal distribution characteristics and spatiotemporal variability of the known data points involved in the interpolation, and must be determined by simultaneously solving a system of Kriging equations. Specifically, the logic is as follows: Kriging interpolation uses "minimizing the variance of the interpolation estimate" as its objective function, while the unbiased constraint requires that the mathematical expectation of the interpolation estimate equals the actual value of the interpolation point. This is achieved by introducing the Lagrange multiplier. Constrained optimization problems can be transformed into unconstrained extremum problems, achieving the synergistic satisfaction of the objective function and constraints.

[0104] In the process of setting up Lagrange multipliers, The range of values ​​is determined by the characteristics of the known data points involved in the interpolation. When the known data points are evenly distributed in the spatiotemporal neighborhood of the point to be interpolated and the spatiotemporal variability function value fluctuates little, the range is determined by the characteristics of the known data points involved in the interpolation. A value close to 0 indicates that the unbiased constraint has a weak corrective effect on weight allocation; when the known data points are sparsely distributed and the spatiotemporal variability function values ​​differ significantly, It will display non-zero values, which can be adjusted by changing the weighting coefficients. The allocation ratio is used to compensate for the bias caused by uneven data distribution, ensuring that the interpolation results meet both the unbiasedness requirement and maintain the minimum variance characteristic. Meanwhile, The solution and the weight coefficients of each known point This is done simultaneously by transforming the Kriging equations into matrix form and inverting the coefficient matrix, which yields the results at the same time. and The precise selection of values ​​for both the matrix and the interpolation method are mutually restrictive and complementary, ensuring the rationality and accuracy of the interpolation calculation. During the solution process, it is necessary to ensure matrix invertibility. If singular matrices occur, the equation system can be reconstructed by adjusting the range of known points, thus obtaining an effective solution suitable for the current scenario. The value is n, which represents the total number of known points involved in the interpolation calculation. This is the number of known data points selected in step S23 that have a significant spatiotemporal correlation with the points to be interpolated. The number of these points needs to be adjusted reasonably based on the range parameter and grid density to ensure sufficient data to support interpolation accuracy without increasing computational cost due to excessive data volume. Solving the equation system can be achieved through matrix operations. After transforming the equation system into matrix form, the weight coefficients of each known point are obtained through matrix inversion and other operations. With Lagrange multipliers During the solution process, it is necessary to ensure that the matrix is ​​invertible. If a singular matrix occurs, the system of equations can be reconstructed by adjusting the range of known points.

[0105] S25. Based on the weight coefficients of the known points, the fused environmental parameter values ​​of each known point are linearly combined to obtain the estimated environmental parameters of the regular grid points to be interpolated at the target time; based on the fused environmental parameter values ​​and the estimated environmental parameters, a regular gridded environmental data cube is formed.

[0106] Specifically, the core logic of linear combination is to allocate the influence of each known data point according to the weight coefficient, so that the known data points with strong correlation and high credibility dominate the interpolation result. The calculation formula can be expressed as the estimated value of the interpolation point is equal to the sum of the product of the fused environmental parameter value of each known point and the corresponding weight coefficient, ensuring that the interpolation result can make full use of the information of high-quality known data.

[0107] After estimating the environmental parameters of a single grid point to be interpolated at the target time, the process iterates through all grid points to be interpolated across the entire construction site, as well as all monitoring time points, calculating the estimated environmental parameters for each grid point at each time. Subsequently, the fused environmental parameter values ​​of all known spatial areas are integrated with the estimated environmental parameters of each grid point to form a regular gridded environmental data cube. This data cube uses the spatial coordinates (x-axis and y-axis) and timestamps of the grid points as a three-dimensional index, with each index corresponding to a unique environmental parameter value. This allows for complete coverage of the entire construction site space and the entire monitoring time series, realizing the conversion of discrete data into continuous, full-domain data, and providing comprehensive and accurate data support for subsequent multi-dimensional environmental state feature extraction and anomaly early warning.

[0108] In one alternative embodiment, extracting a multidimensional environmental state feature vector from the environmental data cube includes the following steps:

[0109] S31. Calculate the spatial average of all grid point values ​​of each type of environmental parameter in the regular gridded environmental data cube to obtain the spatial average of the corresponding type of environmental parameter at the construction site.

[0110] Specifically, the regular gridded environmental data cube comes from the output of step S25. It contains environmental parameter values ​​of all grid points across the entire construction site at various times, covering multiple types of environmental parameters such as temperature, humidity, dust concentration, and noise. Before calculation, the data cube is classified and split according to the type of environmental parameter to ensure that all grid point values ​​of the same type of parameter are grouped together, avoiding cross-parameter interference with the calculation results.

[0111] Spatial averaging is performed using the arithmetic mean method. It iterates through all valid grid point values ​​for the corresponding parameter type, removes invalid values ​​due to interpolation failures, and then calculates the average of all valid values ​​as the spatial mean for that parameter type. The formula for calculating the spatial mean is as follows:

[0112]

[0113] In the above formula, For the first Spatial mean of class environment parameters; For the first The total number of valid grid points corresponding to the class parameter, i.e., the number of grid points after removing invalid values, must satisfy the following condition: ( To preset a minimum effective grid point threshold, and to avoid mean distortion due to insufficient effective data (to set a minimum effective grid point threshold). For the first Class parameter number The environmental parameter values ​​of each valid grid point are extracted directly from the regular gridded environmental data cube.

[0114] The core function of spatial mean is to characterize the overall level of the corresponding environmental parameter across the entire construction site, mitigating the impact of minor local fluctuations and reflecting the macroscopic distribution characteristics of the environmental parameter. For example, the spatial mean of dust concentration can intuitively reflect the overall dust pollution level of the current construction site, providing a basic reference for subsequent judgments on whether the overall environmental status is abnormal. During the calculation process, the number of grid points participating in the averaging must be recorded simultaneously. If the proportion of invalid grid points is... ( This represents the total number of grid points corresponding to this parameter. (To set a threshold for the percentage of invalid values), the interpolation results of the data cube need to be recalibrated to ensure the reliability of the spatial mean.

[0115] S32. For the pre-selected key environmental parameters in the regular gridded environmental data cube, calculate the spatial gradient of the key environmental parameters along the preset dominant direction, and extract the maximum value of the spatial gradient as the maximum value of the spatial gradient.

[0116] Specifically, the pre-selection of key environmental parameters needs to be determined in conjunction with the key points of environmental monitoring at the construction site. Parameters that have a significant impact on construction safety and personnel health are usually selected, such as dust concentration, noise, and concentration of toxic and harmful gases. The pre-selection list can be dynamically adjusted according to the type of construction and regulatory needs at the construction site.

[0117] The setting of the preset dominant direction needs to be calibrated in conjunction with the actual construction site scenario. It is mainly determined based on factors such as the operating trajectory of construction machinery, material transportation routes, and pollutant diffusion patterns. Common dominant directions include the extension direction of the main construction road, the prevailing wind direction, and the gradient direction of the construction area. The direction setting can be optimized through analysis of the site layout plan and historical environmental monitoring data. The core of spatial gradient calculation is to quantify the rate of change of key environmental parameters along the dominant direction, reflecting the degree of spatial distribution difference of these parameters. The calculation formula is as follows:

[0118]

[0119] In the above formula, For the first Class of key environmental parameters along the dominant direction Spatial gradient values ​​of grid point pairs; , These are two adjacent grid points along the dominant direction of this parameter (the first one). The, the (each) environmental parameter values; The grid resolution is the spatial distance between adjacent grid points in the dominant direction, which is equal to the Euclidean distance between the grid horizontal and vertical coordinate intervals (in regular grids, it can be simplified to the horizontal or vertical coordinate interval, depending on the angle between the dominant direction and the grid axis).

[0120] By traversing all adjacent grid point pairs along the dominant direction, a series of spatial gradient values ​​are calculated. ( After determining the total number of adjacent grid point pairs in that direction, the maximum value is extracted as the spatial gradient maximum. The calculation formula is as follows:

[0121]

[0122] In the above formula, For the first Maximum spatial gradient of key environmental parameters; The function for finding the maximum value iterates through all gradient values ​​to select the peak value. This indicator can accurately locate the region with the most dramatic spatial changes in key environmental parameters. For example, the region corresponding to the maximum value of the dust concentration gradient may be the location of a dust pollution source or a diffusion boundary, providing directional guidance for subsequent accurate investigation of anomaly sources, while also supplementing local difference features that cannot be covered by the spatial mean.

[0123] S33. Compare the values ​​of each environmental parameter in the regular gridded environmental data cube with the preset environmental quality standard threshold, and calculate the proportion of the area of ​​grid points that exceed the environmental quality standard threshold to the total monitored area to obtain the proportion of the area exceeding the standard.

[0124] Specifically, the setting of environmental quality standard thresholds should be based on relevant environmental monitoring specifications and industry standards. Corresponding thresholds should be set for different types of environmental parameters. At the same time, the thresholds can be fine-tuned in combination with the internal management requirements of the construction site to ensure the compliance and applicability of the thresholds.

[0125] The comparison process employs a grid-by-grid verification method. For each grid point value of each environmental parameter, it is determined whether it exceeds the corresponding threshold, and the grid point is marked as exceeding the threshold (meeting the requirements). , For the first Environmental quality standard thresholds for class parameters For the first The environmental parameter value of the k-th grid point along the dominant direction of the class of key environmental parameters) or qualified grid points (satisfying the requirements) Then, the relevant areas and proportions are calculated sequentially, using the following formulas:

[0126]

[0127]

[0128]

[0129]

[0130] In the above formula, The area of ​​a single grid point; , These are the horizontal and vertical coordinate intervals of the regular grid (i.e., the components of the grid resolution in the horizontal and vertical directions). For the first Total area of ​​grid points exceeding the class parameter limit; For the first The number of grid points exceeding the limit for class parameters; The total monitored area of ​​the entire construction site; This represents the total number of rule grids across the entire construction site. For the first The percentage of the area exceeding the standard for a given parameter ranges from [0,1]. A larger value indicates a wider range of exceedances for that environmental parameter, and a higher overall environmental risk. This indicator can quantify the coverage of abnormal states, overcome the limitations of judging exceedances at a single grid point, and reflect the overall impact of environmental anomalies.

[0131] S34. The reciprocal of the variance of each environmental parameter value in the regular gridded environmental data cube within the sliding time window is used as a spatiotemporal stability index characterizing the degree of environmental fluctuation.

[0132] Specifically, the sliding time window setting must be consistent with the window logic in step S11. The window length is calibrated according to the fluctuation cycle of the environmental parameters to ensure that it can capture the short-term fluctuation characteristics of the parameters while avoiding numerical oscillations caused by excessively short windows. The window only includes continuous historical data before the current moment and does not include future data to ensure the real-time nature of the indicator calculation.

[0133] Variance calculation is performed for the environmental parameter time series of each grid point. It iterates through all parameter values ​​of that grid point within the sliding time window, first calculating the mean within the window, and then solving for the variance. The specific formula is as follows:

[0134]

[0135]

[0136]

[0137] In the above formula, For the first Class parameter number The average time of each grid point within the sliding time window; The length of the sliding time window (consistent with step S11); For this grid point within the window Environmental parameter values ​​at each moment; The variance of the parameter values ​​at this grid point within the window is calculated using unbiased variance (denominator is taken as...). This avoids underestimation of variance when the sample size is small; This serves as the spatiotemporal stability index for this grid point. A larger variance value indicates more drastic fluctuations in the parameters of this grid point over time, resulting in poorer stability. Taking the reciprocal of the variance as the spatiotemporal stability index reveals a positive correlation between the index value and stability. The larger the value, the smaller the fluctuation of environmental parameters, the stronger the spatiotemporal stability, and the easier it is to perform unified analysis and weight allocation of subsequent feature vectors.

[0138] After the calculation is completed, the spatiotemporal stability index of all grid points for the same environmental parameter is comprehensively processed, and the global mean is taken as the spatiotemporal stability feature corresponding to the parameter. This ensures that the index can reflect the overall fluctuation of the parameter across the entire construction site, rather than the local fluctuation of a single grid point, and is consistent with the global characteristics of other feature dimensions.

[0139] S35. The spatial mean, the maximum spatial gradient, the proportion of the area exceeding the standard, and the spatiotemporal stability index are concatenated to obtain a multidimensional environmental state feature vector.

[0140] Specifically, the splicing process needs to follow a fixed order to ensure the consistency of the dimension and order of the feature vectors, which facilitates subsequent comparison with the dynamic benchmark model. The splicing order can be as follows: arranged in order according to the type of environmental parameters, with the spatial mean, maximum spatial gradient, over-standard area ratio, and spatiotemporal stability index corresponding to each parameter as a set of features, and all feature groups of parameters are spliced ​​in order to form a complete vector.

[0141] For example, if a construction site monitors three parameters—dust concentration, noise, and temperature and humidity—the feature vectors would be in the following order: spatial mean of dust concentration, maximum spatial gradient of dust concentration, proportion of area exceeding dust concentration limits, spatiotemporal stability index of dust concentration; spatial mean of noise, maximum spatial gradient of noise, proportion of area exceeding noise limits, spatiotemporal stability index of noise; spatial mean of temperature and humidity, maximum spatial gradient of temperature and humidity, proportion of area exceeding temperature and humidity limits, and spatiotemporal stability index of temperature and humidity, forming a 12-dimensional feature vector.

[0142] Before splicing, each feature index is normalized to map all index values ​​to the [0,1] interval, eliminating the impact of differences in the dimensions of different indicators on subsequent deviation calculations. The normalization adopts a linear normalization method, and the maximum and minimum values ​​of each index are determined based on historical monitoring data. The specific formula is as follows:

[0143]

[0144] In the above formula, These are the normalized eigenvalues; The original value of the feature (which can be the spatial mean, the maximum spatial gradient, the proportion of the area exceeding the standard, or a spatiotemporal stability index); , These are the minimum and maximum values ​​of the feature, obtained statistically from historical monitoring data. These values ​​need to be stored in the system database in advance and can be updated periodically based on new historical data to ensure normalization accuracy. After normalization, the features are concatenated in a fixed order to form a multidimensional environmental state feature vector.

[0145] In one optional embodiment, a dynamic benchmark model is constructed based on historical site environmental monitoring data, relating it to the construction phase and external weather conditions. The deviation of the current multidimensional environmental state feature vector from the dynamic benchmark model is calculated, including the following steps:

[0146] S41. According to the construction schedule, the historical multidimensional environmental state feature vector is divided into corresponding construction stages to obtain historical feature vector subsets for different construction stages; external weather data of the same historical period as the historical data of site environmental monitoring is obtained, and the external weather data is used as the working condition label and associated with the corresponding historical feature vector subsets; wherein, the historical multidimensional environmental state feature vector is calculated based on the historical data of site environmental monitoring, and the calculation method is consistent with the calculation method of multidimensional environmental state feature vector.

[0147] Specifically, the historical multidimensional environmental state feature vector is calculated and generated based on historical data of construction site environmental monitoring. The calculation method is completely consistent with the extraction method of the multidimensional environmental state feature vector in step S35, ensuring that the dimensions and statistical logic of historical features and current features are consistent, and ensuring the effectiveness of subsequent benchmark comparisons.

[0148] The division of construction phases follows the pre-set construction schedule on the site, breaking down the project into several core phases according to the construction process. Common phases include earthwork excavation, foundation construction, main structure construction, and decoration and finishing. The boundaries of the phases are based on the key nodes in the schedule. Each historical feature vector is assigned to a unique construction phase based on the construction progress corresponding to its collection time. A time mapping table needs to be established during the phase division process to accurately match the collection timestamps of historical feature vectors with the time nodes in the construction schedule. If a certain moment falls within a phase transition period, the phase corresponding to the dominant construction procedure is used.

[0149] External weather data is obtained through API connections to authoritative meteorological platforms or local meteorological monitoring equipment. It is crucial to ensure that the data is synchronized with historical environmental monitoring data, meaning that each set of historical feature vectors is associated with external weather data at its collection time (or concurrently). Weather data needs to be refined into standardized operating condition labels. Label classification is determined based on the influence patterns of environmental parameters, with core dimensions including weather type (sunny, cloudy, rainy, snowy, etc.), wind speed, and precipitation intensity. Examples include combined labels such as "sunny-light breeze" and "rainy-moderate rain," with each label corresponding to a specific weather condition. During association, timestamp alignment is used to embed the operating condition labels into the attribute fields of the corresponding historical feature vectors, forming a three-dimensional associated data structure of "construction phase - weather label - historical feature vector," laying the foundation for subsequent operating condition combination classification.

[0150] S42. Based on the subset of historical feature vectors and external weather data, for each combination of working conditions consisting of a specific construction stage and specific external weather conditions, calculate the mean vector and covariance matrix of all historical feature vectors belonging to the combination of working conditions. The mean vector and covariance matrix together constitute the benchmark sub-model corresponding to the combination of working conditions in the dynamic benchmark model.

[0151] Specifically, the combination of working conditions is the Cartesian product of the construction stage and the weather condition label. That is, each construction stage and each type of weather condition form a unique combination, such as "earthwork excavation stage - rainy day - moderate rain", "main structure construction stage - sunny day - light wind", etc., to ensure that all possible working conditions are covered by a corresponding baseline sub-model.

[0152] For any combination of working conditions, first select all historical feature vectors belonging to that combination to form the feature vector set of that combination. , For the next combination 1 historical feature vector (dimension 1) ), This represents the total number of historical feature vectors under this combination. The mean vector is calculated as the arithmetic mean of the corresponding dimensions of all feature vectors in the set, using the following formula:

[0153]

[0154] In the formula, For the first The mean vector corresponding to the combination of various working conditions has the same dimension as the historical feature vector. For the next combination One historical feature vector; The total number of historical feature vectors under this combination must satisfy the following condition: ( To preset a minimum sample size threshold and avoid distortion of the mean vector due to insufficient sample size. The mean vector represents the baseline level of the environmental state under this working condition combination and is the core reference benchmark for subsequent deviation calculations.

[0155] The covariance matrix is ​​used to reflect the correlation and dispersion among the feature dimensions under this working condition combination. The formula is as follows:

[0156]

[0157] In the formula, For the first The covariance matrix corresponding to the combination of various working conditions has a dimension of . ( (for feature dimensions) For vectors Transpose; take the denominator Unbiased estimation is employed to ensure the accuracy of matrix calculations when the sample size is small. The diagonal elements of the covariance matrix represent the variance of each feature dimension, while the off-diagonal elements represent the covariance of the corresponding two feature dimensions. This quantifies the degree of linear correlation between features, providing variance and covariance information for Mahalanobis distance calculation and eliminating the influence of correlation and dimensional differences between features. The mean vector for each combination of operating conditions is also provided. With covariance matrix Together they form a baseline sub-model, and all baseline sub-models are integrated to form a complete dynamic baseline model, which is stored in a database and can be iteratively updated periodically based on newly added historical data.

[0158] S43. Determine the current working condition combination based on the actual construction stage and actual external weather conditions at the current moment, and select the reference sub-model corresponding to the current working condition combination from the dynamic reference model.

[0159] Specifically, the current actual construction phase is obtained in real time through the connection with the construction site progress management system. If the system is not updated in real time, the deviation between the preset progress plan and the actual construction period can be calibrated to ensure accurate phase determination. The current actual external weather conditions are obtained in real time through the connection with the meteorological platform or local meteorological equipment, and are extracted into standardized labels according to the S41 working condition labeling rules to ensure consistency with the historical weather label format.

[0160] The matching of working condition combinations adopts the principle of precise matching, that is, the combination of the current construction stage and the current weather label must be completely consistent with the preset working condition combination in the dynamic benchmark model. If there is a unique match, the corresponding benchmark sub-model is directly retrieved; if there is no complete match (such as no historical data for special weather conditions), the benchmark sub-model corresponding to the working condition combination with the highest similarity is selected (similarity is prioritized for matching construction stage, and then for matching weather type). At the same time, the matching deviation is recorded to provide a basis for subsequent early warning result calibration.

[0161] S44. Based on the mean vector and covariance matrix in the selected benchmark sub-model, calculate the Mahalanobis distance between the multidimensional environmental state feature vector and the mean vector at the current moment, and use the Mahalanobis distance as the deviation.

[0162] Specifically, the core advantage of Mahalanobis distance over Euclidean distance is that it can eliminate the influence of dimensional differences and linear correlations among various feature dimensions, better meet the deviation measurement needs of multi-dimensional feature vectors, and accurately reflect the degree of difference between the current environmental state and the baseline state.

[0163] The formula for calculating Mahalanobis distance is as follows:

[0164]

[0165] In the formula, The deviation at the current moment (i.e., Mahalanobis distance); This is the multidimensional environmental state feature vector calculated at the current moment (from the output of step S35); This is the mean vector corresponding to the selected baseline sub-model; Covariance matrix If the covariance matrix has singular values ​​(such as perfect linear correlation between features), then a small perturbation term is added to make the matrix invertible, ensuring that the calculation proceeds normally.

[0166] Mahalanobis distance The range of values ​​is The larger the value, the greater the deviation between the current feature vector and the baseline mean vector, meaning the more significant the difference between the current environmental state and the baseline state under the corresponding working condition combination. At that time, the current feature vector is completely consistent with the mean vector, indicating that the environmental state is at the baseline level; as As the value increases, the degree of deviation gradually increases. When it exceeds the preset threshold, an environmental anomaly warning is triggered, providing a quantitative basis for environmental management and control decisions.

[0167] The aforementioned construction site environmental monitoring method based on multi-source fusion unifies the spatiotemporal benchmark of multi-source heterogeneous sensor data through spatiotemporal alignment processing. It then innovatively introduces and calculates a real-time data quality factor to dynamically evaluate the reliability of each sensor's data. Based on this factor, it weights and fuses neighboring sensor data to improve data reliability and generates a comprehensive confidence sequence characterizing the overall reliability of regional data. Next, it intelligently repairs the fused sequence using physical constraints between environmental parameters and generates a refined environmental data cube covering the entire region through spatial interpolation algorithms, combining geographic coordinates and the confidence sequence, thus transforming discrete point data into continuous surface information. Finally, it extracts multi-dimensional feature vectors from this data cube and constructs a dynamic benchmark model based on construction stages and weather conditions for comparative analysis, achieving intelligent early warning based on the deviation of the overall environmental pattern. This collectively achieves full-chain optimization of the construction site environment from bottom-level data acquisition and midstream fusion processing to upper-level intelligent analysis, significantly improving the quality and reliability of environmental monitoring data, realizing panoramic continuous perception of the construction site environment, and greatly enhancing the accuracy and foresight of anomaly early warning.

[0168] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0169] Based on the same inventive concept, this application also provides a system for implementing the above-mentioned construction site environmental monitoring method based on multi-source fusion. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the construction site environmental monitoring system based on multi-source fusion provided below can be found in the limitations of the construction site environmental monitoring method based on multi-source fusion described above, and will not be repeated here.

[0170] In one exemplary embodiment, such as Figure 3 As shown, a construction site environmental monitoring system 30 based on multi-source fusion is provided to implement the methods in the above-described embodiments. The system includes:

[0171] The spatiotemporal data synchronization module 31 is used to acquire the raw data packets output by multiple environmental sensors of different types deployed at different locations on the construction site. Based on the sensor identifiers and timestamps in the raw data packets, the module performs spatiotemporal alignment processing on the sensor data corresponding to all environmental sensors to generate a spatiotemporally aligned multi-sensor time series matrix.

[0172] The adaptive data fusion module 32 is used to calculate the real-time data quality factor of each environmental sensor at each moment based on the local statistical characteristics and factory accuracy indicators of the data of each sensor in the multi-sensor time series matrix; based on the real-time data quality factor, the sensor data of a group of environmental sensors monitoring the same environmental parameter and located in close proximity are weighted and fused to generate a fused environmental parameter sequence for several spatial regions and its corresponding comprehensive confidence sequence; wherein, the real-time data quality factor characterizes the comprehensive reliability of the current reading of the environmental sensor; the comprehensive confidence sequence is the mean sequence of the real-time data quality factors of all sensors in the environmental sensor cluster.

[0173] The data reconstruction module 33 is used to perform physically guided sequence prediction and repair on missing segments in the fused environmental parameter sequence based on the physical constraint relationship between various environmental parameters in the sensor data, and to convert the discrete fused environmental parameter sequence into a regular gridded environmental data cube covering the entire construction site through an interpolation algorithm based on the geographic coordinates and comprehensive confidence sequence of each spatial region.

[0174] The dynamic benchmark early warning module 34 is used to extract multi-dimensional environmental state feature vectors from the environmental data cube, construct a dynamic benchmark model based on historical environmental monitoring data of the construction site and related to the construction stage and external weather conditions, calculate the deviation of the multi-dimensional environmental state feature vectors from the dynamic benchmark model at the current moment; when the deviation exceeds the threshold, an environmental anomaly early warning is triggered.

[0175] Embodiments of this application also provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the aforementioned method embodiments.

[0176] Embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0177] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0178] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A construction site environmental monitoring method based on multi-source fusion, characterized in that, The method includes: S1. Obtain the raw data packets output by multiple environmental sensors of different types deployed at different locations on the construction site. Based on the sensor identifiers and timestamps in the raw data packets, perform spatiotemporal alignment processing on the sensor data corresponding to all environmental sensors to generate a spatiotemporally aligned multi-sensor time series matrix. S2. Based on the local statistical characteristics and factory accuracy indicators of the sensor data in the multi-sensor time series matrix, calculate the real-time data quality factor of each environmental sensor at each moment; based on the real-time data quality factor, perform weighted fusion on the sensor data of a group of environmental sensors monitoring the same environmental parameter and located in close proximity, generating a fused environmental parameter sequence for several spatial regions and its corresponding comprehensive confidence sequence; wherein, the real-time data quality factor characterizes the comprehensive reliability of the current reading of the environmental sensor; the comprehensive confidence sequence is the mean sequence of the real-time data quality factors of all sensors in the environmental sensor cluster; S3. Based on the physical constraint relationship between the environmental parameters in the sensor data, perform physical-guided sequence prediction and repair on the missing segments in the fused environmental parameter sequence, and based on the geographic coordinates of each spatial region and the comprehensive confidence sequence, use an interpolation algorithm to convert the discrete fused environmental parameter sequence into a regular gridded environmental data cube covering the entire construction site. S4. Extract multidimensional environmental state feature vectors from the environmental data cube, construct a dynamic benchmark model based on historical environmental monitoring data of the construction site and related to the construction stage and external weather conditions, calculate the deviation of the multidimensional environmental state feature vectors from the dynamic benchmark model at the current moment; when the deviation exceeds the threshold, trigger an environmental anomaly warning.

2. The method according to claim 1, characterized in that, The process involves calculating the real-time data quality factor for each environmental sensor at each moment based on the local statistical characteristics and factory accuracy indicators of the sensor data in the multi-sensor time series matrix; and, based on the real-time data quality factor, performing weighted fusion on the sensor data of a group of environmental sensors monitoring the same environmental parameter and located in spatial proximity to generate a fused environmental parameter sequence for several spatial regions and its corresponding comprehensive confidence sequence, including: S11. For the reading of environmental sensor i at time t in the multi-sensor time series matrix, calculate the corresponding historical mean and historical standard deviation based on the historical data of the environmental sensor within the sliding time window. S12. Combining the historical mean, historical standard deviation, and factory accuracy index of environmental sensor i, the real-time data quality factor of environmental sensor i at time t is calculated by fusing the anomaly deviation evaluation term, the data continuity evaluation term, and the inherent accuracy weight term; wherein, the calculation formula for the real-time data quality factor is: in, Indicates environmental sensors At any moment The real-time data quality factor; Indicates environmental sensors At any moment The original readings; and These represent the sensors. The historical mean and the historical standard deviation; This is an adjustable coefficient used to control the sensitivity of anomaly detection; Indicates environmental sensors Data missing rate within the recent time window; This is the preset missing rate threshold; Indicates environmental sensors The aforementioned factory precision specifications; This indicates the total number of environmental sensors in the environmental sensor cluster; , , Let be the weight coefficient, and satisfy... ; S13. For an environmental sensor cluster whose spatial distance is less than a set threshold and which monitors the same environmental parameter, calculate the fused environmental parameter value of the corresponding environmental sensor cluster at time t based on the real-time data quality factor of each environmental sensor in the cluster and its original reading at time t; wherein, the calculation formula for the fused environmental parameter value is: in, Represents an environmental sensor cluster At any moment The values ​​of the fusion environment parameters; This refers to the collection of the environmental sensor cluster; For traversing the set Index of all sensors in the system; Indicates environmental sensors At any moment The real-time data quality factor; Indicates environmental sensors At any moment The original readings; S14. The mean value of the real-time data quality factor of all environmental sensors in the environmental sensor cluster is used as the comprehensive confidence level of the fused environmental parameter value at time t; all time points and all environmental sensor clusters are traversed to generate the fused environmental parameter sequence and its corresponding comprehensive confidence level sequence for each spatial region.

3. The method according to claim 2, characterized in that, The method, based on the geographic coordinates of each spatial region and the comprehensive confidence sequence, uses an interpolation algorithm to transform the discrete fused environmental parameter sequence into a regular gridded environmental data cube covering the entire construction site, including: S21. Obtain the geographic coordinates of the center point of each spatial region, the fusion environment parameter values ​​corresponding to each time point, and the comprehensive confidence level. S22. Based on the spatial distance of the geographic coordinates, the time difference between adjacent times, and the comprehensive confidence level, a spatiotemporal variogram model incorporating confidence level weights is constructed; wherein, the spatiotemporal variogram model is used to quantify the correlation of data in the spatiotemporal dimension; the comprehensive confidence level is used to adjust the correlation measure between data points, and the higher the comprehensive confidence level, the greater the weight of the data point in the calculation of the spatiotemporal variogram model; the expression of the spatiotemporal variogram model is: in, Represents known data points and The spatiotemporal variation function values ​​between; and Representing data points respectively and The overall confidence level; Representing data points and The spatial Euclidean distance between them; Representing data points and The absolute time difference between corresponding timestamps; For spatial characteristic range parameters, The spatial characteristic range parameter represents the critical distance at which spatial correlation essentially disappears, and the temporal characteristic range parameter represents the critical time span at which temporal correlation essentially disappears. and Let be the weighting coefficients for the spatial and temporal components, and satisfy . ; The spatial characteristic range parameter and the temporal characteristic range parameter are obtained as follows: from the historical fusion environment parameter sequence and its corresponding historical comprehensive confidence, high-quality data points with historical comprehensive confidence higher than a preset confidence threshold are selected; based on the spatial distance and time difference of the high-quality data points, the spatial characteristic range parameter and the temporal characteristic range parameter are determined by fitting an empirical variogram curve. S23. To meet the requirements of the interpolation rule grid points at the target time, establish a corresponding set of kriging equations based on the spatiotemporal variogram model; wherein, the objective of the kriging equations is to minimize the variance of the interpolation estimate and satisfy the unbiasedness condition. S24. Solve the Kriging equations to obtain the known point weight coefficients for each known point used to calculate the environmental parameter values ​​of the regular grid points to be interpolated; wherein, the Kriging equations are expressed as: in, Indicates the first The known point weight coefficients of the known data points to the interpolation rule grid points; Represents a known point With known points The spatiotemporal variation function values ​​between; Represents a known point With the interpolation rule grid points The spatiotemporal variation function values ​​between; For Lagrange multipliers; This represents the total number of known points involved in the interpolation calculation. S25. Based on the weight coefficients of the known points, the fused environmental parameter values ​​of each known point are linearly combined to obtain the estimated environmental parameters of the regular grid points to be interpolated at the target time; based on the fused environmental parameter values ​​and the estimated environmental parameters, the regular gridded environmental data cube is formed.

4. The method according to claim 1, characterized in that, The extraction of multidimensional environmental state feature vectors from the environmental data cube includes: S31. Calculate the spatial average of all grid point values ​​of each type of environmental parameter in the regular gridded environmental data cube to obtain the spatial average of the corresponding type of environmental parameter at the construction site. S32. For the pre-selected key environmental parameters in the regular gridded environmental data cube, calculate the spatial gradient of the key environmental parameters along the preset dominant direction, and extract the maximum value of the spatial gradient as the maximum value of the spatial gradient. S33. Compare the environmental parameter values ​​in the regular gridded environmental data cube with the preset environmental quality standard threshold, and calculate the proportion of the area of ​​grid points exceeding the environmental quality standard threshold to the total monitored area to obtain the proportion of the area exceeding the standard. S34. The reciprocal of the variance of each environmental parameter value in the regular gridded environmental data cube within the sliding time window is used as a spatiotemporal stability index characterizing the degree of environmental fluctuation. S35. The spatial mean, the maximum spatial gradient, the proportion of the area exceeding the standard, and the spatiotemporal stability index are concatenated to obtain the multidimensional environmental state feature vector.

5. The method according to any one of claims 1 to 4, characterized in that, The dynamic benchmark model, constructed based on historical site environmental monitoring data and correlated with construction phases and external weather conditions, calculates the deviation of the multidimensional environmental state feature vector from the dynamic benchmark model at the current moment, including: S41. According to the construction schedule, the historical multidimensional environmental state feature vector is divided into corresponding construction stages to obtain historical feature vector subsets for different construction stages; external weather data of the same historical period as the historical data of the construction site environmental monitoring is obtained, and the external weather data is used as a working condition label and associated with the corresponding historical feature vector subset; wherein, the historical multidimensional environmental state feature vector is calculated based on the historical data of the construction site environmental monitoring, and the calculation method is consistent with the method of calculating the multidimensional environmental state feature vector; S42. Based on the historical feature vector subset and the external weather data, for each combination of working conditions consisting of a specific construction stage and specific external weather conditions, calculate the mean vector and covariance matrix of all historical feature vectors belonging to the working condition combination, and the mean vector and the covariance matrix together constitute the benchmark sub-model in the dynamic benchmark model corresponding to the working condition combination. S43. Determine the current working condition combination based on the actual construction stage and actual external weather conditions at the current moment, and select the reference sub-model corresponding to the current working condition combination from the dynamic reference model; S44. Based on the mean vector and covariance matrix in the selected benchmark sub-model, calculate the Mahalanobis distance between the multidimensional environmental state feature vector and the mean vector at the current time, and use the Mahalanobis distance as the deviation.

6. A construction site environmental monitoring system based on multi-source fusion, used to implement the method according to any one of claims 1 to 5, characterized in that, The system includes: The spatiotemporal data synchronization module is used to acquire the raw data packets output by multiple environmental sensors of different types deployed at different locations on the construction site. Based on the sensor identifiers and timestamps in the raw data packets, the module performs spatiotemporal alignment processing on the sensor data corresponding to all environmental sensors to generate a spatiotemporally aligned multi-sensor time series matrix. An adaptive data fusion module is used to calculate the real-time data quality factor of each environmental sensor at each moment based on the local statistical characteristics and factory accuracy indicators of the sensor data in the multi-sensor time series matrix; based on the real-time data quality factor, the sensor data of a group of environmental sensors monitoring the same environmental parameter and located in close proximity are weighted and fused to generate a fused environmental parameter sequence for several spatial regions and its corresponding comprehensive confidence sequence; wherein, the real-time data quality factor characterizes the comprehensive reliability of the current reading of the environmental sensor; the comprehensive confidence sequence is the mean sequence of the real-time data quality factors of all sensors in the environmental sensor cluster; The data reconstruction module is used to perform physically guided sequence prediction and repair on missing segments in the fused environmental parameter sequence based on the physical constraint relationship between various environmental parameters in the sensor data, and to convert the discrete fused environmental parameter sequence into a regular gridded environmental data cube covering the entire construction site based on the geographic coordinates of each spatial region and the comprehensive confidence sequence through an interpolation algorithm. The dynamic benchmark early warning module is used to extract multi-dimensional environmental state feature vectors from the environmental data cube, construct a dynamic benchmark model based on historical environmental monitoring data of the construction site and associated with the construction stage and external weather conditions, calculate the deviation of the multi-dimensional environmental state feature vectors from the dynamic benchmark model at the current moment; when the deviation exceeds a threshold, an environmental anomaly early warning is triggered.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 5.

Images